cs.CV [Total: 118]
cs.CL [Total: 49]
cs.AI [Total: 8]
cs.GR [Total: 3]
cs.HC [Total: 2]
cs.LG [Total: 10]
cs.MA [Total: 2]
cs.RO [Total: 6]
cs.SE [Total: 1]
eess.AS [Total: 1]
eess.IV [Total: 1]
eess.SP [Total: 1]

cs.CV [Back]

[1] What Does the Encoder Actually Decide? A Controlled Comparison of Vision Backbones on Joint Tree Segmentation and Stereo Depth cs.CVPDF

Yida Lin, Bing Xue, Mengjie Zhang, Sam Schofield, Richard Green

TL;DR: 本文通过固定数据集、解码器、损失函数、训练计划和评估指标，系统比较了不同视觉主干网络在联合树分割和立体深度估计任务上的性能。研究发现卷积和混合架构优于Transformer，参数量并非性能决定因素，且分割与深度任务的排名高度一致。

Details

Motivation: 解决在机器人修剪树木任务中，如何基于实际性能而非声誉选择视觉主干网络，以同时获取像素级语义分割（树/非树）和立体深度信息的问题。

Result: 在约2500万参数预算下，从零开始训练多个架构家族（CNN、Transformer、混合模型等）。最佳编码器（卷积/混合）在树像素深度估计上达到δ1指标，在分割上获得mIoU；相当比例的纯视觉Transformer训练失败；小参数量模型性能可超越大两个数量级的模型。

Insight: 创新点在于构建了参数共享的公平比较框架，揭示了在特定任务（细植被）上传统CNN架构的持续优势、参数量与性能的非相关性，以及边界F1指标对揭示模型退化（全树分割）的关键作用。

Abstract: A robot pruning trees needs two facts per pixel: whether it belongs to a tree, and its distance. Both are usually obtained via task heads attached to a vision backbone chosen by reputation rather than measurement. Holding dataset, decoders, losses, schedule, and evaluation fixed, we ask: how much does the encoder choice change joint semantic segmentation and stereo depth on thin vegetation? We build a hard parameter-sharing network with one encoder feeding both branches, swapping only the encoder without downstream retuning. We evaluate [N] encoders across [M] architecture families (CNNs, transformers, hybrids, MLP-mixers, state-space models) near a ~25M budget, trained from scratch. Depth is evaluated on tree pixels only; segmentation uses boundary F1 and background IoU to prevent “label-everything-tree” shortcuts. Three findings stand out. First, the strongest encoders are convolutional and hybrid, not transformers: [BestEncoder] leads with [MIoU] segmentation mIoU and [Delta] depth $δ_1$, while [X] of [Y] plain vision transformers collapse when trained from scratch. Second, parameter count does not predict quality — [SmallEncoder] at only [P]M parameters outranks models two orders of magnitude larger. Third, segmentation and depth rankings agree strongly (Spearman $ρ$ = [RhoValue]), showing no task conflict. Finally, [K] of [N] encoders collapse to degenerate all-tree segmentation — exposed by boundary F1 but hidden by region IoU.

[2] Don’t Just Look, Intervene: Perturbation Based Region Labeling for VQA Images cs.CVPDF

Marko Jojic, Zhaonan Li, Ben Zhou

TL;DR: 本文提出了一种名为CSGR（Counterfactual Search for Grounding Regions）的可扩展流程，用于自动标注视觉问答（VQA）任务中决定模型答案的关键图像区域。该方法通过干预候选区域并测量其对模型答案分布的影响来识别关键区域，并将生成的标注用于增强现有基于视觉依据的训练方法。

Details

Motivation: 视觉语言模型（VLMs）应依赖直接决定正确答案的视觉证据，但用于视觉推理依据的监督标注通常获取成本高昂或依赖于特定数据集的标注原语。因此，本文旨在通过模型因果干预自动生成关键区域标注，以提供可扩展的监督信号。

Result: 实验将CSGR标注集成到三种现有的基于视觉依据的训练方法（注意力引导、视觉思维链微调和潜在视觉推理）中。在领域内和领域外评估中，与仅使用交叉熵微调的基线相比，CSGR标注相比其他自动区域标注机制带来了最一致的性能提升，表明其捕获了有用的区域级信息。

Insight: 创新点在于提出了“模型因果视觉证据”作为标注目标，即通过反事实干预改变模型答案分布来定义关键区域。这提供了一种与具体模型和训练方式解耦的、可扩展的自动标注方案，能够为多种视觉依据训练方法提供有效的监督信号。

Abstract: Vision Language Models (VLMs) should rely on visual evidence that directly determines the correct answer, but supervision for grounding visual reasoning is often expensive to obtain manually or tied to dataset-specific annotation primitives. We instead introduce model-causal visual evidence as an annotation target, defined as the set of image regions whose counterfactual intervention changes a model’s answer distribution for a given image-question pair. Based on this principle, we introduce Counterfactual Search for Grounding Regions (CSGR). CSGR is a scalable pipeline that proposes candidate regions, perturbs them, measures their effect on answer sensitivity, and aggregates this evidence across multiple judges to approximate answer-critical regions in VQA data. To assess whether CSGR annotations contain a useful supervision signal, we plug them into three existing grounding-aware training routines: attention steering, Visual CoTfinetuning, and latent visual reasoning. These experiments test whether the proposed annotation scheme can provide a useful supervision signal across multiple ways of consuming region labels, rather than introducing a new way of using them. Across competing automatic region-labeling mechanisms, CSGR annotations provide the most consistent gains over Cross Entropy-only finetuning in both in-domain and out-of-domain evaluations, indicating that the proposed labeling scheme captures useful region-level information.

[3] Part Grounding, Not Action Knowledge: Locating the Bottleneck in VLM Affordance Prediction cs.CV | cs.LG | cs.ROPDF

Sarthak Sattigeri

TL;DR: 该论文通过分析视觉语言模型在可操作性预测任务中的表现，发现模型在低层次操作推理上的主要瓶颈并非缺乏动作知识，而是难以准确定位物体上的操作部件。研究将任务分解为部件定位和动作知识两个步骤，并通过对19个铰接物体和8个不同模型进行实验，证明当明确指定目标部件时，模型的动作预测准确率显著提升。

Details

Motivation: 现有基准测试表明视觉语言模型在低层次操作推理上表现不佳，但整体准确率无法揭示具体失败环节。研究旨在区分可操作性预测中两个被混淆的步骤：识别物体上需要操作的具体部件，以及了解该部件所需的动作，以定位性能瓶颈。

Result: 在开放提示下，所有模型的表现均未超过忽略图像的恒定基线。但当明确命名目标部件后，所有8个模型的平均动作准确率从0.158-0.474大幅提升至0.684-0.947，其中‘推’动作的召回率从0-1/8提升至7-8/8。在真实照片和渲染物体上的辅助实验进一步证实了部件定位是普遍瓶颈。

Insight: 核心创新在于将可操作性预测分解并量化了部件定位与动作知识两个子问题的贡献，通过控制实验揭示了部件定位是当前视觉语言模型的主要性能瓶颈。这一发现挑战了模型缺乏动作知识的普遍假设，并为未来模型改进（如增强视觉基础能力）提供了明确方向。研究同时强调了通过测试平凡替代方案来验证评估指标的重要性。

Abstract: Benchmarks agree that vision-language models reason poorly about low-level manipulation, but an aggregate accuracy score does not say which step fails. We separate two steps that affordance questions conflate: identifying which part of an object to act on, and knowing what action that part requires. Across 19 articulated objects we asked eight models, spanning three developers, what motion a robot should apply. Under an open prompt, push was produced once in 64 evaluations where it was correct, despite being correct for 8 of 19 objects and appearing in the offered label set every time. Inspecting the outputs showed why: models described a different part than the one being scored, e.g. explaining how to pick up a camera rather than press its button. Naming the target part raises action accuracy by 0.32 to 0.63 for every model, from 0.158-0.474 to 0.684-0.947, and push recall from 0-1/8 to 7-8/8. No model beats a constant answer that ignores the image under the open prompt; once the part is named, all eight do. Asked to describe the same part in free prose with no label set, models produce pressing language for 6 to 8 of 8. These results are hard to reconcile with missing action knowledge, and instead point to part grounding as the dominant bottleneck, a pattern that holds across all three model families and does not diminish with capability. Naming the part supplies the grounding variable, so this bounds what a perfect part detector would offer rather than demonstrating a general model of mechanics. Two supporting results agree: on real photographs only three of eight models localize grasp points better than a constant baseline, and on rendered objects none do. We also document two measurement errors of our own, a threshold that let a constant baseline score 0.929 and a labelling rule wrong on 4 of 19 objects, both caught only by testing our numbers against trivial alternatives.

[4] Occlusal Geometry in Closed Form for Orthodontic Report Generation cs.CV | cs.CLPDF

Ajo Babu George, Govind Arun, Sidharth N Krishna, Uma Ranjan

TL;DR: 本文提出了一种用于正畸报告生成的闭合形式咬合几何分析方法。该方法利用已配准的咬合扫描数据直接测量核心咬合参数，通过梯度提升将31个测量值映射到13个报告模板字段，并结合ConvNeXt-Tiny分类器提升准确性。系统在ODIN 2026 Bite2Text测试中排名第三，每个病例在CPU上运行时间少于10秒。

Details

Motivation: 解决传统多模态描述方法无法直接利用已配准咬合扫描数据的问题，通过直接测量咬合几何参数而非推断，提高正畸报告生成的准确性和效率。

Result: 在ODIN 2026 Bite2Text测试阶段获得第三名（BLEU-4 0.2680，METEOR 0.4629），与第一名仅差0.022 BLEU-4；留出测试得分达到BLEU-4 0.458和METEOR 0.677（相对于口内扫描参考）。

Insight: 创新点在于从咬合几何直接推导闭合形式的咬合参数（如覆𬌗、覆盖、中线偏差等），而非依赖传统RAS坐标系；采用测量值到模板字段的映射策略，并结合图像分类器进行多模态融合，在保持高效运行的同时提升报告准确性。

Abstract: Orthodontic report generation from intraoral data is normally cast as multimodal captioning, yet the released Bite2Text scan pairs are supplied already registered in occlusion, which makes several core occlusal quantities directly measurable rather than inferable. The system reported here exploits that property: an anatomical frame is recovered per case from arch taper and arch closure instead of the stated RAS convention, which does not hold across the release, and each arch is reduced to an occlusal ridge profile in arch-angle coordinates yielding overbite, overjet, midline deviation, transverse overlap, crossbite extent, cusp interdigitation lag, and the occlusal curves in closed form. Gradient boosting maps 31 such measurements onto 13 template fields, a field being predicted only where patient-level cross-validation beats its own majority baseline, and a deterministic renderer emits the corpus six-part narrative; a ConvNeXt-Tiny classifier over the five standardised photographic views is fused per field, raising mean field accuracy from 0.601 to 0.683. Reimplementation of the challenge evaluator shows that its BLEU-4 and METEOR are local variants whose F-mean weights recall nine to one, that two clinicians agree on 47 percent of findings for the same patient, and that a constant report consequently outscores a genuine second clinician report by 0.165 captioning. Held-out scores reach BLEU-4 0.458 and METEOR 0.677 against intraoral scan references and 0.278 and 0.507 against photograph references, and the submitted system placed third in the ODIN 2026 Bite2Text test phase at 0.2680 and 0.4629, within 0.022 BLEU-4 of first, running on CPU in under ten seconds per case. The dataset and code are available at https://github.com/GIND123/ODIN_toothfairy4

[5] ArtSociety: Multi-Agent Multimodal Collaboration for Art Emotion Understanding cs.CVPDF

Jian Li, Fanfan Ji, Jinxiang Lai, Ying Tai, Jian Yang

TL;DR: 本文提出了ArtSociety，一个用于艺术情感理解的多智能体多模态协作框架。该框架整合了异构的视觉、语言和推理智能体，并通过无训练的控制器协调它们，以解决AffectiveArt任务中细粒度情感分类、效价/唤醒度预测和属性描述等多个子任务之间的权衡问题。

Details

Motivation: AffectiveArt多维艺术情感理解任务需要联合预测细粒度情感、二元效价/唤醒度及五个属性描述，这些子任务之间存在强烈的经验权衡，单一模型难以同时优化所有任务。

Result: 在官方测试集（1000件艺术品）上，ArtSociety实现了0.8870的综合得分（分类0.7789，描述0.9952）。消融研究表明，决定性的性能提升来自智能体协作和数据侧监督。

Insight: 创新点在于采用异构多智能体协作框架，并设计了两种无训练控制器：针对尾部情感的罕见类别感知投票仲裁器，以及强制视觉证据先于标签承诺的“描述-分类”链式推理智能体。这为解决多任务权衡和利用错误模式去相关性提供了新思路。

Abstract: The AffectiveArt Multidimensional Art Emotion Understanding task asks to jointly predict an artwork’s fine-grained emotion (12 classes, 1549:1 head-to-tail ratio), binary valence/arousal, and five attribute-grounded descriptions – sub-tasks that exhibit strong empirical trade-offs, so the single-model solutions we tried do not jointly optimize all of them well. We present ArtSociety, a multi-agent framework that assembles heterogeneous multimodal experts – a DINOv2-Giant vision agent (A1), a scene-grounded CoT fine-tuned MLLM (A2), and three closed-source reasoning agents (A3-A5) – and coordinates them with two training-free controllers: (i) a rare-class-aware voting arbiter that lowers the agreement threshold for tail emotions, exploiting decorrelated error patterns across agent families; and (ii) a description-first reasoning agent whose DESCRIBE-then-CLASSIFY chain of thought forces visual evidence before label commitment, yielding near-perfect grounded descriptions. A task-routing policy directs the hard emotion task to the full five-agent ensemble while assigning the near-saturated valence/arousal and generative description tasks to the single strongest reasoning agent. On the official test set (1,000 artworks), ArtSociety achieves an Overall Score of 0.8870 (Classification 0.7789, Description 0.9952). An eleven-variant ablation study reveals that, once method and scale saturate at around 0.76, the decisive gains come from agent collaboration and data-side supervision – a 30B MoE model trained on older data does not outperform an 8B model trained on better data. Code is available at https://github.com/swordlidev/ArtSociety

[6] SJD-SV: Speculative Jacobi Decoding with Semantics Verification for Autoregressive Image Generation cs.CVPDF

Baoquan Zhang, Bingqi Shan, Shihao Fang, Kenghong Lin, Xutao Li

TL;DR: 本文提出了一种名为SJD-SV的新方法，用于加速自回归图像生成。该方法通过识别语义感知的令牌子序列，并在该子序列级别而非逐个令牌级别进行验证，以解决现有推测性雅可比解码（SJD）中存在的令牌歧义问题。该方法是一个即插即用的模块，可集成到现有SJD及其变体中，并在多个数据集上显著提升了性能。

Details

Motivation: 现有推测性雅可比解码（SJD）方法在加速自回归图像生成时，存在令牌歧义问题，但其原因未能得到很好解释。本文通过可视化分析发现，视觉令牌通常对应局部、细小且模糊的视觉细节，单个令牌难以准确表达特定语义，从而导致了歧义。

Result: 在多个数据集上的广泛实验表明，现有的SJD方法在集成本文提出的SJD-SV方法后，均取得了显著的性能提升。

Insight: 核心创新点在于利用令牌间的强相关性来识别语义感知的令牌子序列，并将验证粒度从单个令牌提升到语义子序列级别，从而更有效地解决令牌歧义问题。该方法设计为即插即用模块，具有良好的兼容性和实用性。

Abstract: Speculative Jacobi Decoding (SJD) is an important approach for accelerating autoregressive image generation. Although SJD has shown superior performance, recent studies point out that it usually suffers from a token ambiguity issue during token verification but its reason can not be well explained. To figure out this reason, in this paper, we conduct a visualization analysis on vision token and find that different from text tokens, vision tokens generally corresponds to some local, small, and unclear vision details, which means only using single token is difficult to accurately express a certain semantic, thereby causing token ambiguity issue. To this end, we propose a novel Speculative Jacobi Decoding with Semantics Verification (called SJD-SV), for accelerating autoregressive image generation. The key idea is that leveraging the strong correction characters between tokens to recognize semantic-aware token subsequence and then instead of perform token-by-token verification, turning to perform verification on semantic-aware token subsequence level for accelerating image generation. In particular, our method is plug-in, which can be directly integrated into existing SJD and its variants. Extensive experiments on various datasets show that existing SJD methods achieve significant performance improvement after integrating our SJD-SV method.

[7] Pixel-wise Planarity for High-Precision Monocular Plane Segmentation cs.CVPDF

Ahmetcan Yavuz, Alpay Ozkan, Rémi Pautrat, Shaohui Liu, Marc Pollefeys

TL;DR: 本文提出了一种基于像素级平面性预测的单目平面分割框架，通过预训练的深度和法线估计骨干网络，结合专门的平面性置信度头，在推理时通过轻量级区域生长方法整合深度、法线和平面性信息，以增强几何一致性。该方法旨在解决现有方法因区域分组不精确和几何监督不一致导致的过分割和错误检测问题。

Details

Motivation: 解决单目RGB图像平面分割中因不精确的区域分组和几何不一致的监督导致的过分割和错误平面检测问题。

Result: 在多个数据集上，该方法相比先前的最先进方法，在几何精度和分割质量上均有所提升，同时提高了计算效率。

Insight: 创新点在于引入像素级平面性置信度预测头，并与深度和法线预测结合，通过几何一致的轻量级区域生长进行分割；客观分析认为，该方法通过显式建模平面性并严格评估标注的几何不一致性，提升了分割的鲁棒性和精度。

Abstract: Plane segmentation from a single RGB image remains challenging due to imprecise region grouping and geometrically inconsistent supervision, often leading to over-segmentation and false planar detections. We propose instead a pixel-wise planarity prediction framework for robust monocular plane segmentation. Building on a pretrained monocular geometric backbone predicting depth and surface normals, we introduce a dedicated planarity head that estimates per-pixel planarity confidence. During inference, predicted depth, normals, and planarity are combined in a lightweight region-growing procedure that enforces geometric consistency when forming plane segments. We further analyze existing plane ground-truth annotations and demonstrate substantial geometric inconsistencies under strict distance thresholds. Across multiple datasets, our method achieves improved geometric precision and segmentation quality compared to prior state-of-the-art approaches, while improving computational efficiency. Our code and models are available at https://github.com/alpayozkan/PixelwisePlanarity.

[8] Evaluation of MLLM-Agnostic Plug-and-Play Keyframe Selection Methods for Long Video Understanding cs.CV | cs.AIPDF

Dilip Sarkar, Md. Safayet Islam, Liang Liang

TL;DR: 本文对五种无需训练、即插即用的关键帧选择方法进行了全面评估，这些方法旨在解决多模态大语言模型因视觉令牌和计算预算限制而无法处理长视频所有帧的问题。评估在三个MLLM和三个长视频理解基准上进行，结果表明QAaF在大多数设置中性能最佳，FOCUS次之。

Details

Motivation: 动机在于解决MLLMs处理长视频时因计算和令牌限制无法分析所有帧的难题，并对比现有低成本、无需训练的关键帧选择方法，为社区提供统一的评估基准。

Result: 在三个长视频理解基准（如视频问答数据集）上使用三个MLLM进行评估，QAaF在15个综合评估设置中的13个取得最佳性能，达到当前SOTA水平，FOCUS总体排名第二。

Insight: 创新点在于首次对多种MLLM无关的即插即用关键帧选择方法进行系统化横向评估，揭示了不同方法在统一实验设置下的性能差异，为后续研究提供了重要参考；客观来看，其评估框架本身具有方法论价值，有助于推动该领域标准化。

Abstract: Multimodal large language models (MLLMs) cannot process every frame of a long video because of limitations in visual-token and computational budgets. Three primary approaches have been proposed to enhance their long-video understanding capabilities: (i) Retraining an MLLM on a large video corpus and/or extending its input length; (ii) Training an adapter for a specific MLLM that takes the entire video and the query as input and selects the most relevant video frames; and (iii) Developing a training-free, plug-and-play (PaP) adapter that is MLLM-agnostic. We refer to the third approach as PaP keyframe selection. A PaP method may use only candidate video frames without considering the query, or it may use both candidate video frames and the query. The first approach is prohibitively expensive. The second approach requires substantial training time and computational resources, but it is accessible to many because an adapter contains significantly fewer trainable parameters than an entire MLLM. The third approach has the lowest computational cost and is therefore broadly accessible. To the best of our knowledge, only five PaP methods have been reported within the past year. All of these methods have been evaluated on one or more video question-answering benchmarks and have demonstrated improvements in long-video understanding. However, the methods were evaluated on different benchmarks using different MLLMs. We present a comprehensive evaluation of these five methods using three MLLMs across three long-video understanding benchmarks. Our results show that QAaF achieves the best performance in 13 of the 15 aggregate evaluation settings, while FOCUS ranks second overall. These results provide a common experimental reference for comparing training-free keyframe-selection methods for MLLMs.

[9] Preserving Subject-Clarity in Image Outpainting with Multiscale Wavelet Supervision cs.CVPDF

Abhilash Neog, Taewan Kim, Yi Wu, Xu Chen, Jian Jiao

TL;DR: 本文提出了一种用于图像外绘（outpainting）的主题清晰度保持框架，旨在解决现有扩散模型方法在扩展图像边界时导致主题保真度下降的问题。该方法结合了视觉语言模型（VLM）引导的语义条件与多尺度小波监督，以在生成过程中保持主题的局部细节和结构一致性。

Details

Motivation: 商业和广告图像常因构图不佳、主题被裁剪、文本或徽标被截断以及上下文不足而影响主题清晰度，即图像清晰传达其主要主题的能力。现有基于扩散的外绘方法虽然能生成视觉上合理的内容，但往往导致主题结构不一致、语义漂移或细节丢失，从而降低主题保真度。

Result: 在四个广告和自然图像基准测试中，该方法提高了主题清晰度，与匹配的监督微调相比，平均将主题中心的DreamSim误差和FID分别降低了3.0%和2.4%；与每个数据集上最强的SOTA方法相比，平均分别降低了10.8%和7.7%。

Insight: 创新点在于将VLM引导的语义条件与多尺度小波监督相结合，以在扩散模型框架内实现主题局部细节的精确保持，同时不增加推理成本。客观来看，其主题中心的数据构建流程和兼容现有骨干网络的设计也具有借鉴价值。

Abstract: Commercial and advertising images are frequently affected by poor framing, partially cropped subjects, truncated text or logos, and insufficient context, all of which can reduce subject clarity, i.e., the ability of an image to clearly communicate its primary subject. Image outpainting offers a scalable solution by extending image boundaries and recovering missing content and context. However, existing diffusion-based outpainting methods often produce visually plausible completions while degrading subject fidelity through structural inconsistencies, semantic drift, or loss of fine-grained detail. To address this limitation, we propose a subject clarity outpainting framework that combines vision-language model (VLM)-guided semantic conditioning with multiscale wavelet supervision for subject-localized detail preservation. To support training, we develop a subject-centric data curation pipeline that constructs subject-intersecting outpainting pairs from advertising and natural images. The resulting objective introduces no additional inference cost and is designed to be compatible with diffusion-based backbones. Across four advertising and natural-image benchmarks, our method improves subject clarity, reducing subject-centered DreamSim error and FID on average by 3.0% and 2.4% over matched supervised fine-tuning, and by 10.8% and 7.7% over the strongest state-of-the-art approach per dataset, respectively.

[10] (How) Do MLLMs Report Bistable Images Like Humans? cs.CV | cs.AIPDF

Ryota Takatsuki, Tomoki Doi, Amane Watahiki, Anil K. Seth, Hitomi Yanaka

TL;DR: 该论文研究多模态大语言模型（MLLMs）在处理双稳态图像（如鸭兔图）时的报告行为是否与人类相似，并探究其内部计算机制。通过使用LLaVA系列模型，论文从可调节性和排他性两个维度进行分析，发现视觉和语言线索能系统性影响模型报告，且报告具有排他性。

Details

Motivation: 动机是探究MLLMs在面对支持多种互斥解释的双稳态图像时，其报告行为是否表现出与人类相似的特性，以及支撑这种行为的内部计算过程是什么。

Result: 在经典鸭兔图和合成的视觉字谜上测试，行为结果显示视觉和语言操纵能系统地以符合人类预期的方式改变模型报告，且响应主要保持排他性。机制上，这些效应源于竞争的图像-标记表示、自底向上和自顶向下调节的不同路径，以及排他性报告与物体计数编码之间的联系。

Insight: 创新点在于将人类认知心理学中的双稳态图像范式应用于评估MLLMs，并揭示了模型内部表示如何支持类似人类的可调节和排他性报告行为，为理解MLLMs的视觉-语言整合机制提供了新视角。

Abstract: Bistable images such as the duck-rabbit are classic stimuli in which one image supports multiple mutually incompatible interpretations, typically reported one at a time in humans. We ask whether multimodal large language models (MLLMs) show similar report behavior and what internal computations support it. Using the LLaVA family, we study two tractable dimensions: modulability, whether reports can be biased by bottom-up visual cues and top-down linguistic priors, and exclusivity, whether responses commit to a single interpretation. We test both on the canonical duck-rabbit and on synthetic Visual Anagrams to mitigate memorization confounds. Behaviorally, both visual and linguistic manipulations systematically shift reports in human-consistent ways, while responses remain predominantly exclusive. Mechanistically, these effects arise from competing image-token representations, distinct pathways for bottom-up and top-down modulation, and a link between exclusive reporting and object-count encoding. Code and data are available at https://github.com/rtakatsky/mllm-bistable-images.

[11] Sampling headroom is not selection gain: a compute-value audit of test-time scaling for video world models cs.CV | cs.AIPDF

Yuhua Jiang, Junjie Lu, Feifei Gao

TL;DR: 本文提出计算价值审计框架，用于评估视频世界模型中测试时扩展的有效性。研究发现，增加采样数量虽能提升候选视频质量，但现有选择机制无法可靠识别最优样本，导致额外计算投入无法转化为实际性能增益。

Details

Motivation: 针对视频世界模型中测试时扩展的计算效率问题，研究旨在厘清额外采样是否真能通过选择机制提升生成质量，而非仅扩大候选池规模。

Result: 在Physics-IQ数据集上，候选池从4扩至16时理论质量提升9.23 IQ，但Flow/Cycle/VideoReward等选择器未能有效利用该空间；12种自适应策略均未超越均匀计算分配，仅回收42-69%计算成本。

Insight: 创新点在于提出四阶段审计框架，揭示采样空间与选择增益的脱钩现象；核心启示是部署价值取决于能否建立可靠决策机制，使收益覆盖完整计算成本。

Abstract: Test-time scaling (TTS) can improve generation only when additional compute produces better candidates and the system can reliably identify them. This distinction is especially important for video world models, where a wider sample pool may contain stronger rollouts without improving the output that is ultimately selected. We introduce the Compute-Value Audit (CVA), a sequential framework that asks whether extra sampling creates opportunity, observable signals provide a reliable state, that state supports a beneficial action, and the resulting gain exceeds the full entry fee of generation and verification. On 192 Physics-IQ scenes, expanding the pool from 4 to 16 candidates increases oracle quality by +9.23 IQ (95% CI [+7.44, +11.14]), but Flow, Cycle, and VideoReward fail to recover this headroom reliably. Across three generators, none of twelve adaptive-depth policies outperforms uniform compute; they recover only 42-69% of the measured entry fee. A matched-60-NFE Predict-and-Perturb intervention on VideoPhy2 is likewise negative across three fresh-seed replicas. These negative results are not universal: anchor-explorer passes all four stages in a sparse PRM800K setting, MMLU-Pro exposes the gap between predictive state and useful action, and a privileged paired future establishes a positive video upper bound. Together, these results show that sampling headroom has deployment value only when it can be converted into a reliable decision whose benefit survives the complete compute charge. Code is available at https://github.com/YuhuaJiang2002/sampling-headroom-is-not-selection-gain.

[12] A Comprehensive Review of Multimodal Facial State Analysis: Tasks, Methods, and Resources cs.CVPDF

Xuri Ge, Tianshuo Zhang, Ruihan Li, Hui Ye, Kaiwen Zheng

TL;DR: 本文是一篇关于多模态面部状态分析的综述性论文，全面回顾了该领域的核心任务、代表性方法及数据集。论文指出，传统单模态视觉方法存在环境敏感性和弱可解释性等局限，而多模态方法通过整合视觉、音频、文本、生理等多源互补信息，并结合多任务学习，能够提升面部状态推理的语义理解、模型可解释性以及细粒度表达与跨场景泛化能力。

Details

Motivation: 传统基于视觉的单模态面部状态分析方法受限于环境敏感性和弱可解释性，难以满足对人脸表情、心理建模和人机交互深入理解的需求。因此，需要综述如何通过多模态融合与多任务学习来应对这些挑战。

Result: 作为一篇综述，本文未提出具体模型，因此没有定量实验结果。它系统梳理了在多模态面部状态分析（如面部表情识别、动作单元检测和基于人脸的软生物特征估计）任务上的现有方法、基准数据集和研究进展。

Insight: 论文强调语言模态在提供上下文语义、增强推理和生成解释方面的独特价值，并突出了多模态学习与多任务学习相结合对于实现可解释、自适应面部状态分析的重要性，为未来研究方向提供了清晰的指引。

Abstract: Facial state analysis plays a crucial role in understanding human expressions, psychological modeling, and human computer interaction. Traditional unimodal vision-based methods are often limited by environmental sensitivity and weak interpretability. Multimodal facial state analysis addresses these issues by integrating complementary cues from visual, audio, textual, physiological, and other related modalities. This survey emphasizes two key aspects: on one hand, multimodal learning enables contextual semantic understanding for improved facial state reasoning and leverages interpretable language generation to enhance model explainability; on the other hand, multi-task learning allows simultaneous analysis of expressions, action units (AUs), and face-based soft biometrics (e.g., age, gender), effectively capturing fine-grained expressions and improving cross-scene generalization. This survey reviews core tasks, representative methods, and datasets in multimodal facial state analysis, focusing on facial expression recognition, AU detection, and face-based soft biometric estimation, and emphasizing the unique value of language in providing contextual semantics, enhancing reasoning, and generating explanations. The survey aims to provide an up-to-date overview of the literature and to highlight future research directions for multimodal, interpretable, and multi-task adaptive facial state analysis.

[13] Interpretable Temporal Video Reasoning with EventGraph and EventField cs.CV | cs.CL | cs.IRPDF

Durgendra Narayan Singh

TL;DR: 本文提出了一种结构化的时序视频推理流程，该流程围绕离散的EventGraph、连续的EventField和人类可读的EventGlyph视图构建。在EPIC-KITCHENS数据集的一个校准子集上，该方法在时序推理问答任务中取得了0.98的总体准确率，显著优于基线方法。结果表明，结构化时序表征能够通过保留符号结构、捕捉时序连续性并提供人类可读的诊断信息，同时支持高性能和可检查性。

Details

Motivation: 旨在解决视频时序推理任务中，如何构建既高性能又具备可解释性的结构化表征的问题。

Result: 在EPIC-KITCHENS数据集的10个视频、50个时序推理问题的子集上，EventField+Glyph方法取得了0.98的总体准确率，比基于描述的基线方法高出0.40，比仅使用视觉语言模型（VLM）的直接问答方法高出0.20，且结果具有统计显著性。

Insight: 创新点在于将离散的图结构（EventGraph）与连续的场表征（EventField）相结合，并辅以人类可读的可视化（EventGlyph），从而构建了一个兼具高性能、时序连续性和强可解释性的视频推理框架。

Abstract: We present a structured temporal video reasoning pipeline built around a discrete EventGraph, a continuous EventField, and a human-readable EventGlyph view. On a calibrated EPIC-KITCHENS subset of 10 videos and 50 temporal reasoning questions, EventField+Glyph achieves 0.98 overall accuracy, which is higher than the caption baseline by +0.40 (paired p = 1.1 \times 10^{-5}) and direct VLM-only QA by +0.20 (p = 0.0063) on this subset. We further evaluate annotation-source variations, including manual, heuristic, and heuristic+Gemini pipelines, and find that the best structured method stays above the caption baseline across settings. We also include cross-video pair benchmarking and an appendix gallery of glyph outputs for all studied videos. Overall, the results indicate that structured temporal representations can support both performance and inspectability by preserving symbolic structure, capturing temporal continuity, and providing human-readable diagnostics for video reasoning.

[14] BEACON: Behavior and Appearance Control for Subject-Specific Video Generation cs.CV | cs.AIPDF

Pokrzywa Baptiste, Nabyl Quignon, Yara Bahram, Muhammad Osama Zeeshan, Antitza Dantcheva

TL;DR: 本文提出了BEACON框架，用于生成保留特定人物视觉身份和个性化表情行为的人类中心视频。该方法通过将视觉身份与表情行为解耦，并利用参考图像和参考视频作为互补条件信号，从而生成更具表现力且身份保持性好的视频。

Details

Motivation: 现有最先进方法通常仅基于单张参考图像生成视频，缺乏对人物特定时间动态表情的建模，导致生成的表情变化有限且缺乏人物特异性。

Result: 在MEAD和RAVDESS数据集上的实验表明，仅需对预训练的Wan视频扩散模型微调约2000个样本并更新约1%的参数，BEACON在面部表现力上超越了现有最先进的视频生成方法，同时保持了有竞争力的身份保持能力。

Insight: 核心创新在于将视觉身份与表情行为解耦，并引入参考视频作为补充条件来捕捉人物特定的面部动态，从而支持身份-表情的迁移，这是一种轻量级且高效的框架设计。

Abstract: Generating human-centric videos that preserve both visual identity and person-specific expressive behavior remains a fundamental challenge. In addition to reproducing appearance, a model must replicate the facial behaviors that characterize how a subject expresses emotion over time. However, most state-of-the-art methods condition generation on a single reference image, which contains no information about these temporal dynamics. As a result, they tend to preserve the subject’s visual identity but often produce expressions with limited variation and weak subject specificity. To mitigate this issue, we introduce BEACON, a lightweight framework for person-specific video generation that produces more expressive videos by disentangling visual identity from expressive behavior. BEACON conditions generation on two complementary signals: a reference image encoding the identity and a reference video capturing subject-specific facial dynamics. By conditioning on these complementary signals, BEACON generates videos that better preserve both the subject’s appearance and characteristic facial dynamics, while also supporting identity-expression transfer. Our experiments on the MEAD and RAVDESS datasets show that by fine-tuning on approximately 2,000 pairs and updating about 1% of the pretrained Wan video diffusion model, BEACON improves facial expressivity over state-of-the-art video generation methods while maintaining competitive identity preservation.

[15] TryOnReward: Learning Foveated Consistency for Reinforcement Fine-Tuning of Virtual Try-On cs.CV | cs.AIPDF

Xueheng Li, Yong Liu, Xiaolong Fu, Wen Xue, Chengjun Xie

TL;DR: 本文提出了TryOnReward，一个专为虚拟试穿（VTON）设计的细粒度奖励模型，用于解决现有通用视觉语言模型（VLMs）在评估试穿质量时缺乏判别力的问题。该模型采用焦点校准目标和边际感知监督，联合优化成对偏好和每维度质量分数，并构建了大规模人工标注数据集和基准进行训练与评估。

Details

Motivation: 虚拟试穿的目标是生成符合人类偏好的视觉合理结果，但经典的保真度指标与人类判断相关性弱，通用VLMs无法提供试穿质量评估所需的细粒度判别力，这在强化学习微调（RFT）中会导致严重的奖励黑客问题。

Result: 实验表明，TryOnReward在人类偏好对齐方面显著优于通用评判模型，当作为RFT奖励函数时，能在多个基线模型上持续产生人类更偏好的试穿结果。

Insight: 创新点在于设计了焦点校准目标，将每个质量维度锚定在相关区域以避免全局捷径学习，并联合利用相对（成对偏好）和绝对（每维度评分）质量信号进行监督。从客观角度看，构建大规模人工标注数据集和专门基准也为该领域提供了重要的评估资源。

Abstract: Virtual Try-On (VTON) aims to dress a person with the reference garment, producing visually reasonable results aligned with human preferences. Turning this preference-oriented goal into an actionable objective relies on a scoring function aligned with human taste. However, classic fidelity metrics exhibit weak correlation with human judgments, and generic VLMs fail to provide the discriminative granularity demanded by try-on quality evaluation, which hinges on faithfully preserving garment and person details. This shortcoming is further exacerbated in the reinforcement fine-tuning (RFT) optimization and leads to severe reward hacking. To this end, we present TryOnReward, a fine-grained reward model tailored for VTON. Built on a vision-language backbone, it adopts a foveation calibration objective that grounds each quality dimension in the relevant region to avoid global shortcut learning. Meanwhile, TryOnReward jointly optimizes pairwise preferences and per-dimension quality scores via margin-aware supervision, leveraging both relative and absolute quality signals. For model training and evaluation, we build TryOnReward-100K, a human-annotated per-dimension rating dataset, alongside TryOn-Bench and TryOnRewardBench, two benchmarks covering diverse real scenarios. Extensive experiments confirm that TryOnReward significantly outperforms generic judges in human preference alignment, and when serving as the RFT reward function, it consistently yields human-preferred try-on results across multiple baselines.

[16] SomBench: Benchmark Dataset for Advancing Machine Learning in Lunar Science cs.CV | cs.LGPDF

Himanshu Patil, Gabby Nyirjesy, Rachel A. Slank, Vishal Gaur, Daniela Szwarcman

TL;DR: SomBench是一个用于月球科学机器学习的统一基准数据集，整合了来自四个轨道任务的十种仪器的30多个共配准数据层，覆盖了从1米到20公里/像素的空间分辨率。该数据集通过图像锚定的分块流程生成预训练就绪的多模态图块视图，并提供了涵盖撞击过程、火山历史和极地挥发物等任务的基准测试套件。

Details

Motivation: 现有月球轨道任务的多仪器观测数据在采样、投影和惯例上存在异质性，限制了机器学习研究的可复现性，因此需要构建一个统一、空间对齐且适合机器学习的数据集来推动月球科学中的机器学习应用。

Result: 使用ResNet-50和SwinV2-B模型进行的基线实验证实，每个基准任务都可以从发布的数据中学习，为未来的模型开发建立了参考点。

Insight: 创新点在于构建了一个大规模、多模态、空间对齐的月球数据集，并设计了防泄漏的数据划分和预训练就绪的图块视图，为月球科学中的机器学习研究提供了标准化的基准平台。

Abstract: Lunar orbital missions, such as Lunar Reconnaissance Orbiter, Kaguya/SELENE, Gravity Recovery and Interior Laboratory, and Lunar Prospector, among others, provide rich multi-instrument observations, but their heterogeneity in sampling, projection, and conventions limits reproducible machine learning (ML). We introduce SomBench, a unified, spatially-aligned, ML-ready lunar dataset aggregating 30+ co-registered layers from ten instruments across four missions, spanning 1 meter to 20 kilometer/pixel and covering 82 degree latitude in 90 Lunar Transverse Mercator zones with two polar stereographic caps. An image-anchored tiling pipeline yields pretraining-ready multimodal tile views with leakage-safe splits, distributed as netCDF with Parquet catalogs. An application benchmark suite spans impact processes, volcanic history, and polar volatiles. Baseline experiments with ResNet-50 and SwinV2-B models confirm that each benchmark task is learnable from the released inputs, establishing reference points for future model development.

[17] Structure-Token Evidence-Anchored Reasoning for Scientific Chart Understanding cs.CVPDF

Alberlucia Rafael Soarez, Camila Ferreira, Daniel Kim, Mariana Costa, Alejandro Torres

TL;DR: 本文提出STEER方法，通过冻结Llama-3.2-Vision编码器并插入三个模块（图表结构图编码器、证据锚定逐步推理、弱解析器-强推理器对齐），来解决大视觉语言模型将科学图表误当作自然图像处理的问题，从而提升图表理解和推理的准确性。

Details

Motivation: 现有大视觉语言模型在处理科学图表时，往往将其视为自然照片，无法有效利用坐标轴、图例和几何标记中的量化信息，导致推理过程缺乏依据，可能生成未从图表中读取的合理数字。

Result: 在ChartQA数据集上，STEER达到82.70的平均松弛准确率，优于ChartGemma的80.16和相同混合数据训练的LLaVA-CoT骨干模型的76.40；在CharXiv推理任务上达到33.60（对比InternVL Chat V1.5的29.20），在ChartQAPro CoT上达到40.70（对比Qwen2-VL-7B的37.17），尤其在OCR捷径消失的任务上优势更明显。

Insight: 创新点包括：将图表结构显式建模为图结构以绑定刻度、图例项和标记；强制每个算术推理步骤引用图节点作为证据；采用弱解析器-强推理器对齐策略，仅使用专用表格提取器作为节点属性的教师，从而增强推理的可解释性和准确性。

Abstract: Scientific charts encode quantities in axes, legends, and geometric marks, yet large vision-language models still treat them as natural photographs. Visual in-context examples do not expose the coordinate frame; unconstrained chain-of-thought can name a plausible number that was never read from a bar. We present STEER (Structure-Token Evidence-anchored Reasoning), which freezes a Llama-3.2-Vision encoder and inserts three modules: a chart structure graph encoder (CSGE) that binds ticks, legend items, and marks; evidence-anchored step reasoning (EASR) that forces every arithmetic step to cite a graph node; and weak-parser strong-reasoner alignment (WPSR) that uses a specialized table extractor only as a teacher of node attributes. On ChartQA, STEER reaches 82.70 average relaxed accuracy versus 80.16 for ChartGemma and 76.40 for a LLaVA-CoT backbone trained on the same mix. Gains widen on CharXiv reasoning (33.60 vs. 29.20 InternVL Chat V1.5) and ChartQAPro CoT (40.70 vs. 37.17 Qwen2-VL-7B), where OCR shortcuts disappear. Ablations show that dropping node serialization or numeric candidate constraints undoes most of the reasoning lift.

[18] Frame-Synchronous Hand Gesture Detection by Projected Winding Order cs.CV | cs.GR | cs.HC | cs.LGPDF

Amey Thakur

TL;DR: 本文提出了一种无需分类器、训练数据或校准的帧同步手势检测方法，专门用于检测手掌绕长轴旋转的精确时刻。通过投影后的二维叉积构造标量s，其零点对应手掌侧向的瞬间，符号变化指示手掌朝向，实现了对旋转手势的精确帧级检测。该方法在双手交互场景中通过重新设计场景表示和覆盖谓词，解决了手部交叉时的跟踪问题，并在包含精确真值的数据集上验证了高检测精度和低延迟性能。

Details

Motivation: 传统视频手势识别通常基于逐帧分类，适用于控制任务但无法满足同步需求，即需要将输出精确对齐到手势实际发生的帧。本文旨在解决手掌旋转这一常见动作的精确同步问题，避免分类器依赖和训练数据需求。

Result: 在包含精确真值的240个接近失误序列数据集中，该方法检测到95%的翻转动作且无假阳性，平均检测延迟为6.7毫秒，仅为观察间隔的六分之一，优于基于采样报告的方法（23.0毫秒）。

Insight: 创新点在于将手势检测建模为连续量的零点检测问题而非分类任务，通过投影叉积的代数形式实现了对图像镜像、手部尺度和左右手的不变性；在双手交互中，通过覆盖谓词和基于噪声统计的参数推导，解决了手部交叉时的跟踪难题。

Abstract: Gesture recognition on video is normally posed as classification: label each frame, then act on the label. That is adequate for control, where a command may be obeyed several frames late without a user noticing, and inadequate for synchronisation, where an output must be aligned to the frame on which the gesture physically occurred. We take the synchronisation problem for one common movement, the rotation of an open hand about its long axis, and show that it admits an exact solution needing no classifier, no training data and no calibration. Let s be the normalised two-dimensional cross product of the two palm edges at the wrist and the outer knuckles, under the projection the camera already performs. We prove that s factorises as k(theta)cos(theta) with |k| > 0 everywhere, so s vanishes exactly when the palm is edge-on and its sign tracks the face presented to the camera. Detection is therefore a zero crossing of one scalar, which yields an instant rather than an interval, and we prove the criterion invariant to image mirroring, hand scale and handedness, all from its algebraic form rather than from the landmark estimator. A two-handed interaction uses four fingertips as a window onto a restyled version of the same scene. We give a coverage predicate that stays correct when the hands cross, where the usual triangulation does not, and derive each free parameter of the three stylisation operators from a noise statistic rather than by inspection. Against a corpus with exact ground truth the criterion detects 95% of flips with no false positive in 240 near-miss sequences, and places each within 6.7 ms on average of the instant it occurred, a sixth of the interval at which the hand is observed, against 23.0 ms for reporting the bracketing sample. Reporting an event more finely than one samples follows from treating a gesture as the zero of a continuous quantity rather than a label.

[19] RxScribe Bench: A Multi-Axis Benchmark for Evaluating Vision-Language Models on Indian Outpatient Prescriptions cs.CVPDF

Somil, Vijay Saini, Vidit Verma, Riya, Aastha Batta

TL;DR: RxScribe Bench是一个用于评估视觉语言模型在手写处方数字化任务上的基准测试，它将性能分解为与临床严重性相关的四个维度，而不是汇总为一个单一分数。该基准要求模型根据处方图像和输出模式生成结构化记录，并与人工标注的真实值进行逐字段比较，每个字段还标注了可见性和清晰度。评估发现，在四个维度上没有单一模型表现全面领先。

Details

Motivation: 现有的处方转录准确性评估通常报告一个混合的单一数字，将不同临床风险的错误（如虚构药物与误读清晰剂量）等同对待，这无法反映真实的临床风险差异。

Result: 在真实处方图像上对前沿视觉语言模型进行了独立冷启动评估，结果显示没有单一模型在所有四个维度（正确性、幻觉、参与度、鲁棒性）上都表现最佳。

Insight: 创新点在于提出了一个多维度评估基准，将性能分解为与临床严重性直接相关的四个独立轴（正确性、幻觉、参与度、鲁棒性），特别是鲁棒性维度引入了最低证据阈值机制，避免在证据不足时报告结果，这为医疗AI评估提供了更细粒度和临床相关的分析框架。

Abstract: Prescription transcription errors are not interchangeable. A model that fabricates a drug and a model that misreads a legible dose pose very different clinical risks, yet prescription-transcription accuracy is typically reported as a single blended figure that treats the two as equivalent. We introduce RxScribe Bench, a benchmark for evaluating vision-language models on handwritten prescription digitization that decomposes performance into four axes tied to clinical severity, rather than folding everything into a single aggregated score. Given only a prescription image and an output schema, a model produces a structured record, which is then compared field by field against a human-authored ground truth of identical shape, with each field also labeled for visibility and legibility. The four axes isolate distinct failure modes, namely Correctness, Hallucination, Engagement, and Robustness. The Robustness axis withholds its hard-field results rather than reporting one when the supporting sample falls below a minimum-evidence threshold. We evaluate frontier vision-language models on real prescriptions across independent cold runs per image, and find that no single model wins across all four axes.

[20] Multimodal-Multiresolution Foundation Model for Lunar Remote Sensing cs.CV | cs.AI | cs.LGPDF

Paolo Fraccaro, Gabby Nyirjesy, Daniela Szwarcman, Himanshu Patil, Vishal Gaur

TL;DR: 本文提出了一种用于月球遥感的多模态基础模型，该模型在包含近200万个配准图块、涵盖11种模态和两种空间尺度（1米/像素和100米/像素）的SomBench数据集上从头开始预训练。模型基于TerraMind掩码令牌架构，并引入了两个月球特定扩展：将采集几何作为显式上下文，并联合训练米级和百米级图块以实现单一权重覆盖双分辨率。模型在四个基准测试（WAC/NAC陨石坑检测、不规则月海斑块分割和极地冰前景回归）中表现优异，匹配或超越了ImageNet预训练基线。

Details

Motivation: 解决月球遥感中缺乏大规模预训练多模态基础模型的问题，旨在利用多分辨率、多模态数据提升对月球地表特征的表示和理解能力。

Result: 在四个基准测试中，预训练模型匹配或超越了ImageNet预训练基线及随机初始化对照模型。在WAC陨石坑检测任务上，仅用50%数据训练的预训练模型性能优于使用全数据集的ImageNet基线。LoRA适配策略在陨石坑检测和IMP分割任务上匹配或超越了全微调，且参数更少。

Insight: 创新点包括：将采集几何作为显式上下文输入；联合训练双分辨率图块以实现单一权重覆盖；采用FlexiViT补丁嵌入支持不同补丁尺寸无需重训练；模态独立输入支持灵活的多模态微调。模型架构和预训练策略均带来性能增益。

Abstract: We present a multimodal foundation model for lunar remote sensing, pretrained from scratch on SomBench, a geographically partitioned corpus of nearly two million co-registered tile bundles spanning 11 modalities at two spatial scales (1 m/pixel and 100 m/pixel). The model adapts the TerraMind masked-token architecture with two lunar-specific extensions: acquisition geometry is provided as explicit context, and meter- and hundred-meter-scale tiles are trained jointly so that a single set of weights covers both resolutions. FlexiViT patch embeddings allow adaptation to different patch sizes without retraining, while modality-wise inputs enable flexible multimodal fine-tuning. Qualitative generation experiments suggest the model learns meaningful cross-modal correspondences, including terrain derivatives from elevation and illumination-consistent reflectance from geometry. We evaluate on four benchmarks: crater detection at WAC and NAC scales, irregular mare patch (IMP) segmentation, and polar ice prospectivity regression. Across tasks, the pretrained model matches or outperforms ImageNet-pretrained baselines and an architecturally identical random-init control. On multimodal ice prospectivity regression, pretrained variants achieve the best results, while the random-init model outperforms most baselines, suggesting gains arise from both the architecture and pretraining. Label efficiency is notable for WAC crater detection, where the pretrained model trained on 50% of the data exceeds the strongest ImageNet baseline trained on the full dataset. Among adaptation strategies, LoRA matches or surpasses full fine-tuning on crater detection and IMP segmentation while using far fewer trainable parameters, whereas full fine-tuning performs best for ice prospectivity regression. We release the pretrained checkpoint, benchmark datasets, and fine-tuning code to support reproducible lunar AI research.

[21] LLaDA-UI: Bringing Block-wise Diffusion to Vision-Language GUI Agents cs.CV | cs.AIPDF

Zhangxuan Gu, Haoxing Chen, Qi Qin, Yi Xin, Kai Gan

TL;DR: 本文提出了LLaDA-UI，一个基于混合专家架构、参数为16.7B的块级扩散视觉语言GUI智能体。它采用两阶段训练流程：首先进行通用多模态预训练以对齐视觉编码器和扩散语言主干，然后在多样化的GUI数据上进行监督微调。实验表明，该模型在多个平台的GUI基准测试中超越了Qwen2.5-VL-7B等模型，证明了块级扩散范式在GUI智能体中的实用性。

Details

Motivation: 扩散大语言模型因其块并行、任意顺序生成的高解码效率，在延迟敏感的应用中具有吸引力。GUI智能体需要实时感知屏幕状态并输出结构化的、空间定位的动作，是测试该范式的理想场景。本文旨在探索能否将扩散大语言模型扩展为强大的多模态GUI智能体，同时保持其并行解码优势。

Result: 在广泛采用的定位基准测试和跨多个平台的导航基准测试中，LLaDA-UI显著优于Qwen2.5-VL-7B模型，并在六个报告的GUI基准测试中的四个上超越了Qwen3-VL-8B模型。

Insight: 主要创新点是将块级扩散生成范式成功扩展到多模态GUI智能体领域，并保持了其并行解码的高效率优势。从客观角度看，其两阶段训练策略（通用预训练+GUI特定微调）以及采用混合专家架构来构建一个参数可控但能力强大的模型，是值得借鉴的设计思路。

Abstract: Diffusion large language models (dLLMs) achieve high decoding efficiency through block-parallel, arbitrary-order generation, making them attractive for latency-sensitive applications. GUI agents represent a natural testbed for this paradigm, as they must repeatedly perceive screen states and emit structured, spatially grounded actions in real time. However, whether dLLMs can be extended into capable multimodal GUI agents while preserving their parallel decoding advantage remains an open question. We present LLaDA-UI, a 16.7B-parameter MoE-based, block-wise diffusion vision-language GUI agent. LLaDA-UI follows a two-stage training pipeline: general multimodal pre-training aligns a native-resolution vision encoder with the LLaDA2.0-mini-base diffusion language backbone, followed by GUI-agent supervised fine-tuning on diverse mobile, desktop, web, and grounding data. Across widely adopted grounding benchmarks and navigation benchmarks spanning multiple platforms, LLaDA-UI substantially outperforms Qwen2.5-VL-7B and surpasses Qwen3-VL-8B on four of six reported GUI benchmarks. These results establish block-wise diffusion as a practical generative paradigm for multimodal GUI agents.

[22] Target-Checked Reliability Score Refinement for Video Question Answering cs.CVPDF

Guoxiang Ren, Rohitash Chandra

TL;DR: 本文提出了一种无需重新训练模型或改变其答案的可靠性分数优化方法，用于提升视频问答（VideoQA）中多选问题的置信度校准。该方法通过收集多个固定视频-语言模型在不同视频采样下的选项概率列表，构建响应图来表征跨视图变化和跨模型一致性，并利用目标试点数据进行基于直方图的梯度提升（HGB）和正则化逻辑回归分数优化，仅在视频级检查表明稳定改进时才替换原始分数。

Details

Motivation: 视频-语言模型在回答多选问题时可能表现出高置信度但答案错误，因此需要在不重新训练模型或改变答案的情况下，通过目标偏移（target shift）来改进答案级别的可靠性分数。

Result: 在HERBench（963个问题）和Perception Test（911个问题）两个独立测试集上，该方法将三个模型的平均风险覆盖曲线下面积（AURC）分别降低了16.64%和18.87%，均显著优于七种无需训练的基线方法，并提升了AUROC、降低了校准误差，在50%覆盖率下将错误率降低了约6.5个百分点。

Insight: 创新点在于利用多模型、多视频采样的响应图来表征不确定性，并结合目标试点数据进行轻量级后处理优化可靠性分数，仅在稳定改进时才替换原始分数，从而在不改变模型答案的情况下有效提升置信度校准和错误检测能力。

Abstract: Video-language models can answer multiple-choice questions with high confidence yet be wrong. We study whether answer-level reliability scores can be improved under target shift without retraining the models or changing their answers. We collect option-probability lists from three fixed video-language models under four deterministic video samplings and represent cross-view changes and cross-model agreement as a response graph. Using a labeled target pilot, we compare the original score, defined as the probability assigned to the chosen answer, with a histogram-based gradient-boosting (HGB) score trained on the development datasets and a regularized logistic-regression score trained on the target pilot. A candidate replaces the original score only when repeated video-level checks indicate a positive, stable improvement. We develop this rule on public VideoQA benchmarks and Video Hallucination Diagnosis (VHD), a controlled diagnostic dataset for shared high-confidence errors. Ranking quality is measured by the area under the risk-coverage curve (AURC), where lower is better. On a held-out 963-question HERBench split, the method reduces mean AURC across the three models by 16.64% (95% confidence interval (CI), 12.12 to 22.61%); the smallest model-level gain is 11.39%. On a separate held-out 911-question Perception Test split, the mean reduction is 18.87% (95% CI, 15.43 to 22.14%). For InternVL3.5, the target check retains the original scores. Using the same outputs, the method outperforms seven training-free baselines in mean AURC on both datasets. It also improves AUROC, reduces calibration error, and lowers the error rate at 50% coverage by 6.50 and 6.58 percentage points.

[23] Harnessing Image Question Dependence for Better VLM Test-time Reinforcement Learning cs.CV | cs.LGPDF

Xinrui He, Ting-Wei Li, Junting Wang, Mengting Ai, Xinyu He

TL;DR: 本文提出TTIQ框架，通过利用图像-问题依赖关系改进视觉语言模型（VLM）的测试时强化学习。该方法通过教师强制生成原始图像-问题对及其消融变体的响应，利用词元级似然变化估计输入依赖，构建奖励函数以促进联合基于图像和问题的响应，而非仅依赖共识。

Details

Motivation: 现有基于共识的测试时强化学习受限于自生成学习信号的可靠性，其增益主要源于答案归一化而非内容修正，且初始VLM响应常因未能联合利用图像和问题信息而产生错误，导致奖励信号可能固化而非纠正这些错误。

Result: 在八个VQA数据集和多种VLM规模上的实验表明，TTIQ在所有模型尺度上均取得最佳平均性能，且能泛化到不同VLM家族，在未见数据集上无需额外训练即可提升性能。

Insight: 创新点在于利用图像和问题的依赖关系构建响应级奖励和词元级策略信用分配，强调联合基于双输入的响应而非仅追求流行答案；客观分析认为其通过消融变体量化输入依赖的方法为VLM适应性提供了更细粒度的学习信号。

Abstract: Test-time reinforcement learning can adapt vision-language models (VLMs) to unlabeled target data, but its effectiveness is fundamentally limited by the reliability of self-generated learning signals. To assess the reliability of consensus-based learning signals, we analyze VLM test-time reinforcement learning across diverse VQA datasets and model sizes, revealing two limitations. First, gains from consensus-based test-time training largely come from answer normalization rather than content correction. Second, many initial VLM responses are incorrect due to the model’s limited ability to jointly use the image and the question; consensus rewards derived from these outputs may preserve the resulting grounding errors rather than correct them. Motivated by these, we propose TTIQ, a test-time reinforcement learning framework that harnesses image-question dependence for better vlm adaptation. TTIQ teacher-forces each sampled response under the original image-question pair and its image- and question-ablated variants, using the resulting token-level likelihood changes to estimate dependence on each input. It combines image and question dependence with calibrated confidence to construct a response-level reward that favors jointly grounded responses, and uses the token-level signals to assign greater positive policy credit to tokens supported by both inputs. This design favors responses that are jointly grounded in the image and the question and sufficiently confident, rather than merely popular. Experiments across eight VQA datasets and multiple VLM sizes show that TTIQ achieves the best average performance at every model scale. It further generalizes across VLM families, while models trained on one dataset improve performance on unseen datasets without further training.

[24] Adaptive Conformal Redistribution for Inter-class Transitional Uncertainty in Medical Image Classification cs.CV | cs.AI | cs.LG | stat.CO | stat.MLPDF

Saibal Ghosh, Samarup Bhattacharya, Sanjoy Kumar Saha, Umapada Pal, Tapabrata Chakraborti

TL;DR: 本文提出了一种名为自适应保形重分配（AdaConRed）的后处理决策规则，用于解决医学图像分类中过渡类别导致的决策模糊问题。该方法通过一个五阶段流程，包括视觉-语言生成增强、冻结编码器、轻量级分类器、自适应预测集构建以及基于模型输出的重分配，将模糊的预测集转化为精确的类别分配。

Details

Motivation: 医学图像分类中，过渡类别的特征分布与相邻类别重叠，导致决策边界模糊。保形预测虽能提供不确定性感知的预测集，但在需要单一决策的临床筛查中无法直接使用，因此需要一种能将模糊预测集转化为可操作决策的方法。

Result: 在OSCC口腔病变和ISIC皮肤病变基准测试上，误覆盖水平为0.2时，AdaConRed显著提升了性能。在3类OSCC基准上，总体准确率从73.54%提升至77.38%，口腔癌准确率从64.29%提升至82.14%，良性准确率从56.57%提升至70.20%。在ISIC基准上，总体准确率从85.83%提升至87.19%，黑色素瘤准确率从66.04%提升至68.34%。该方法在相同骨干网络和重分配规则下优于LAC、APS和RAPS。

Insight: 创新点在于将保形预测从不确定性量化扩展到可操作的决策支持，通过熵调制、边界感知的非保形分数构建自适应预测集，并仅使用推理时的模型输出来重分配过渡样本。这为存在过渡疾病类别的临床场景提供了实用的后处理解决方案，且性能提升集中在临床关键类别上。

Abstract: Medical image classification is frequently complicated by transitional categories whose feature distributions overlap those of adjacent classes, producing ambiguous decision boundaries. Conformal prediction returns uncertainty-aware prediction sets, but these are not directly actionable in clinical screening, where a single decision is required. This work proposes adaptive conformal redistribution (AdaConRed), a label-free post-conformal decision rule that converts ambiguous prediction sets into refined class assignments. A five-stage pipeline is developed. Vision-language generative augmentation addresses minority-class scarcity; a frozen DermFoundation encoder provides embeddings; a lightweight multi-layer perceptron performs classification; an entropy-modulated, margin-aware nonconformity score constructs adaptive prediction sets; samples predicted as transitional with multi-label sets are reassigned to the most probable alternative class within the set, using only model outputs at inference. Evaluation uses the OSCC oral lesion and ISIC skin lesion benchmarks at a miscoverage level of 0.2. On the 3-class OSCC benchmark, overall accuracy improves from 73.54% to 77.38%, with oral cancer accuracy rising from 64.29% to 82.14% and benign accuracy from 56.57% to 70.20%. Reassignment of transitional samples reduces OPMD accuracy from 84.78% to 80.16%, consistent with the asymmetric cost of missed malignancy. On ISIC, overall accuracy improves from 85.83% to 87.19%, melanoma accuracy rising from 66.04% to 68.34%. AdaConRed outperforms LAC, APS and RAPS under an identical backbone and redistribution rule. Conformal prediction can be extended beyond uncertainty quantification toward actionable decision support where transitional disease categories are present, with gains concentrated in the clinically critical malignant categories. Code repository: https://github.com/saibal436ghosh/AdaConRed.

[25] GroundBench: A Factorized, Counterfactual Benchmark for Locating VLM Affordance Failures cs.CV | cs.LG | cs.ROPDF

Sarthak Sattigeri

TL;DR: 论文提出了GroundBench，一个用于诊断视觉语言模型（VLM）在可操作性（affordance）任务中失败原因的因子化、反事实基准。该基准通过六个分支-合并条件，分离了视觉定位、机械推理和类别-动作关联等混杂因素，并通过反事实提问测试模型是否依赖文本捷径而非真实视觉推理。

Details

Motivation: 现有评估发现，在操作提示中直接命名目标部件能显著提升VLM的动作准确性，但这提供了真实系统本应推断的信息，混淆了模型失败的根本原因（如视觉定位能力不足）。因此，需要一个新的诊断基准来分离这些解释因素，定位VLM的具体能力缺陷。

Result: 在三个OpenAI模型（包括GPT-5）上的1,068次预测评估表明：仅提供目标区域（无类别信息）时，动作准确率在或低于0.53的多数基线；仅提供类别信息（无位置）时，准确率提升至0.68-0.74。所有超出基线的增益都发生在提供的部件类别本身就能决定动作的情况下。反事实测试中，GPT-5在74对样本上实现了0.86的配对加权遵从率，但未能处理所有观察到的‘推转为垂直举’案例。

Insight: 创新点在于设计了一个因子化、反事实的基准测试框架，能够系统地解耦并量化VLM在可操作性任务中依赖的不同信息源（如位置、类别、文本关联），从而揭示模型性能提升是源于真正的视觉推理还是文本层面的类别-动作关联捷径。这为更精细地评估和诊断VLM的 grounding 能力提供了方法论。

Abstract: A companion evaluation found that naming the target part in a manipulation prompt increased action accuracy by 0.32-0.63 across eight vision-language models, with no model outperforming a constant baseline until the part was named. However, naming the part supplies information that a real system must infer, confounding visual grounding, mechanical reasoning, and category-to-action association. We introduce GroundBench, a diagnostic benchmark that separates these explanations through six branch-and-merge conditions, each adding a controlled information bundle, and a counterfactual re-ask targeting a real alternate part visible in the same image. Across three OpenAI models and 1,068 predictions, supplying the target region without its identity leaves action accuracy at or below the 0.53 majority baseline (0.26, 0.26, and 0.53), although the models largely reproduce the supplied region. Supplying identity without location instead yields 0.74, 0.68, and 0.68. Every above-baseline gain in this curated set occurs where the supplied part category itself determines the action. A no-vision control leaves GPT-5’s scores unchanged or improved, providing evidence consistent with substantial category-to-action association. GPT-4o mini declines on one condition, so this interpretation is not universal. Adding joint type and motion axis does not improve accuracy across six model-stratum comparisons. On 74 counterfactual pairs from 32 objects, GPT-5 achieves 0.86 pair-weighted compliance with a 0.07 shortcut rate but fails all observed push-to-lift-vertical cases. GroundBench identifies which supplied information changes affordance behavior and tests whether apparently grounded performance can be reproduced through textual shortcuts.

[26] Task-Based CT Protocol Optimization Using Reinforcement Learning and Virtual Imaging Trials cs.CV | cs.AI | cs.LGPDF

Jiaqi Zou, David Fenwick, Vahid Tarokh, Nicholas Felice, Jayasai Rajagopal

TL;DR: 本文提出了一种结合强化学习和虚拟成像试验的CT协议优化框架，旨在平衡图像质量与辐射剂量。该方法利用预训练的视觉变换器提取患者定位图嵌入，指导强化学习代理在少量协议测试中高效搜索最优参数组合，显著减少了穷举测试的需求。

Details

Motivation: CT协议优化需要权衡诊断图像质量和辐射剂量，但由于采集与重建参数相互依赖，穷举测试不切实际，因此需要一种高效的任务驱动优化方法。

Result: 在包含63个计算人体模型的虚拟试验中，仅评估每个患者8个协议（约穷举测试的2%），即可恢复98.2%的穷举搜索最优目标；无患者特定模拟时，代理评分达到89.7%恢复率，而基于定位图的策略比盲策略提升10.7个百分点（p=0.02）。

Insight: 创新点在于将强化学习与虚拟成像试验结合，并利用预训练视觉变换器提取患者特定嵌入来指导协议优化，实现了任务驱动、剂量感知的协议选择，大幅提升了优化效率。

Abstract: Protocol optimization in computed tomography (CT) aims to improve diagnostic image quality while reducing radiation dose, but the interdependence of acquisition and reconstruction parameters makes exhaustive testing impractical. We propose a virtual imaging trial framework with reinforcement learning for efficient CT protocol optimization. Sixty-three computational human models with liver lesions were imaged using a validated CT simulator across 468 combinations of acquisition and reconstruction parameters, including tube voltage, tube current, reconstruction kernel, slice thickness, and pixel size. The optimization objective balanced liver lesion detectability, quantified by detectability index d-prime, against radiation dose. A Proximal Policy Optimization agent was trained and conditioned on patient-specific CT localizer embeddings derived from a pretrained vision transformer. On held-out patients, evaluating only 8 protocols per patient, about 2% of exhaustive testing, recovered 98.2% of the exhaustive-search oracle objective. With no patient-specific simulation, surrogate scoring alone achieved 89.7% recovery. Conditioning on the localizer improved zero-simulation recovery by 10.7 percentage points over the localizer-blind policy (paired 95% CI 2.9-19.5; p=0.02). These results show that the proposed framework can substantially reduce exhaustive protocol testing while enabling task-based, dose-aware protocol selection before the diagnostic scan.

[27] Pedestrian Crossing Intent Classification From Event-Based Vision Using Convolutional Spiking Neural Networks With Temporal Augmentation cs.CV | cs.AIPDF

Henok Teklu, Mustafa Sakhai, Maciej Wielgosz, Matej Mertik

TL;DR: 本文提出了一种基于事件相机的行人过马路意图分类方法。通过将真实驾驶视频转换为合成动态视觉传感器（DVS）事件流，并结合模拟数据增强，训练了一种新型卷积脉冲神经网络（Conv-SNN）。该方法在多个数据集上实现了高精度分类，同时具有低功耗和高效处理稀疏时间数据的优势。

Details

Motivation: 解决自动驾驶中实时预测行人过马路意图的安全关键问题，克服传统基于帧的深度网络在处理运动模糊、高动态范围和类别不平衡等挑战性条件时的局限性，特别是其时间分辨率低和能效差的问题。

Result: 在JAAD DVS测试集上达到95.83%准确率和F1=0.9695，在正常天气的CARLA DVS上达到97.79%准确率和F1=0.9478，在恶劣天气的CARLA DVS上达到94.78%准确率和F1=0.8369。模型参数量仅1.07M，在CPU上训练，其精度达到或超过了先前基于帧的方法在JAAD数据集上报告的水平。

Insight: 创新点在于构建了从真实视频到合成事件流的端到端处理流程，并提出了结合剪辑一致性DVS数据增强的卷积脉冲神经网络架构。该方法能够原生处理稀疏的时间表征，在保持高精度的同时提高了能效，为事件相机在自动驾驶感知任务中的应用提供了有效方案。

Abstract: Anticipating whether a pedestrian will cross the road is safety-critical for autonomous vehicles, requiring real-time inference under challenging conditions including motion blur, high dynamic range, and class imbalance. Conventional frame-based deep networks process redundant RGB data at fixed frame rates, limiting their temporal resolution and energy efficiency. In this work we present an end-to-end pipeline that (i) converts real-world driving footage from the Joint Attention in Autonomous Driving (JAAD) dataset into synthetic dynamic vision sensor (DVS) event streams using the v2e simulator, (ii) augments training with the CARLA-simulated DVS sequences of the DVS-PedX dataset under both normal and adverse weather conditions, and (iii) trains a novel convolutional spiking neural network (Conv-SNN) with clip-consistent DVS augmentation to classify pedestrian crossing intent as binary: crossing or non-crossing. We detail all architectural decisions, the exact leaky-integrate-and-fire neuron dynamics with surrogate-gradient learning, the class-balanced loss formulation, JAAD oversampling at 6x, and a 70/15/15 stratified splitting protocol. The trained model achieves 95.83% accuracy and F1 = 0.9695 on the JAAD DVS test set, 97.79% accuracy and F1 = 0.9478 on normal CARLA DVS, and 94.78% accuracy and F1 = 0.8369 on adverse-weather CARLA DVS, all from a 1.07M-parameter architecture trained on CPU. Compared to prior frame-based approaches on JAAD, our method closes or surpasses the reported accuracy while operating natively on sparse temporal representations. We include a thorough analysis of the convergence behaviour across all 15 training epochs, domain transfer characteristics, and a quantitative comparison with representative related work.

[28] Global-Local Contextual Progressive Expansion Network for Martian Landslide Segmentation in Multimodal Remote Sensing Imagery cs.CVPDF

Leo Thomas Ramos, Sidike Paheding, Abel A. Reyes-Angulo, Rajaneesh A., Sajinkumar K. S.

TL;DR: 本文提出TransCPLES网络，一种结合上下文渐进层扩展特征提取与Transformer上下文推理的U型网络，用于火星滑坡分割。该模型旨在捕捉局部地貌模式和更广泛的空间依赖性，以在MMLSv2多模态火星滑坡数据集上实现更可靠的分割。实验表明，该模型在地理分布不同的样本上取得了最佳综合性能，在不同滑坡范围下分割一致，前景判别稳定，并在精度与计算成本间取得了良好平衡。

Details

Motivation: 火星滑坡自动分割是理解其表面过程的重要任务，有助于未来太空探索。然而，由于滑坡形态多变、前景区域稀疏或不规则，且轨道观测结合了异质的光谱和地形线索，这仍是一个相对未被充分探索的开放挑战。

Result: 在七波段多模态火星滑坡数据集MMLSv2上的实验表明，TransCPLES在地理分布不同的样本上取得了最佳综合性能，分割结果在不同滑坡范围下一致，前景判别稳定，与多种最先进的卷积、注意力机制和Transformer分割模型相比，在精度和计算成本间取得了有利的平衡。

Insight: 论文宣称的创新点是提出TransCPLES网络，它耦合了上下文渐进层扩展特征提取与Transformer上下文推理，以同时捕获局部地貌模式和全局空间依赖。从客观角度看，这是首次在该领域进行如此全面的深度学习模型评估，并将渐进扩展与Transformer结合用于行星遥感图像分割，是一个新颖的架构设计。

Abstract: Automated landslide segmentation on Mars is one of the important tasks for understanding its surface processes, and all will aid in future space exploration. However, it remains a relatively underexplored open challenge because landslide morphology is highly variable, foreground regions are often sparse or irregular, and orbital observations combine heterogeneous spectral and topographic cues. In this context, this work investigates the capability of deep learning to address Martian landslide segmentation through an extensive assessment of modern neural segmentation models. To the best of our knowledge, this is the first study to provide such a comprehensive exploration in this domain. We further propose TransCPLES, a U-shaped network that couples Contextual Progressive Layer Expansion feature extraction with Transformer-based contextual reasoning, enabling the model to capture local geomorphic patterns and broader spatial dependencies for more reliable landslide delineation. Experiments on MMLSv2, a seven-band multimodal Martian landslide dataset, show that TransCPLES achieves the best overall performance when evaluated on geographically distinct samples, with consistent delineation across different landslide extents, stable foreground discrimination, and a favorable balance between accuracy and computational cost compared with several state-of-the-art convolutional, attention-based, and Transformer-based segmentation models. With this work, we hope to provide a useful reference and encourage further research and development in deep learning for planetary remote sensing. Code will be available after publication.

[29] RIGOR: Rig-Informed Geometry for Omnidirectional Reconstruction cs.CVPDF

Tingjun Huang, Dmitry Rudshin, Mathieu Meyer, Pietro Bonazzi, Marc Pollefeys

TL;DR: RIGOR是一种针对重力对齐的全景视频的大规模重建流水线，它通过冻结前馈式透视主干网络，并将每个全景图像视为一个四视图虚拟装置，以检测和修复局部不一致的预测，通过循环四视图一致性检索闭环，并在全局优化前对候选重访进行几何验证，从而改善长轨迹下的重建一致性。

Details

Motivation: 解决现有前馈式3D重建模型在长轨迹、重复结构、弱纹理和动态物体等挑战性环境中预测不一致的问题，并利用全景相机提供的宽空间覆盖和丰富视觉信息来缓解这些挑战，同时避免对360度图像进行额外微调。

Result: 在具有挑战性的建筑工地序列上，所提出的一致性机制在轨迹精度和重建几何方面均优于前馈式基线模型。

Insight: 创新点在于将全景图像作为四视图虚拟装置来构建一致性机制，包括局部不一致性检测与修复、基于循环共识的闭环检索以及几何验证，这些机制被整合到一个Sim(3)位姿图中以校正累积的旋转、平移和尺度漂移。

Abstract: Recent developments in feed-forward 3D reconstruction resulted in models which can recover dense scene representations and camera motion solely from an image stream. However, such predictions are prone to becoming inconsistent over long trajectories, specifically in demanding environments with repetitive structures, weak textures and dynamic objects or people. One way to mitigate those challenges is to use an omnidirectional camera, which provides wide spatial coverage and captures richer visual information. Yet, the majority of models do not offer support for 360-degree imagery or require additional fine-tuning. To bridge these two aspects, we present RIGOR: a large-scale reconstruction pipeline for gravity-aligned omnidirectional videos that retains a frozen feed-forward perspective backbone and exploits each panorama as a four-view virtual rig. The rig structure is used to detect and repair locally inconsistent predictions, to retrieve loop closures through cyclic four-view consensus, and to geometrically verify candidate revisits before global optimization. Verified constraints drive a Sim(3) pose graph that corrects accumulated rotation, translation, and scale drift along the sequence. We demonstrate that the proposed consistency mechanisms improve both trajectory accuracy and reconstructed geometry over a feed-forward baseline on challenging construction-site sequences. The code is made available under this link: https://github.com/TangentH/RIGOR.

[30] Towards Practical Precision Agriculture: Real-Time Fruit Detection and Video Analytics on Embedded Edge Hardware cs.CVPDF

Ivica Dimitrovski, Vlatko Spasev, Ivan Kitanovski, Petre Lameski, Dane Boshev

TL;DR: 该研究提出了一个面向精准农业的端到端实时水果检测、跟踪与计数框架，部署在NVIDIA Jetson Orin Nano嵌入式边缘硬件上。它使用轻量级YOLO26s检测器在四个公共水果数据集上进行训练，并通过TensorRT优化部署，实现了高帧率、低延迟和低能耗的实时视频分析流水线。

Details

Motivation: 解决静态图像基准测试无法满足实际果园视频分析的计算与实时性需求的问题，旨在开发一个能在嵌入式边缘设备上实时运行的实用化精准农业监测系统。

Result: 在四个水果检测任务上，平均mAP@50:95在0.4957到0.8656之间。在Jetson平台上，TensorRT FP16实现了66.76-74.56 FPS的推理速度，延迟为13.41-14.98 ms，相比PyTorch FP32仅损失少量精度（mAP下降0.0020-0.0054）但能耗降低约64-66%。完整流水线达到44.96-54.11 FPS，在APPLE MOTS视频序列上，HOTA为0.345-0.538，计数F1分数为0.611-0.803。

Insight: 创新点在于构建了一个集成了硬件加速解码、ByteTrack跟踪和运动感知越线分析的完整嵌入式实时视频分析流水线，并系统评估了不同量化精度（FP32/FP16/INT8）的权衡。研究洞察表明，实用的边缘水果监测不仅需要高效的检测模型，还依赖于支持可靠时序关联的数据采集几何（如侧向视角优于正向穿越）。

Abstract: Static-image benchmarks do not capture the computational and temporal requirements of practical orchard video analytics. This study presents an end-to-end framework for real-time fruit detection, tracking, and counting on the NVIDIA Jetson Orin Nano Super. A lightweight YOLO26s detector is trained independently on four public datasets representing apples, mangoes, blueberries, and strawberries under a common protocol. The models are deployed on embedded platform using PyTorch and TensorRT at FP32, FP16, and INT8 precision. APPLE MOTS is then used for temporal video analytics because it provides orchard sequences with persistent fruit identities, enabling evaluation of multi-object tracking and unique-fruit counting. The selected FP16 TensorRT detector is integrated into an NVIDIA DeepStream pipeline combining hardware-accelerated decoding, ByteTrack tracking, and motion-aware line-crossing analytics. Across the four detection tasks, mean test mAP@50:95 ranges from 0.4957 to 0.8656. On the Jetson, TensorRT FP16 achieves 66.76-74.56 images/s at 13.41-14.98 ms prediction latency, while reducing mAP@50:95 by only 0.0020-0.0054 and gross energy consumption by approximately 64-66% relative to PyTorch FP32. The complete detector-tracker-analytics pipeline reaches 44.96-54.11 FPS and sustains the configured 30-FPS input rate without output-frame loss. On held-out orchard video sequences, HOTA ranges from 0.345 to 0.538, event-level counting F1 from 0.611 to 0.803, and relative count error from 6.2% to 51.6%. Performance varies across acquisition geometries: near-lateral row viewing yields the most stable tracking and counting, whereas forward traversal remains association- and recall-limited despite spatially adaptive counting geometry. These results show that practical edge-based fruit monitoring requires efficient detection and acquisition geometries that support reliable temporal association.

[31] Multimodal Foundation Models Adaptation based on Domain-Aware Relaxed Orthogonal Subspace for Remote Sensing cs.CVPDF

Han Luo, Ruoyu Yang, Yinhe Liu, Yanfei Zhong

TL;DR: 本文提出了一种名为域感知松弛正交子空间适应（DROS）的统一框架，用于解决预训练基础模型在遥感任务中因领域偏移而导致的子空间不匹配问题。该方法将低秩适应重新定义为数据条件化的子空间学习和灵活的子空间适应，并通过扩展到多模态设置（MM-DROS）来促进跨模态交互。

Details

Motivation: 预训练基础模型在计算机视觉中取得了显著成功，但其高微调成本限制了实际部署。现有的参数高效微调方法（如LoRA）在应用于领域偏移显著的遥感任务时，由于固定子空间未考虑下游激活分布，可能导致子空间不匹配，从而影响适应效果。

Result: 在多个遥感基准测试上的广泛实验表明，DROS实现了最先进的性能，甚至超过了全微调，且没有额外的推理开销。

Insight: 创新点在于将低秩适应重新定义为数据条件化的子空间学习，通过基于下游训练分布估计的二阶激活统计量来初始化权重分解，使其反映遥感数据实际诱导的特征几何结构，并采用松弛正交参数化实现灵活的几何变换。在多模态扩展中，通过跨模态特定子空间共享变换结构，实现了高效的跨模态交互。

Abstract: Pretrained foundation models (FMs) have achieved remarkable success in computer vision, yet their high fine-tuning cost limits practical deployment. Parameter-efficient fine-tuning (PEFT) methods such as Low-Rank Adaptation (LoRA) improve efficiency by constraining updates to a predefined low-rank subspace. However, when applied to remote sensing tasks with substantial domain shifts, the fixed subspace is constructed without observing the downstream activation distribution and can therefore provide a poor coordinate system for adaptation, a phenomenon herein termed subspace mismatch. To address this issue, a unified framework is introduced, termed Domain-aware Relaxed Orthogonal Subspace adaptation (DROS), which reformulates low-rank adaptation as data-conditioned subspace learning and flexible subspace adaptation. Specifically, the weight decomposition is conditioned on second-order activation statistics estimated from the downstream training distribution, so that the initialization reflects the feature geometry actually induced by the remote-sensing data, followed by flexible geometric transformations enabled by a relaxed orthogonal parameterization. Furthermore, the framework is extended to multimodal settings (MM-DROS) by sharing transformation structures across modality-specific subspaces, facilitating efficient cross-modal interaction. Extensive experiments on multiple remote sensing benchmarks demonstrate that DROS achieves state-of-the-art performance, even surpassing full fine-tuning, without additional inference overhead.

[32] MARC: Morphology-Aware Regression of Consensus for Cell Segmentation in Subcellular Spatial Transcriptomics cs.CVPDF

Xinyu Shu, Andrew Zhang, Jean Yang, Jinman Kim

TL;DR: 本文提出了MARC框架，用于预测亚细胞空间转录组学中细胞分割的多方法共识支持图。该框架通过留一法共识伪目标和前景-联合共识损失进行训练，能够在不执行多方法推理的情况下近似显式跨方法共识，从而支持大规模研究中的稳健分割评估。

Details

Motivation: 在亚细胞空间转录组学中，准确的细胞分割是构建细胞级表征的基础，但可靠的边界真值难以获取，且手动标注耗时。现有方法依赖多计算密集型流程来构建显式共识，效率低下。

Result: 在Xenium肾脏组织的4642个保留切片上评估，MARC相对于显式计算跨方法共识图，平均Dice得分为0.90，平均交并比为0.82，平均细胞级Spearman相关性为0.79。

Insight: 创新点包括使用留一法共识伪目标进行训练，以及前景-联合共识损失来聚焦监督；客观分析认为，该方法通过回归预测共识图，避免了多方法推理的计算开销，为大规模分割评估提供了高效解决方案。

Abstract: Accurate cell segmentation remains a major bottleneck in subcellular spatial transcriptomics (SST), in which morphological images and spatially resolved RNA transcripts are used to partition tissues into individual cellular instances. As segmentation serves as the foundation for constructing cell-level representations, boundary errors can lead to incorrect transcript assignments and compromise downstream analyses. However, reliable ground-truth boundaries are unavailable because they must be inferred from incomplete morphological and transcript signals. Furthermore, manual annotation of a large number of cells is time-consuming. Agreement among complementary segmentation methods provides a practical surrogate for identifying well-supported and ambiguous regions, but explicit consensus construction requires executing multiple computationally intensive pipelines. In this study, we propose MARC (Morphology-Aware Regression of Consensus), a framework that predicts a multi-method consensus-support map for SST segmentation. MARC is trained with leave-one-method-out consensus pseudo-targets and a Foreground-Union Consensus Loss that focuses supervision on candidate and consensus foreground. We evaluated MARC on 4,642 held-out tiles from Xenium kidney tissue, achieving a mean Dice score of 0.90, a mean intersection-over-union of 0.82, and a mean cell-level Spearman correlation of 0.79 against explicitly computed cross-method consensus maps. We demonstrate that the predicted consensus maps localise weakly supported regions while preserving consensus-based rankings and identifying low-consensus cells for manual review. These results show that MARC closely approximates explicit cross-method consensus without multi-method inference and therefore has the potential to facilitate robust, consensus-aware evaluation of cell segmentation in large-scale SST studies.

[33] FFVO: A Feedforward Pose Decoder for Long-Horizon Visual Odometry cs.CVPDF

Meng-Li Shih, Shih-Yang Su, Yuliang Zou, Hao Xiang, Haidong Zhu

TL;DR: 本文提出了一种名为FFVO的前馈视觉里程计方法，专门用于长序列视频中的相机位姿估计。该方法通过紧凑的相机令牌表示、分层局部到全局的时间解码器以及中间轨迹监督，解决了计算成本高、长上下文歧义和时间不稳定性等挑战。

Details

Motivation: 为了解决长视频中相机位姿估计面临的计算成本高、长上下文歧义和时间不稳定性等问题，作者旨在开发一种高效且时间稳定的前馈相机位姿解码器。

Result: 在Waymo Open Dataset、KITTI和一个大规模专有基准测试上的广泛评估表明，该方法优于现有的前馈方法，并显著减少了抖动和漂移。

Insight: 创新点包括使用紧凑的相机令牌表示进行高效时间聚合，采用分层局部到全局时间解码器以缓解几何歧义，以及引入中间轨迹监督来提升时间稳定性。这些设计使得FFVO在长时域视觉里程计中表现出色。

Abstract: Stable and reliable 4D spatial understanding is fundamental for autonomous driving systems. While feedforward reconstruction networks can estimate camera motion and 3D structure in one pass, pose estimation over long videos remains challenged by computational cost, long-context ambiguity, and temporal instability. To address these challenges, we propose Feedforward Visual Odometry (FFVO), a pose-specialized adaptation of joint reconstruction architectures for efficient and temporally stable camera-pose estimation. FFVO uses (i) a compact camera-token representation for computationally efficient temporal aggregation, (ii) a hierarchical local-to-global temporal decoder that mitigates geometric ambiguity by separating short-range motion aggregation from sequence-level integration, and (iii) intermediate trajectory supervision that promotes temporal stability. Extensive evaluation on the Waymo Open Dataset (WOD), KITTI, and a large-scale proprietary benchmark demonstrates that our method performs favorably against existing feedforward approaches, and greatly reduces jitter and drift. These results support FFVO as an effective feedforward camera-pose decoder in long-horizon visual odometry settings.

[34] Rank-Consistent Set Reasoning for Co-Salient Object Detection cs.CVPDF

Yuan Xiang, Matteo Rossi, Yingzhou Chen

TL;DR: 本文提出了一种名为Rank-Consistent Set Reasoning (RCSR)的监督式密集预测框架，用于协同显著目标检测（Co-SOD）。该方法将图像组建模为无序集合，通过在多尺度上对区域与学习到的组槽（group slots）的匹配程度进行排序，并使用稳健的修剪统计量聚合这些排序，以抑制偶然的成对匹配并防止异常图像主导共享表示。模型还引入了组置换目标和困难干扰物增强，以学习集合层面的目标属性。

Details

Motivation: 解决协同显著目标检测中需要同时考虑单张图像的显著性和图像组内一致性的问题，并避免现有方法可能依赖图像序列顺序、语义标签或外部模型（如自然语言分支、开放词汇检测器）的局限性。

Result: 在CoCA、CoSal2015和CoSOD3k数据集上进行了评估，并测试了模型对组大小鲁棒性、干扰物抑制、顺序不变性和跨数据集迁移的能力。摘要未提及具体的定量结果（如精度指标）或是否达到SOTA水平。

Insight: 创新点在于将图像组明确建模为无序集合，并引入基于排序一致性的集合推理机制（包括组槽、排序聚合、秩一致性门控）来捕获组级共性，同时通过组置换目标和数据增强防止模型记忆图像顺序或孤立视觉显著性，这为集合级别的视觉任务提供了新的建模思路。

Abstract: Co-salient object detection (Co-SOD) requires a model to find foreground regions that are salient in individual images and supported by the image group. We present \emph{Rank-Consistent Set Reasoning} (RCSR), a supervised dense-prediction framework that models a group as an unordered set rather than as a sequence of images or a semantic label. The core idea is to rank how strongly each spatial region agrees with a small collection of learned group slots at every image scale, and to aggregate these ranks with a robust trimmed statistic. This suppresses accidental pairwise matches and prevents one atypical group member from dominating the shared representation. A set encoder builds group slots directly from multi-scale visual features, while a rank-consistency gate measures whether the ordering of candidate regions is stable across group members. The gated slots are decoded jointly with per-image features to produce co-saliency maps. The model contains no natural-language branch, no open-vocabulary detector, and no external segmentation model. We further introduce a group permutation objective and hard-distractor augmentation so that the model learns the properties of a set-level target rather than memorizing image order or isolated visual saliency. We formulate an evaluation protocol for CoCA, CoSal2015, and CoSOD3k, together with tests of group-size robustness, distractor rejection, order invariance, and cross-dataset transfer.

[35] Hyper-LLaVA: Hyperbolic Uncertainty-aware Modality-Balanced Routing for Multimodal Continual Instruction Tuning cs.CVPDF

Kunlun Xu, Yanqin Zhang, Wenwen Qiang, Jiahuan Zhou

TL;DR: 本文提出了Hyper-LLaVA，一种用于多模态持续指令调优（MCIT）的超几何不确定性感知模态平衡路由方法。该方法通过超几何空间建模样本到任务分布的相似性来改进模态内任务匹配，并通过量化每个模态的任务匹配模糊度来实现模态间的自适应平衡，从而提升了参数路由能力。

Details

Motivation: 现有最先进的MCIT方法在参数路由时，依赖样本到任务中心的相似性和等权重的跨模态融合，存在两个根本缺陷：一是模态内样本到任务中心距离未能充分利用任务内多样性信息，二是不同模态在不同任务上的可靠性不同，具有任务间模糊性的模态容易误导路由结果。

Result: 论文提出的Hyper-LLaVA方法在性能上大幅超越了现有的最先进方法，取得了显著的提升。

Insight: 创新点在于引入了超几何空间来建模任务特征分布，以更好地捕获任务内多样性；同时，通过量化每个模态的任务匹配不确定性，实现了跨模态的自适应平衡路由，从而更鲁棒地处理多模态输入。

Abstract: Multimodal Continual Instruction Tuning (MCIT) aims to exploit the incrementally accumulated knowledge to process multimodal inputs of diverse tasks, where parameter routing plays an important role. State-of-the-art methods rely on sample-to-task center similarity and cross-modal fusion with equal weight during routing. However, such solutions face two fundamental flaws: (1) Within each modality, the sample-to-task center distance is sub-optimal for routing since the abundant intra-task diversity information is underleveraged. (2) Different modalities exhibit varying reliability across tasks, where the modality with inter-task ambiguity can easily misguide the routing result. To address these problems, we propose Hyperbolic Uncertainty-aware Modality-Balanced Routing (Hyper-LLaVA) to improve parameter routing capacity based on cross-modality task feature uncertainty modeling. Specifically, to improve intra-modality task matching, Hyper-LLaVA accesses the sample-to-task distribution similarity in the Hyperbolic space. Besides, to alleviate the degradation brought by unreliable modalities, Hyper-LLaVA quantifies the task matching ambiguity within each modality to achieve adaptive balancing between task matching across modalities. Based on the complementary intra- and inter-modality task matching enhancement, our Hyper-LLaVA outperforms state-of-the-art approaches by large margins. Our source code is available at https://github.com/zhoujiahuan1991/ICML2026-Hyper-LLaVA

[36] Restore What Matters: Lessons from Joint Restoration and Recognition cs.CVPDF

Lanqing Guo, Xijun Wang, Minchul Kim, Yu Yuan, Wes Robbins

TL;DR: 本文提出了一种联合恢复与识别（JR^2）的新范式，旨在解决传统‘先恢复后识别’流程中视觉质量提升与识别性能脱节的问题。该方法基于物理模拟、神经科学原理和视觉学习，通过任务信号指导模型仅恢复下游任务真正需要的部分，实现选择性恢复。

Details

Motivation: 传统识别流程通常采用‘先恢复后识别’的工作流，但视觉上令人愉悦的图像往往不能带来识别性能的提升。本文旨在打破这一局限，提出一种任务驱动的恢复方法，使恢复过程直接服务于识别任务。

Result: 在IARPA-BRIAR基准测试上，该方法取得了稳定的性能提升（例如，TAR@0.01% FAR提升0.6，FNIR@1% FPIR降低2.5）。同时，通过质量门控机制跳过了约70%的干净帧，有效降低了计算成本。

Insight: 论文的核心创新在于提出了任务驱动的选择性恢复范式，将物理先验、神经科学中的选择性注意机制与端到端的识别损失相结合。其关键见解是：更好的视觉质量既非必要也非充分条件，恢复模块必须是任务驱动、选择性的且具有物理感知能力的。

Abstract: Recognition pipelines typically adopt a restore-then-recognize workflow, yet decades of experience show that generating visually pleasing images seldom translates to improved recognition. We propose a Joint Restoration-for-Recognition (JR$^2$) paradigm: restore only what downstream tasks truly require, with task signals dictating where, how much, and whether restoration is necessary. JR$^2$ rests on three pillars: (i) Physics, employing optics-accurate turbulence simulation, extensible to blur and noise, to ground restoration in real image formation; (ii) Neuroscience, drawing on selective attention and neuroplasticity to direct model capacity toward identity-critical regions and frames while bypassing already-clean inputs; and (iii) Vision & Learning, coupling recognition loss end-to-end through restoration and alignment so that low-level edits maximize high-level identity stability. Evaluations on IARPA-BRIAR show consistent improvements (e.g., TAR@0.01% FAR +0.6; FNIR@1% FPIR -2.5), while a quality gate skips ~70% of clean frames, reducing cost. Ablations confirm physics priors enhance realism, joint training prevents catastrophic forgetting, and selective restoration suffices in many cases. We conclude that better-looking images are neither necessary nor sufficient; restoration modules must be task-driven, selective, and physically aware. Code, pretrained models, and recipes are provided for integration.

[37] StepPrune: Adaptive Sequential Visual Token Selection across Multimodal Large Language Models cs.CVPDF

Hansen Zhang, Landi He, Mingde Yao, Lijian Xu

TL;DR: 本文提出StepPrune方法，将多模态大语言模型中的视觉令牌剪枝问题建模为一个自适应的顺序决策过程。该方法根据先前选择的令牌和文本上下文逐步构建保留子集，并通过学习到的STOP动作自动确定其大小，实现了对视觉复杂度的自适应和依赖上下文的令牌选择。

Details

Motivation: 现有视觉令牌剪枝方法（如top-K）通常独立评估令牌并对所有输入应用统一的预算，忽略了选择依赖的交互以及不同样本间视觉复杂度的差异，导致计算效率低下或性能损失。

Result: 在LLaVA-1.5、LLaVA-NeXT、Qwen2.5-VL和InternVL3等模型上的实验表明，StepPrune在LLaVA-1.5、Qwen2.5-VL和InternVL3上所有评估剪枝率下取得了最佳的平均归一化性能保持率。在LLaVA-1.5上，剪枝88.9%视觉令牌时仍能保持94.6%的完整前缀归一化性能，并将预填充延迟从59.95 ms减少到40.05 ms，实现了1.50倍的预填充加速。

Insight: 主要创新点包括：1）将视觉令牌剪枝建模为顺序决策过程，考虑了令牌间的依赖关系；2）引入可学习的STOP动作实现自适应预算分配；3）提出方差保持噪声门作为离散选择过程的可微分替代，便于训练；4）分组选择机制支持高分辨率输入处理。

Abstract: Visual prefixes account for a major portion of the per-layer computation in multimodal large language models (MLLMs), making visual-token pruning a direct approach to accelerating inference. Existing top-K methods typically evaluate tokens independently and apply a uniform budget to all inputs, overlooking both selection-dependent interactions and variations in visual complexity across samples. In contrast, we propose StepPrune, which formulates visual-token pruning as an adaptive sequential decision process. Conditioned on previously selected tokens and textual context, StepPrune progressively constructs the retained subset and automatically determines its size through a learned STOP action. During training, a variance-preserving noise gate provides a differentiable surrogate for the discrete selection process, whereas during inference, unselected tokens are physically removed before language-model prefill. A grouped selection mechanism further extends StepPrune to high-resolution inputs. Experiments across LLaVA-1.5, LLaVA-NeXT, Qwen2.5-VL, and InternVL3 show that StepPrune achieves the best average normalized performance retention across all evaluated pruning rates on LLaVA-1.5, Qwen2.5-VL, and InternVL3, while remaining competitive on the substantially longer AnyRes prefixes of LLaVA-NeXT. On LLaVA-1.5, StepPrune retains 94.6% of the full-prefix normalized performance while pruning 88.9% of the visual tokens. At a mean retained count of 64, StepPrune reduces prefill latency from 59.95 ms to 40.05 ms, corresponding to a 1.50x prefill speed-up.

[38] Realtime-Venus: A full-duplex interaction system with asynchronous delegation cs.CV | eess.ASPDF

Ruixiang Zhao, Hualei Wang, Renhe Sun, Enzhi Zhou, Jincenzi Wu

TL;DR: 论文提出了Realtime-Venus，一个支持异步委托的实时全双工交互系统。该系统包含两个独立的90亿参数模型：Realtime-Venus-Omni用于视听交互，Realtime-Venus-Audio用于语音交互。系统采用双循环运行时架构，在维持前台连续交互的同时，异步执行后台推理和工具调用任务。

Details

Motivation: 为了解决数字和物理环境中自然交互所需的连续感知与即时响应问题，特别是需要将对话动态地锚定在不断变化的视觉和听觉上下文中的挑战。

Result: 在评估的在线模型中，Realtime-Venus-Omni在八个视频基准测试中的六个上取得了最高分，包括StreamingBench (70.2%)和Daily-Omni (81.3%)。Realtime-Venus-Audio在多个音频理解和口语问答基准上领先，并在Full-Duplex-Bench v1.5上对75%的用户打断做出响应，其续话率指标超过了Gemini 3.1 Live和GPT-4o。

Insight: 核心创新在于将全双工交互、异步任务委托与连续感知整合到一个统一的因果时间线框架中。通过分离的视听与纯音频模型以及双循环运行时，实现了前台不间断交互与后台异步执行的协同，这为构建更自然、响应更及时的智能体系统提供了新架构。

Abstract: Natural interaction in digital and physical environments requires continuous perception and timely responses. Spoken dialogue relies on acoustic and linguistic cues, while video interaction also requires grounding the conversation in evolving visual context. We present Realtime-Venus, a proactive full-duplex interaction system with two separately trained 9B models: Realtime-Venus-Omni for audio-visual interaction and Realtime-Venus-Audio for spoken interaction. Each model serves as a complete conversational frontend, integrating continuous perception, conversational control, and native speech generation through a shared causal timeline for user inputs, model outputs, and delegation events. A dual-loop runtime coordinates live interaction with background reasoning and tool execution. Foreground interaction continues while Realtime-Venus-Harness executes tasks asynchronously and returns results for integration into the ongoing dialogue. Both models follow a common post-training recipe combining offline understanding, proactive full-duplex trajectories, and delegation workflows. Among the evaluated online models, Realtime-Venus-Omni achieves the highest scores on six of eight video benchmarks, including StreamingBench (70.2%), OVO-Bench (64.7%), and Daily-Omni (81.3%). Across eight audio understanding and spoken question answering benchmarks, Realtime-Venus-Audio leads the compared models on MMAU (78.0%), MMAU-Pro (63.2%), Llama Questions (83.8%), and Speech CMMLU (67.8%), while matching the best VoiceBench AlpacaEval score of 4.81. On Full-Duplex-Bench v1.5, Realtime-Venus-Audio responds to 75% of user interruptions and achieves continuation rates of 97%, 88%, and 86% under backchannels, other-directed speech, and background speech, respectively, exceeding Gemini 3.1 Live and GPT-4o on all three continuation metrics.

[39] Beyond OCR Accuracy: Text-Centric VQA Under Image Degradation with Modular and End-to-End cs.CVPDF

Ritali Vatsi, Rachapudi Jagadeesh, Shruti Singh Baghel, Himani Sharma, Amit Shukla

TL;DR: 本文对基于模块化OCR流水线和端到端视觉语言模型在图像退化条件下的文本中心视觉问答（VQA）性能进行了实证比较。研究发现，经过微调的模块化流水线在精确匹配准确率上显著优于端到端基线，同时指出传统OCR错误率指标不能可靠预测下游VQA性能，因为语义推理可以补偿识别失败。

Details

Motivation: 解决现实世界中图像退化（如运动模糊、低分辨率、压缩伪影）对文本中心VQA任务的影响，并比较模块化OCR流水线与端到端模型在此条件下的鲁棒性差异。

Result: 在4013张退化图像和7000个问答对上的评估显示，微调后的模块化流水线达到57.50%的精确匹配准确率，而端到端基线仅为38.00%，领域特定微调带来最高29.50个百分点的提升。

Insight: 创新点在于揭示了语义推理能补偿OCR识别错误，使得传统OCR错误率指标（如字符错误率、词错误率）无法有效预测VQA性能；同时强调了在真实视觉条件下进行任务感知评估的重要性，并提出了结合ResNet-18、自注意力空间建模和可变形卷积的自定义检测器架构SA-DBNet用于模块化流水线。

Abstract: Text-centric Visual Question Answering (VQA) requires reading and reasoning over text embedded in images, a task made substantially harder when images suffer from real-world degradation such as motion blur, low resolution, or compression artifacts. While modular OCR-based pipelines and end-to-end vision-language models are both widely used for this task, their comparative robustness under degraded conditions remains underexplored. We present an empirical study comparing two modular pipelines with SA-DBNet, a custom detector architecture combining ResNet-18 with self-attention spatial modeling and deformable convolutions against an end-to-end vision-language baseline, evaluated on 4013 degraded images with 7000 question-answer pairs. Fine-tuned modular pipelines achieve up to 57.50% exact-match accuracy versus 38.00% for the end-to-end baseline, with domain-specific fine-tuning yielding a gain of up to 29.50 percentage points. Critically, we find that conventional OCR error metrics like Character Error Rate and Word Error Rate are unreliable predictors of downstream VQA performance, as semantic reasoning can compensate for recognition failures when contextual cues are present. These findings highlight the importance of task-aware evaluation for text-centric VQA systems under realistic visual conditions. Codes are available here

[40] DiVA: Enabling Interactive Digital Life Simulation via Video Models cs.CVPDF

Cheng Chen, Hao Ouyang, Qiuyu Wang, Ka Leong Cheng, Wen Wang

TL;DR: DiVA是一个深度交互的数字生活模拟器，它通过结合多模态大语言模型作为路由器和精心设计的堆叠视频管道，实现了与数字角色进行长期、开放式、多轮交互的体验。其核心创新在于将生成过程建模为等待视频、动作视频及其间过渡的三部分耦合系统，并引入锚定视频延续模块来确保平滑过渡和长期视觉质量。

Details

Motivation: 为了解决当前数字角色交互体验中存在的长期连续性差、视觉质量退化（如相机抖动、不一致性）以及难以处理复杂姿态变化（如从坐到站）等问题，旨在创建一个能够维持高保真身份、连贯性和动态性的可持续交互模拟系统。

Result: 通过将核心生成模块替换为主流的长视频生成、延续和插值方法进行综合比较，并进一步与额外的长格式音频驱动化身模型对比，结果表明DiVA在维持长期视觉质量和真实感方面显著优于现有方法，验证了其作为可持续交互模拟的有效性。

Insight: 主要创新点在于将交互视频生成分解为等待、动作和过渡三个耦合阶段，并设计了锚定视频延续模块来编码前序动作信息以稳定角色状态，从而确保平滑过渡和长期一致性。这种系统设计有效解决了音频驱动模型中常见的视觉退化问题，并支持复杂的姿态变化。

Abstract: We present DiVA, a deeply interactive digital life simulator pioneering a new paradigm for long-term, open-ended interactive experiences within digital character worlds. DiVA’s architecture pairs a Multimodal Large Language Model (MLLM) as a router with a meticulously designed stacked video pipeline for seamless, multi-turn interactions with action and audio response. To maintain continuity and avoid degradation, we model generation as a three-part coupled system: waiting video, action video, and the transitions between them. These transitions are critically handled by our Anchored Video Continuation (AVC) module, which returns the character to stable states to prevent degradation. By encoding information from the preceding action video segment, AVC ensures smooth transitions, significantly reducing camera jitter and inconsistencies common in current video transition methods. This design also enables complex pose changes (e.g., sitting to standing) typically difficult for audio-driven models. These system designs together ensure high-fidelity identity, coherence, and dynamics for extended experiences. To validate our pipeline design, we comprehensively compare our system against alternatives by replacing our core generation module with mainstream long-video, continuation, and interpolation methods. We further analyze the necessity of the three-stage design, anchor-state selection, transition naturalness, spatial grounding, and the quality-latency trade-off, and we expand the comparison to additional long-form audio-driven avatar models. Results confirm DiVA is markedly superior in maintaining long-term visual quality and realism, validating its effectiveness as a sustainable, interactive simulation.

[41] Physically Typed and Geometry-Aware Representations for Earth Foundation Models cs.CV | cs.LGPDF

Rajiv Ranjan

TL;DR: 本文探讨了地球基础模型是否应显式保留物理场类型（如标量场、矢量场、张量场）的几何变换特性，而非仅依赖标准嵌入与数据增强。核心贡献是一个分阶段的证伪研究计划：首先在计算可控的ERA5数据集上比较传统嵌入、增强匹配嵌入、类型等变嵌入及Hodge/Helmholtz变体在时空、方向及低数据条件下的性能，仅当显式几何类型化带来可复现改进时，才推进到多模态地球基础模型的构建。

Details

Motivation: 当前地球观测基础模型学习到的高维语义嵌入缺乏固有的物理变换规律，而物理场（标量、矢量、张量等）在旋转、反射和局部坐标系变换下行为不同。本文旨在验证显式保持这些几何类型区分是否能带来实际性能提升，而非假设更复杂的架构必然更好。

Result: 摘要未提供具体定量结果，但提出了一个评估框架：将在ERA5数据集上系统比较传统嵌入、增强匹配、类型等变及Hodge/Helmholtz变体模型在空间、时间、方向变换及低数据场景下的表现，以确定显式几何类型化是否产生可复现的改进。

Insight: 创新点在于提出了一个分阶段、可证伪的研究方法，而非直接设计复杂架构；强调将物理场的几何类型（如标量、极矢量、轴矢量、协向量）显式融入基础模型，以探索其相对于现有仅考虑球面几何或隐式学习的方法是否具有实际优势。

Abstract: Earth-observation (EO) foundation models have become exceptionally effective at learning se mantic, high-dimensional geospatial embeddings, while modern weather and climate models have demonstrated that Earth-specific geometry, spherical operators, meshes, and hybrid physical solvers can materially improve prediction. Yet these two advances are not equivalent. A conventional latent embedding has no inherent physical transformation law, whereas scalar fields, tangent polar-vector fields, axial/pseudovector quantities, covectors, and higher-order tensors transform differently under rotations, reflections, and changes of local coordinate frame. This proposal asks whether a general purpose Earth foundation model should preserve those distinctions explicitly, or whether standard embeddings plus augmentation already learn everything that matters. The central contribution is therefore not a more complicated architecture by assumption, but a staged falsification program. A compute-conscious ERA5 dry run first compares conventional, augmentation-matched, typed equivariant, and Hodge/Helmholtz variants under spatial, temporal, orientation, and low-data shifts. Only if explicit geometric typing yields reproducible improvements does the program advance toward a multimodal Earth foundation model in which semantic embeddings coexist with physically typed fields. The proposed gap is narrower and more defensible than claiming that current models ignore geometry entirely: several systems already respect spherical domain geometry, and emerging work explicitly learns scalar/vector fields on spheres. The unresolved question is whether foundation-scale, multimodal, parity-aware field typing produces practical gains beyond those existing approaches.

Flavian Pegado, Ronit Hire, Shreyas Rajesh, Soham Phade

TL;DR: 本文提出了BLInD模型，这是一个仅依赖车辆历史状态（如速度、曲率、转向灯、车辆类型）来预测未来自车轨迹的多模态概率分布的紧凑网络。它无需摄像头、激光雷达、地图或物体跟踪输入，具有低延迟特性，适用于安全关键部署。模型在多个公开数据集上验证了其泛化能力，并集成到自动紧急制动（AEB）任务中，显著降低了误报率。

Details

Motivation: 解决现有学习型分布模型依赖场景上下文，而仅基于车辆状态的模型通常只能预测单一路径的问题，旨在开发一个不依赖外部传感器、能同时实现多模态预测、泛化能力强且适合实时部署的驾驶员意图预测模型。

Result: 在Wayve、Waymo和nuScenes数据集上，流匹配模型取得了最佳Top-k ADE/FDE结果（例如Wayve上为0.15/0.37米），自回归模型也达到了相当的覆盖率。在AEB触发任务中，与基线策略相比，AR模型将误报率降低了13.7倍（至0.11%），流匹配模型降低了25.1倍（至0.06%），同时保持了高真阳性率。模型在NVIDIA DRIVE Orin ECU上的推理时间分别为0.87毫秒（AR）和2.9毫秒（流匹配）。

Insight: 创新点在于首次同时实现了仅基于车辆历史状态（blind）、学习得到的多模态轨迹分布，并证明了其跨域泛化能力。这为下游系统（如AEB）提供了一个可控且合理的意图采样接口，其紧凑和低延迟的设计使其非常适合在汽车ECU上实时部署。

Abstract: We present BLInD (Blind Learned Intent Distribution), a compact network that maps recent vehicle-state history (e.g. speed, curvature, indicator, and vehicle type) to a top-k distribution of future ego trajectories, with no camera, LiDAR, map, or object-track inputs. We find that vehiclestate history alone is sufficient to learn a useful multimodal distribution over near-term ego trajectories, and its low-latency nature makes it well-suited for safety-critical deployment. We investigate two distribution architectures, autoregressive (AR) and flow-matching, and train on both mixed-platform opensource and Wayve datasets. Both generalize without datasetspecific adaptation; the flow-matching model achieves best topk ADE/FDE of 0.15/0.37 m on Wayve, 0.15/0.36 m on Waymo, and 0.28/0.59 m on nuScenes, with the AR model reaching comparable coverage. Integrating the distributions into an AEB trigger task, a strict all-candidates policy reduces false positives from 1.51% to 0.11% with AR (13.7x reduction, 94.9% TP) and to 0.06% with flow-matching (25.1x reduction, 98.7% TP) compared to a 1-CTRV policy with 100% true positive score. BLInD runs in 0.87 ms with the AR head and 2.9 ms with the flow-matching head on an NVIDIA DRIVE Orin ECU making it compatible with real-time deployment on automotive ECUs. While existing learned distribution models rely on scene context and blind vehicle-state models typically collapse to a single path, BLInD is learned, blind, and cross-domain simultaneously, a combination not demonstrated by prior work. These results show that such a distribution provides a controllable and plausible intent sampling interface for downstream systems, with AEB as one instantiation.

[43] SGWIB:Sliced Gromov-Wasserstein Information Bottleneck for Video Highlight Detection cs.CV | cs.AIPDF

Hanjuan Huang, Yung-Chieh Yeh, Hsing-Kuo Pao

TL;DR: 本文提出了一种名为SGWIB的视频高光检测方法，该方法基于信息瓶颈原则，通过引入切片Gromov-Monge间隙（SGMG）正则化器来保持片段间的时间结构，并利用上下文解耦模块减少特定领域（如体育）的上下文偏差。

Details

Motivation: 视频高光检测需要既学习判别性的片段表示，又保持片段间的时间关系。传统信息瓶颈方法直接应用于此任务会忽略片段间的关系结构并扭曲时间组织，因此需要一种结构感知的表示学习方法。

Result: 在MrHiSum和MoSu数据集上的实验表明，SGWIB在Kendall’s tau、Spearman’s rho、mAP@50和mAP@30指标上均优于其他单模态方法，在MrHiSum数据集上视觉模型将先前最佳结果分别提升了0.031、0.031、0.87和0.75，达到了新的SOTA水平。

Insight: 创新点在于提出了结构感知的SGMG正则化器来在信息瓶颈框架中保持时间结构，并设计了上下文解耦模块以减少领域特定偏差，这为视频理解任务中学习结构保持的紧凑表示提供了新思路。

Abstract: Video highlight detection aims to identify temporally important segments that capture the most informative or engaging events in a video. Reliable prediction therefore requires not only discriminative segment representations but also preservation of the temporal relationships among neighboring and distant segments. The information bottleneck principle has proven effective for learning compact and task-relevant representations, yet it has not been explored for video highlight detection, and applying conventional formulations directly would overlook inter-segment relational structure and distort highlight relevant temporal organization during compression. We therefore introduce the Sliced Gromov-Monge Gap (SGMG), a structure aware regularizer that measures the excess relational distortion induced by a prescribed source-to-bottleneck mapping relative to an optimal sliced structural correspondence. Building on SGMG, we develop SGWIB, an information-bottleneck framework for single-modal video highlight detection that learns compact bottleneck representations while preserving inter-segment temporal structure. We further introduce Home-Away-Related Contextual Pseudo-Labels and a contextual disentanglement module that reduce sports-specific contextual bias by separating highlight oriented information from contextual patterns. Experiments on MrHiSum and MoSu show that SGWIB attains the best Kendall’s tau, Spearman’s rho, mAP@50, and mAP@30 among the compared single-modal methods on both datasets. On MrHiSum, the visual model improves the strongest previous results by 0.031, 0.031, 0.87, and 0.75 on these four metrics, respectively. These results show that structure-aware information-bottleneck regularization combined with contextual disentanglement improves segment-level highlight prediction.

[44] LPA-CWM: A Learned Physical Adjudicator for Motion Reasoning with Counterfactual World Models cs.CV | cs.AI | cs.ROPDF

Kunwei Wu, Xiang Liu, Guocai Yao, Junming Chen, Zhikang Chen

TL;DR: 本文提出LPA-CWM，一种用于反事实世界模型中运动推理的学习物理仲裁器。该方法通过轻量级学习模块评估不同掩码干预下生成响应的可靠性，并加权聚合以提升运动轨迹提取的准确性。

Details

Motivation: 现有反事实世界模型在聚合不同目标帧掩码下的预测响应时，采用均匀加权忽略了响应可靠性的差异，导致运动推理性能受限。

Result: 在DAVIS和Kinetics子集上，LPA-CWM相比均匀加权的CWM分别将平均DCA指标提升了60.0%和29.0%，并在TAP-Vid First基准上提高了跟踪精度。

Insight: 创新点包括将响应聚合建模为候选可靠性学习问题，引入轻量级学习仲裁器；同时提出了完整性感知运动对应评估协议，综合衡量轨迹的定位、完整性、可见性与连续性。

Abstract: Counterfactual world models (CWM) extract motion from pretrained video predictors by comparing factual and intervened predictions. However, responses generated under different target-frame masks vary in reliability, while uniform aggregation weights them equally. We formulate response aggregation as candidate reliability learning and propose LPA-CWM with a lightweight Learned Physical Adjudicator (LPA). Trained on dense MOVi-F trajectories, the 3.0M-parameter LPA compares visual context and response structure across an unordered candidate set to predict relative weights, while the CWM predictor and intervention generator remain frozen. The weighted responses undergo windowed localization and one paired re-evaluation to recover motion. We also introduce Completeness-aware Motion Correspondence (CMC), a ground-truth-anchored evaluation protocol that jointly measures localization, trajectory completeness, visibility, and continuity, counting missing predictions as failures on visible dynamic points. On the evaluated DAVIS and Kinetics subsets, LPA-CWM improves $\mathrm{DCA}_{\mathrm{avg}}$ over Uniform CWM by 60.0% and 29.0%, respectively, and also improves tracking accuracy under TAP-Vid First. A quick overview is available at https://LPA-CWM.github.io.

[45] RA-CoA: Training-free Fashion Image Captioning via Retrieval-Augmented Chain-of-Attributes cs.CV | cs.AI | cs.LGPDF

Abhirama Subramanyam Penamakuri, Shreya Shukla, Anand Mishra

TL;DR: 本文提出了一种无需训练的时尚图像描述框架RA-CoA，通过检索增强的属性链方法，将时尚图像描述分解为检索相关属性集和属性级推理两个阶段，以提升细粒度属性描述的准确性。

Details

Motivation: 针对时尚图像描述需要细粒度视觉推理和领域特定术语，以及时尚库存快速更新带来的可扩展性挑战，现有通用视觉语言模型在属性覆盖和精度上不足，易产生幻觉或误识别细节。

Result: 在不同视觉语言模型家族和提示范式下的广泛评估表明，RA-CoA显著提升了描述质量，在METEOR分数上平均比零样本描述提高了26.3%。

Insight: 创新点在于将时尚图像描述解耦为检索和推理两个可解释阶段，无需微调即可提升细粒度属性精度；客观分析认为其模型无关性和训练免费设计增强了实际应用的适应性和可扩展性。

Abstract: Fashion Image Captioning (FIC) plays a vital role in enhancing user experience and product search in e-commerce platforms. Unlike natural scene image captioning, FIC requires fine-grained visual reasoning and knowledge of domain-specific terminology to capture subtle attributes such as neckline and closure types, graphic patterns, and dress silhouettes. Moreover, as fashion inventories evolve rapidly with new trends, styles, and frequently emerging vocabulary, developing training-free captioning solution becomes essential for scalability and real-world adaptability. Instruction-tuned vision-language models (VLMs) offer a promising solution to fashion image captioning dueto their strong zero-shot capabilities and natural language fluency. However, these general-purpose models often lack attribute-level coverage and precision, and tend to hallucinate or misidentify fine-grained fashion details, making them less suitable for high-fidelity applications like product cataloging or personalized recommendations. To address this, we propose RA-CoA (Retrieval-Augmented Chain-of-Attributes), a novel, training-free framework that disentangles fashion image captioning into two interpretable stages: (i) retrieval of relevant attribute sets from a product knowledge base, and (ii) attribute-level reasoning to generate the final caption. RA-CoA is a model-agnostic approach that works with frozen VLMs to improve fine-grained attribute precision in product captions without the need for fine-tuning. Extensive evaluations across diverse VLM model families under different prompting paradigms demonstrate that RA-CoA significantly improves caption quality, achieving an average gain of 26.3% METEOR score over zero-shot captioning. We make our code publicly available.

[46] Adapting Open-Weight MLLMs to Generate Point Prompts for Electron Microscopy Segmentation cs.CVPDF

Samia Mohinta, Albert Cardona

TL;DR: 该论文探索了利用开源多模态大语言模型（MLLMs）为电子显微镜图像分割自动生成点提示的方法。通过将图像和自然语言指令与掩码质心坐标配对，训练LoRA适配器，使冻结的MLLM能够根据语言请求输出坐标，进而驱动微SAM分割器。实验表明，经过微调的Qwen3-VL等模型在分割性能上显著提升，并能泛化到未见数据集和语言表述。

Details

Motivation: 解决在电子显微镜图像分割中，如何自动化生成点提示以替代人工输入的问题，探索开源MLLMs作为可解释的、语言驱动的定位与分割桥梁的可行性。

Result: 在三个线粒体数据集上，经过监督微调和奖励优化的Qwen3-VL模型将分割AP50从0.247提升至0.736，接近自动提示生成（APG）的0.773；其他MLLMs也达到或超过APG水平。虽然性能仍低于专用监督检测器（AP50 0.904），但MLLMs生成的坐标更接近标注点集和实例数量，且展现出良好的跨数据集和语言表述泛化能力。

Insight: 创新点在于首次将开源MLLMs适配为电子显微镜图像的点提示生成器，通过LoRA微调实现语言指令到坐标的映射，构建了一个可检查的、语言导向的定位-解码链路，为自动化生物医学图像分析提供了新思路。

Abstract: Promptable models such as microSAM segment electron microscopy (EM) images from point prompts, but automation requires generating prompts without user input. We ask whether open-weight multimodal large language models (MLLMs) can generate them from natural-language requests by returning coordinates to a frozen segmenter. To that end, we convert masks from three mitochondria datasets into training examples, pairing images and instructions with centroid coordinates, then train LoRA adapters while freezing the MLLM backbone and microSAM. We find that Qwen3-VL reaches segmentation AP${50}$ $0.736$ after supervised fine-tuning and reward optimization, up from $0.247$ without adaptation, while automatic prompt generation (APG) achieves $0.773$. In addition, two other MLLMs improve, reaching or exceeding APG. When compared with a supervised centroid-heatmap detector that reaches AP${50}$ $0.904$ for this mitochondria task, Qwen3-VL more closely matches the annotated point set and instance counts. Moreover, training on two public datasets transfers to an unseen third, while training on all three transfers to an independent EM volume. Robustness tests show stable performance under unseen formulations of the natural-language request, while the coordinates can be reused by a second segmenter. To our knowledge, this is the first feasibility study of open-weight MLLMs as EM point generators, providing an inspectable, language-directed link between localization and mask decoding.

[47] Talking to Me or Someone Else? Rethinking Talk-to-Me Detection in Egocentric Videos cs.CV | cs.AIPDF

Feiyu Du, Xi He, Jia Li, Yapeng Tian, Weili Wu

TL;DR: 本文重新审视了第一人称视角视频中的‘对我说话’检测问题，将其重新定义为在线、帧级别的预测任务，并考虑了多种非TTM状态。作者构建了一个包含约90万帧标注的在线TTM数据集，并提出了一个融合多模态社交线索的模型。

Details

Motivation: 现有TTM研究多为离线的片段级识别，与在线交互需求不匹配，且忽略了第一人称视频中多样化的非TTM说话状态。

Result: 在扩展的Ego4D社交交互基准上，提出的多模态模型（融合音频、视觉和语音语义线索）在TTM检测上达到了75.5%的帧级F1分数，优于强基线。

Insight: 将TTM检测从二元分类重构为包含多种非TTM状态（如对他人说话、自言自语、背景）的在线帧级任务，并构建相应数据集；通过融合多模态社交线索提升性能，为分析不同说话状态对识别的影响提供了系统方法。

Abstract: Online understanding of who is talking to the camera wearer is a key capability for egocentric social interaction. However, existing talk-to-me (TTM) studies are commonly formulated as offline clip-level recognition, which is poorly aligned with online interaction and overlooks the diverse non-TTM speaking states that naturally arise in egocentric videos. In this paper, we revisit this problem by reformulating it as an online, frame-level prediction task. Instead of treating TTM as a binary problem against a single negative class, we model it in the presence of diverse and previously underexplored non-TTM states, such as talking-to-others, self-talking, and background conditions. To support this new formulation, we construct an Online TTM Dataset consisting of 406 egocentric video clips with approximately 900K annotated frames, each labeled with frame-level social interaction categories (e.g., background, TTM, talking-to-others, self-talking), by extending the Ego4D social interaction benchmark. In this benchmark, we evaluate five adapted baselines and develop a new model that integrates social cues across modalities. Experimental results show that our multimodal model, which jointly leverages audio, visual, and speech-semantic cues, achieves 75.5% frame-level F1 on TTM, outperforming strong baselines and enabling a systematic analysis of how different speaking states affect TTM recognition.

[48] SignMimic: Robust High-Quality Sign Language Motion Generation via Human-Shape-Oblivious Pose Transfer Guidance cs.CVPDF

Zhewen He, Junyi Yu, Haomian Huang, Zhenhua Li, Yi Fang

TL;DR: 本文提出SignMimic方法，用于解决手语视频模仿生成任务：给定驱动视频和单个参考帧，合成目标手语者复现源动作的视频，同时保持身份和语言形式。该方法通过分解刚性运动、非刚性变形和视角相关补全三个子问题，结合几何与语言先验，实现了高质量、稳定的手语动作生成。

Details

Motivation: 现有手语视频生成方法将刚性运动、非刚性变形和视角相关补全耦合在单一生成器中，导致手形漂移和时空不稳定问题。本文旨在通过解耦这些因素，提升生成视频的质量和稳定性。

Result: 在ASL 50K、How2Sign、CSL News等多个大规模数据集上，SignMimic在视频质量、身份相似性和帧连续性方面达到SOTA水平，且在生成视频上进行反向翻译（SLT）时损失最小。消融实验验证了刚性正则化、非刚性适应和补全模块的作用。

Insight: 创新点包括：（1）基于TNet的SE(3)刚性正则化稳定全局姿态；（2）在正则空间通过NIF2D进行非刚性适应以保持手/脸等精细发音器官和协同发音；（3）在条件视频扩散前使用Pose-MAE风格补全。这种分解式架构为动作生成任务提供了可借鉴的模块化设计思路。

Abstract: We study the challenge of sign language video mimicking: given a driving video and a single reference frame, synthesize a video where the target signer reproduces the source motion while preserving identity and linguistic form. Prior pipelines entangle rigid motion, non-rigid deformation, and view-dependent completion in a monolithic generator, causing handshape drift and spatio-temporal instability. We present SignMimic, which (i) applies a TNet-based model to study SE(3) rigid canonicalization to stabilize global pose, (ii) performs non-rigid adaptation in a canonical space to preserve fine-grained articulators (hands/face) and coarticulation via NIF2D, and (iii) uses Pose-MAE-style completion before conditional video diffusion. This factorization injects geometric and linguistic priors, yielding shape and spatio-temporal consistency. On several large-scale datasets (ASL 50K, How2Sign, CSL News), SignMimic achieves state-of-the-art-level performance on video quality, identity similarity, and frame continuity while also achieving minimal loss when performing back translation (SLT) on generated videos. Ablations confirm the role of rigid canonicalization, non-rigid adaptation, and completion. Code, model checkpoints, and video examples will be released.

[49] SPARK: Representation-Level KV Memory Alignment for Safer Vision-Language Models cs.CVPDF

Mohd Azfar, Izhar Dad Khan

TL;DR: 论文提出了SPARK框架，旨在通过修复视觉语言模型（VLM）预填充阶段形成的多模态键值（KV）记忆来增强模型安全性，而无需在推理时修改模型参数。该方法通过两阶段过程识别并投影掉与有害意图相关的表征方向，并混合修复后和原始的记忆，从而有效抵御跨模态的越狱攻击。

Details

Motivation: 视觉语言模型（VLMs）容易受到将有害意图分布在文本和图像中的越狱攻击，使得单模态安全机制失效。论文旨在探究是否可以直接在多模态KV记忆中缓解这种漏洞，而不在推理时修改模型参数。

Result: 在LLaVA-OneVision-7B、Chameleon-7B、Qwen2-VL-7B和InternVL2-4B等多个模型上，SPARK显著降低了多模态攻击成功率。例如，在LLaVA-OneVision-7B上，纯图像越狱攻击成功率降至4.7%，而MMMU性能仅下降0.6个百分点（47.8 vs. 48.4），语言质量接近基线。在MM-SafetyBench上，攻击成功率从39.2%降至12.4%。即使在白盒自适应联合提示-图像攻击下，攻击成功率也被限制在20.3%，而未防御模型为54.6%。

Insight: 论文的创新点在于提出了一个无需修改模型参数、直接在预填充阶段对多模态KV记忆进行针对性修复的两阶段框架。其核心洞察是，通过识别并投影掉与有害意图相关的表征子空间，并利用图像结构先验锚定修复后的键，可以有效地缓解多模态越狱行为，同时保持模型的通用能力。该方法避免了在推理时使用显式的有害分类器，而是通过基于干预相关子空间能量的头级系数来混合记忆。

Abstract: Vision-language models (VLMs) remain vulnerable to jailbreaks that distribute harmful intent across text and images, making unimodal safety mechanisms insufficient. We investigate whether this vulnerability can be mitigated directly in the multimodal key-value (KV) memory formed during prefill, without modifying model parameters at inference time. We introduce SPARK, a two-stage framework for targeted KV-memory repair. Stage 1 uses a disposable diagnostic adapter to identify harm-associated directions in multimodal key and value representations. Stage 2 projects out these directions, learns a lightweight residual repair, and anchors repaired keys with an image-structural prior to preserve visual grounding. Rather than applying the intervention uniformly, SPARK mixes repaired and original memory using a head-wise coefficient g_h* determined by intervention-relevant subspace energy E_h, requiring no explicit harm classifier at inference. Across LLaVA-OneVision-7B, Chameleon-7B, Qwen2-VL-7B, and InternVL2-4B, SPARK reduces multimodal attack success while preserving general capability. On LLaVA-OneVision-7B, image-only jailbreak attack success falls to 4.7%, while MMMU remains within 0.6 points of the undefended model (47.8 vs. 48.4) with near-baseline language quality. On MM-SafetyBench, attack success decreases from 39.2% to 12.4%. Even under white-box adaptive joint prompt-image attacks, attack success is limited to 20.3%, compared with 54.6% for the undefended model. These results suggest that multimodal jailbreak behavior can be substantially mitigated by selectively repairing intervention-relevant KV subspaces at prefill, particularly when harmful evidence is carried by the visual modality.

[50] S3-Tracker: Self-Supervised Surgical Tissue Tracking With Contrastive Random Walks cs.CV | cs.LG | cs.ROPDF

Jiaming Zhang, Zijian Wu, Mehran Armand, Septimiu Salcudean

TL;DR: 本文提出了一种名为S3-Tracker的自监督手术组织跟踪方法，通过对比随机游走从无标注的内窥镜视频中学习全局像素对应关系并推断点轨迹，旨在解决手术环境下软组织变形导致的稳健点跟踪问题。

Details

Motivation: 内窥镜视频中的稳健点跟踪对于计算机辅助干预和自主机器人手术至关重要，但监督方法依赖大量标注数据，而手术条件使得可靠的轨迹标注具有挑战性。

Result: 该方法在无需标注的情况下进行训练，其性能与现有的半监督方法相当，并能隐式处理组织变形。

Insight: 创新点在于提出了一种自监督的Track-Any-Point方法，利用对比随机游走建立像素对应关系，这展示了在手术环境中实现自监督点跟踪的可行性，并有望减少对标注数据的依赖。

Abstract: Robust point tracking in endoscopic videos is essential for computer-assisted intervention and autonomous robotic surgery, enabling continuous registration between intraoperative video and preoperative imaging despite soft tissue deformation. However, supervised tracking methods depend on large annotated datasets, while surgical conditions make reliable trajectory annotation challenging. We propose a self-supervised Track-Any-Point approach that learns from unlabeled surgical videos by establishing global pixel correspondences and inferring point trajectories through contrastive random walks. Trained without annotations, our method achieves performance comparable to existing semi-supervised approaches while implicitly handling tissue deformation. These findings demonstrate the feasibility of self-supervised point tracking in surgical environments and its potential to reduce reliance on annotated data.

[51] Vision-Language Models for Criterion-Level Grading of Handwritten Examinations in Outcome-Based Education cs.CVPDF

Md Khalid Syfullah, Asif Hasan Tonmoy, Saad Ahmed, S. M. Jahangir Alam

TL;DR: 本研究评估了多种视觉语言模型（VLM）在手写考试基于学习成果的准则级评分任务上的表现。通过比较包括Qwen2.5-VL、InternVL3、Pixtral在内的20种模型配置，研究发现，经过LoRA微调的Qwen2.5-VL模型在评分准确性上超越了人类评分员之间的平均一致性水平。同时，研究揭示了模型在评分可靠性、可重复性以及解释质量方面存在的挑战。

Details

Motivation: 在基于成果的教育中，准则级评分将考试表现与学习成果关联，但人工评分存在工作量大和评分者间差异的问题。本研究旨在探索VLM模型自动化此评分过程的潜力，以减轻教师负担并提高评分一致性。

Result: 在1982条准则级记录上，使用LoRA微调的Qwen2.5-VL模型取得了最佳性能：其与主评分员的二次加权Kappa（QWK）为0.727，平均绝对误差为0.435分，优于人类评分员对间的平均QWK（0.551）。然而，模型在多次运行中的评分一致性（50.2-63.6%的准则分数发生变化）和生成的解释质量方面表现不佳。

Insight: 研究强调了针对具体评分标准进行模型校准的重要性，以及确保评分可重复性的必要性。LoRA微调在指令调优的VLM上表现优于部分微调和少样本提示。一个关键发现是，模型在评分准确性上可能达到实用水平，但其解释生成能力需要单独验证，且注意力引导的归因方法在本任务中未显示出统计显著优势。

Abstract: Criterion-level grading connects examination performance to learning outcomes, but manual marking introduces workload and variation between markers. This study evaluates vision-language models (VLMs) for handwritten outcome-based assessment across five dimensions: accuracy, human agreement, repeated-run reliability, error concentration, and explanation quality. Using 1,982 criterion-level records from 485 undergraduate examination answers, we compare 20 configurations spanning Qwen2.5-VL, InternVL3, Pixtral, a Donut baseline, and a cascade ensemble. Evaluation setups include zero-shot prompting, few-shot prompting, partial fine-tuning, and Low-Rank Adaptation (LoRA). Two independent faculty markers regraded all 291 test criteria, providing a human agreement baseline on the same assessment materials. Qwen2.5-VL with LoRA achieved Quadratic Weighted Kappa (QWK) of 0.727 and mean absolute error of 0.435 marks against the examiner, compared with mean human-pair QWK of 0.551. This comparison reflects calibration to the examiner’s training marks. LoRA outperformed partial fine-tuning for all three instruction-tuned VLMs, while few-shot prompting reduced QWK in every configuration with valid prompted scores. Aggregate reliability and exact repeatability diverged: intraclass correlations ranged from 0.790 to 0.874, yet 50.2-63.6% of criteria changed marks across five sampled runs. Attention-guided deletion showed no statistically significant advantage over random masking, and four faculty reviewers reached no consensus on explanation usefulness. These findings highlight the need for rubric-specific calibration, repeatable scoring, review of consequential errors, and separate validation of explanations. The released evaluation protocol supports criterion-level assessment research and grading tools with teacher oversight.

[52] Learning Continuous Source Responses For Generalizable AI-Generated Image Detection cs.CVPDF

Manni Cui, Ruiqi Liu, Zijian Yu, Hao Tan, Zibo Wei

TL;DR: CuRe 将 AI 生成图像检测改为回归真实—生成混合比例，并在紧凑源响应子空间内分类，以减少二元端点监督诱发的捷径特征。

Details

Motivation: 现有生成图检测器在域内表现较好，但跨生成器泛化弱，常被归因于捷径过拟合；即便抑制捷径，仍沿用二元分类任务，未重新审视训练目标本身的表示影响。

Result: CuRe 用真实与生成内容混合比例作为连续监督信号适配主干，再选择紧凑源响应子空间限制分类器使用潜在捷径特征。在十个公开基准上平均平衡准确率为 89.7%，比第二名高 5.2 个百分点；跨视觉主干的泛化提升一致，对常见图像退化也较稳健。

Insight: 连续监督能让模型学习真实性变化的细粒度，而非只分离两个端点，源响应子空间进一步压缩干扰因素；跨生成器结果积极，但需了解混合比例构造与公开基准覆盖范围。

Abstract: Advances in image generation have made synthetic images increasingly difficult to distinguish from real photographs, raising concerns about the trustworthiness of visual media. Existing AI-generated image detectors often perform well on in-domain data, but their robustness and cross-generator generalization remain limited. These limitations are commonly attributed to overfitting to shortcut cues. Although many methods seek to suppress shortcut learning, most retain binary classification as the training task without reconsidering how the task itself shapes the learned representations. We introduce CuRe, a framework for learning Continuous Source Responses that revisits authenticity detection from the perspective of the training task. CuRe reformulates backbone adaptation as regression of real-generated mixing ratios, providing finer supervision that encourages the model to capture authenticity-related variation beyond binary endpoint separation. We further select a compact source-response subspace to suppress nuisance variation and limit the final classifier’s access to potential shortcut cues. Across ten public benchmarks, CuRe achieves an average balanced accuracy of 89.7%, exceeding the second-best method by 5.2 percentage points. Further experiments demonstrate consistent generalization gains across visual backbones and strong robustness to common image degradations. Code is available at https://github.com/manic-cui/CuRe

[53] Two-Stage Mixture-of-LoRA for Multi-Task Medical Vision-Language Learning cs.CVPDF

Zhanghao Chen, Yuanyuan Li, Zhenyu Lu, Shuo Gao, Guangquan Zhou

TL;DR: 两阶段 Mixture-of-LoRA 为 MedGemma-1.5-4B 配置共享 LoRA 和六个任务专家，先联合训练后逐任务细化，以支持多任务医学视觉语言学习。

Details

Motivation: 医学视觉语言模型需同时做分类、检测、回归和报告生成，不同输出格式、冲突梯度和训练数据失衡使联合适配困难。

Result: 框架由一组共享 LoRA 和六个任务专属专家组成。第一阶段所有任务共同训练共享和专家适配器；第二阶段冻结主干、共享适配器及非目标专家，逐个细化任务专家；分类和回归再进行模态均衡续训。在 FLARE 2026 Task 3 测试集上，分类平衡准确率 0.85、多标签分类 micro-F1 0.48、检测 F1 0.79、回归 MAE 17.39。

Insight: 共享与任务特定适配器配合分阶段更新，能隔离任务冲突；分类和回归额外重采样用于缓和模态失衡，结果限于所测挑战赛任务与数据。

Abstract: Medical vision-language models (VLMs) allow a single model to perform clinical image analysis tasks ranging from diagnosis classification to report generation. However, joint adaptation is challenged by heterogeneous output formats, conflicting task gradients, and imbalanced training data. Hence, we present \textbf{Two-Stage Mixture-of-LoRA}, a framework built on MedGemma-1.5-4B. The framework uses a shared-specific Mixture-of-LoRA architecture comprising one shared LoRA and six task-specific expert LoRAs, together with a two-stage training procedure. In Stage 1, we jointly train the shared LoRA and all task-specific expert LoRAs on all tasks. In Stage 2, we first freeze the backbone, the shared LoRA, and all non-target experts, and refine one task expert at a time. Classification and regression then receive an additional modality-balanced continuation, in which smaller modality groups are repeated to match the largest group. In the FLARE 2026 Task 3 test sets, the proposed method achieves 0.85 balanced accuracy for classification, 0.48 micro-F1 for multi-label classification, 0.79 detection F1, and 17.39 regression MAE. Code is available at https://github.com/YuanYL03/MICCAI-FLARE-2026-Challenge-Task3-2D.

[54] Rethinking Camouflage Image Generation towards a Training-Free Paradigm cs.CVPDF

Haodong Yang, Zhongling Huang, Gong Cheng

TL;DR: FreeCam 提出无需训练的伪装图像生成范式，以冻结修复扩散模型保护前景，并利用多模态场景推理与颜色纹理线索合成隐蔽背景。

Details

Motivation: 现有伪装图像生成依赖专门数据集训练，成本高且跨域能力有限；传统目标保留强调视觉显著性，与降低目标可察觉性的伪装目标相冲突。

Result: FreeCam 以冻结的图像修复扩散框架保留前景，通过上下文推理模块利用冻结多模态先验选择适合隐蔽的环境，再由内在外观模块提取目标颜色和纹理，引导背景生成以实现外观融合。实验报告生成质量和伪装效果达到先进水平，无需任务微调；合成图像还可作为伪装目标检测监督，并降低通用检测器对目标的检测率。

Insight: 伪装生成应优化目标与背景的感知可分性，而非单纯让目标突出且完整；冻结先验避免训练成本，但摘要未给出跨数据集数字和前景保真度权衡。

Abstract: Camouflage image generation (CIG) aims to synthesize realistic camouflaged images by blending foreground objects into concealment-compatible background contexts. Achieving this objective requires jointly satisfying three coupled requirements: foreground preservation to retain target integrity, semantic compatibility to select plausible concealment contexts, and appearance assimilation to reduce visual discrepancies. Recent approaches predominantly rely on task-specific training on camouflage datasets to address these requirements, incurring substantial computational cost and limiting generalization beyond the training domain. To address these limitations, we formulate training-free CIG as a concealment-oriented paradigm that preserves the target while reducing its perceptual separability from the synthesized surroundings, rather than maintaining its visual prominence, without parameter updates. We instantiate this paradigm with FreeCam based on a frozen inpainting diffusion framework to preserve the foreground. Within this framework, a Contextual Reasoning Module exploits frozen multimodal priors to infer an environment favorable to concealment, thereby promoting semantic compatibility, while an Intrinsic Appearance Module extracts low-level color and texture cues from the foreground to guide background synthesis toward appearance assimilation. Extensive experiments demonstrate that FreeCam achieves state-of-the-art generation quality and camouflage effectiveness without task-specific training, while its generated images provide synthetic supervision for camouflaged object detection and reduce target detectability under general object detectors.

[55] Contour-Guided Spectral Routing for Robust Real-Time Pedestrian Detection cs.CVPDF

Sam Williams, Yuan Xiang

TL;DR: 论文提出轮廓引导频谱路由行人检测器，依次融合全局低频结构、高频边界和边界—语义差异信息，以改善小目标、遮挡和天气变化下的实时检测。

Details

Motivation: 驾驶场景行人检测面临小目标证据弱、遮挡导致几何支持不足，以及天气和照明变化扰动外观统计三类相互关联的失效；作者希望统一协调空间、频谱和边界信息。

Result: 检测器先向空间证据加入全局频谱上下文，再使深层表征交换空间和频率线索，最后用边界—语义分歧条件化跨尺度融合。训练时保留小波子带变换，分别扰动高低频系数以模拟雾、雨、雪和弱光。在 CityPersons 上 AP50/AP50:95 为 70.4/44.2，高于 RT-DETR 的 68.1/42.2；完整小波增强配置达到 71.1/44.6。

Insight: 低频有助于稳定整体结构，高频保护小目标轮廓，按预定顺序路由比将频域、注意力和边界模块任意并列更有针对性；摘要未列出各天气子集的分项表现。

Abstract: Real-time pedestrian detection in driving scenes is constrained by three coupled failure modes: tiny targets lose discriminative evidence, occlusion weakens geometric support, and weather or illumination changes distort appearance statistics. We formulate the detector through a unified \emph{contour-guided spectral routing} view rather than treating frequency processing, attention, and boundary reasoning as independent add-ons. The detector routes information in a prescribed order: spatial evidence is first augmented with global spectral context, deep representations then exchange spatial and spectral cues, and cross-scale fusion is finally conditioned on boundary–semantic disagreement. This ordering yields a compact representation pipeline in which low-frequency context stabilizes global structure while high-frequency evidence protects small-object contours. We further retain a wavelet-subband training transformation that perturbs low- and high-frequency coefficients independently, targeting appearance shifts caused by fog, rain, snow, and low illumination. The formulation exposes a single routing variable at each stage and distinguishes reusable signal transforms from the task-specific policy that decides where each signal is injected. On CityPersons, the proposed detector obtains 70.4 AP${50}$ and 44.2 AP${50:95}$, compared with 68.1 and 42.2 for RT-DETR, while the full wavelet-augmented configuration reaches 71.1 and 44.6.

[56] A Generative AI Integrated Multimodal Framework for Low-Latency Multi-Camera Person Re-Identification cs.CV | cs.AI | eess.IVPDF

Leon Fernando, C Dombawala, P. Hettigoda, Vanodhya G. Warnasooriya, Ishara Neranjana

TL;DR: 多摄像头行人重识别系统用成本感知早退级联先检查廉价视觉检索结果，仅在模糊时调用 VLM 语义描述或人脸特征。

Details

Motivation: 行人重识别受视角、光照、遮挡、背景和低分辨率影响，多模态信息可增强鲁棒性但全量调用会增加时延；系统需按查询歧义自适应使用昂贵线索。

Result: 框架融合分割人体的全局视觉嵌入、VLM 自动生成的细粒度语义属性及可选的人脸嵌入。先查看 top-k 检索结果的最佳匹配与其余候选间距，明显无歧义时直接返回，模糊时保留多个假设并按可靠性权重调用额外模态。在 DukeMTMC-reID 的 60.7% 查询及 Market-1501 的 68.4% 查询中，无需语义推理即可解决，同时保持有竞争力的 mAP 和 Rank-1 表现。

Insight: 把昂贵多模态计算保留给歧义样例可降低总体推理负担，而不是对每个查询统一融合；摘要未给出完整准确率和平均时延，因此节省幅度仍需完整实验核查。

Abstract: Person re-identification (ReID) is essential for multi-camera surveillance and tracking, yet remains difficult due to viewpoint and illumination changes, occlusion, background clutter, and low resolution imagery. We propose a generative AI integrated multimodal ReID framework designed explicitly for robustness under missing cues and low latency deployment. The key idea is a cost aware early-exit cascade that prioritizes inexpensive, high confidence evidence and only triggers expensive modalities for ambiguous cases. Our system integrates (i) global visual embeddings from segmented person regions, (ii) automatically generated fine grained semantic attribute descriptions generated by vision-language models (VLMs), and (iii) optional facial embeddings when face observations are reliable. To optimize the balance between accuracy and latency, we use a cost aware early-exit cascade instead of fusing all modalities. Specifically, we first inspect the top-k retrieval results to determine whether the query is unambiguous. If the best match is clearly separated from the remaining candidates, we stop early and return the result to minimize latency; in ambiguous cases, we keep multiple hypotheses and invoke additional modalities (face/semantic) with adaptive reliability weighting to refine the decision. We report person re-identification performance using mAP and Rank-1 accuracy on the Market-1501 and DukeMTMC-reID benchmarks. The proposed adaptive early-exit cascade resolves 60.7% of DukeMTMC-reID queries and 68.4% of Market-1501 queries without invoking semantic reasoning, reducing computational overhead while maintaining competitive retrieval performance.

[57] Lightweight Generalized DeepFake Face Detection with WAVIE: Wavelet Augmented Vision Intermediate Embeddings cs.CV | cs.AIPDF

Arya Pulkit, Aditya Ruhela, Akarshan Kapoor, Arnav Bhavsar

TL;DR: WAVIE 在冻结 CLIP 中间视觉嵌入上加入三层小波增强，以联合空间与频率特征改善跨数据集 DeepFake 检测。

Details

Motivation: DeepFake 检测器往往在训练中见过的伪造方法上有效，却在未见生成流程上显著退化；依赖训练集捷径是影响真实部署泛化的关键风险。

Result: WAVIE 对冻结 CLIP 中间 Transformer 嵌入进行轻量投影，应用三级 Daubechies-6 离散小波变换，增强低频分支、保留高频分支，再逆变换分类。只在 FaceForensics++ 训练时，帧级 AUROC 在 Celeb-DF-v1、Celeb-DF-v2 和 WildDeepFake 分别为 0.852、0.852 和 0.831，优于若干跨域基线；消融支持小波模块和中间特征聚合的作用。

Insight: 冻结视觉主干加频域特征增强能减少对单一伪造数据集的依赖，跨数据集结果支持空间与频谱线索互补；摘要提供帧级 AUROC，视频级判定能力仍需单独验证。

Abstract: Deepfake detection systems often exhibit significant performance degradation when deployed on unseen manipulation methods, limiting their reliability in real-world multimedia environments. This lack of generalization poses critical challenges for misinformation mitigation, digital forensics, and human-centric AI systems. Existing detectors perform well on the forgery methods they are trained on, but their accuracy drops sharply on unseen pipelines. To bridge this generalization gap, we propose WAVIE (Wavelet Augmented Vision Intermediate Embeddings), an end-to-end architecture that combines complementary spatial and frequency cues on top of a frozen CLIP backbone. WAVIE projects intermediate transformer embeddings through a lightweight learnable module, applies a three-level Daubechies-6 (db6) discrete wavelet transform (DWT), refines the low-frequency branch while preserving the high-frequency branch, reconstructs the feature via inverse DWT, and performs classification. Trained only on FaceForensics++, WAVIE achieves AUROC = 0.852 on Celeb-DF-v1, 0.852 on Celeb-DF-v2 and 0.831 on WildDeepFake (WDF) at the frame level, outperforming several state-of-the-art generalization baselines. Extensive ablation studies confirm the importance of both the wavelet module and the intermediate-feature aggregation for cross-dataset performance, highlighting the necessity of jointly leveraging spatial and frequency domains. These results position WAVIE as a strong baseline for deepfake detection in the wild.

[58] Context-Aware Mutual Learning for Blind Image Inpainting and Beyond cs.CVPDF

Haoru Zhao, Yufeng Wang, Zhaorui Gu, Bing Zheng, Haiyong Zheng

TL;DR: CAML 将盲图像修复中的掩码估计与图像填补设为双向互学，使两者分别提供语义上下文和边缘纹理细节。

Details

Motivation: 未知掩码下的图像修复常分成先估计污染区域、再执行修复的两阶段流程，但既有方法主要单向利用掩码给修复提供条件，忽视图像纹理也能反过来帮助识别掩码。

Result: CAML 包含填补引导的上下文互学模块（IGCM），将修复图像中的纹理、边缘等细节用于改善掩码估计；另有估计引导模块（EGCM），利用掩码语义强化图像填补。消融实验验证组件作用；在盲图像修复以及雪、阴影、水印去除任务上，摘要报告达到先进表现，但未提供具体数值。

Insight: 掩码定位与内容恢复存在互补信息，应联合学习而非只把定位结果单向传给修复网络；多任务结果显示框架具迁移潜力，但摘要未给出各任务的绝对指标。

Abstract: Blind image inpainting, aiming to recover contaminated images in the case of unknown masks, is a challenging task. Motivated by the perspective of human vision and knowledge, blind image inpainting can be decomposed into two stages: mask estimation and image inpainting based on the estimated mask. The two-stage idea exhibits evident advantages in enhancing inpainting quality and augmenting the generalization capability of unknown real-world contamination by explicitly employing the estimated mask for image inpainting compared to one-stage scheme. This two-stage idea has also been intuitively implemented. However, existing two-stage methods excessively emphasize the unilateral relationship of mask estimation to image inpainting, and may overlook the mutual relations between them. Specifically, mask estimation can provide more contextual semantics for image inpainting to strengthen the understanding of semantics, and image inpainting can offer more contextual details (e.g., textures and edges) for mask estimation to improve the learning of details. In this work, we propose a novel Context-Aware Mutual Learning (CAML) framework for blind image inpainting that joints mask estimation and image inpainting to mutually exploit contextual information. In the CAML framework, we design the Inpainting-Guided Context-Mutual (IGCM) learner to acquire the complementary contextual details from image inpainting for assisting mask estimation, and the Estimation-Guided Context-Mutual (EGCM) learner to strengthen the understanding of contextual semantics from mask estimation for assisting image inpainting. Ablation studies validate the efficacy of our CAML. Extensive experiments show that our CAML achieves state-of-the-art performance on both blind image inpainting and additional vision tasks, i.e., snow removal, shadow removal, and watermark removal, indicating its superiority.

[59] From Visual Attribution to Clinical Reasoning: Explainable Parkinson’s Disease Screening from Hand-Drawn Patterns cs.CV | cs.AIPDF

Aritra Dey, Utsav Kumar Nareti, Chandranath Adak, Soumi Chattopadhyay, Krishna Gopal Sasmal

TL;DR: 论文提出基于手绘螺旋和曲线图样的可解释帕金森筛查框架，从视觉显著区域提取运动学描述符并生成临床关联解释。

Details

Motivation: 帕金森病早期运动症状会体现在受控手绘图样的震颤、笔画不规则和曲率变化中；传统筛查可能主观、耗时且依赖专家，黑盒分类也难以说明依据。

Result: 框架在受试者互斥协议下建模离线手绘图样的分布式结构失真和细节纹理，再用注意力及梯度定位显著区域，提取轮廓粗糙度、曲率不规则、笔画变化和震颤频率能量等临床描述符，最后由语言推理模块形成症状学解释。在公开帕金森手写数据集上取得有竞争力的预测效果和临床一致解释；摘要未报告具体分数。

Insight: 视觉归因可通过明确的运动学描述符连接到临床症状，帮助从筛查分数转向可审查理由；现有摘要只称结果有竞争力，无法判断准确率或临床验证规模。

Abstract: Parkinson’s disease (PD) manifests early neuromotor impairments that become observable in controlled hand-drawn patterns such as spirals and meanders, where tremor-induced oscillations, stroke irregularity, and curvature instability reflect underlying motor degradation. In this work, we present an explainable framework for PD screening from offline hand-drawn patterns that integrates discriminative visual modeling with clinically grounded reasoning. The predictive model captures distributed structural distortions and fine-grained texture variations. It is evaluated under subject-disjoint protocols to ensure reliable generalization. To move beyond black-box classification, we introduce a multi-stage explainability pipeline that combines visual attribution with structured symptom abstraction. Salient regions are identified using attention- and gradient-based localization, followed by extraction of clinically meaningful motor descriptors quantifying contour roughness, curvature irregularity, stroke variability, and tremor-frequency energy. These descriptors are subsequently translated into coherent clinical rationales through a language-based reasoning module, linking model evidence to established PD symptomatology. By bridging visual attribution and clinical interpretation, the proposed framework advances interpretable document intelligence for neurological screening using hand-drawn patterns. Experimental results on publicly available Parkinson’s disease handwriting datasets demonstrate competitive predictive performance and clinically consistent explanations.

[60] AlayaVista: Streaming World Modeling from Panoramic States to Perspective Video cs.CVPDF

Jiaming Tan, Mingliang Zhai, Zhen Li, Yuwei Wu, Chuanhao Li

TL;DR: AlayaVista 将全景世界演化与相机视角视频合成解耦，计划从单张透视图构建 360° 场景先验并以流式方式生成受控视角视频。

Details

Motivation: 交互视频世界模型需在摄像机移动时保留广阔场景上下文并低延迟生成高保真局部观察；纯透视模型容易丢失屏外内容，完整球面视频或显式 3D 表示成本又高。

Result: AlayaVista 先用预训练全景扩展模型从单张透视图构造 360° 场景先验，再演化相机条件全景潜状态；潜视口渲染器将其转换为目标透视视频潜变量，透视细化器恢复细节、抑制伪影并超分辨。为流式效率，作者将全景生成改为分块自回归，并将全景生成和视角细化蒸馏为少步流程。另构建 MUGEN 数据集，含 1,318 小时、至少 4K 分辨率的真实全景视频及语义和几何标注；摘要未给模型实验结果。

Insight: 全局全景状态可维护屏外上下文，透视渲染再提供局部细节，形成全球景记忆与局部视口的分层表示；本地摘要只报告方法和数据集，尚无性能指标可概括。

Abstract: Interactive video world models must maintain broad scene context under camera motion while producing high-fidelity observations with low latency. Existing approaches face a representation trade-off: perspective models operate on local views and must preserve off-screen content over long rollouts, whereas broader spatial coverage is typically obtained by synthesizing full-sphere videos or constructing explicit 3D representations. Motivated by the complementary roles of global context and selective local acuity in visual perception, we present AlayaVista, a camera-controllable streaming video world model that decouples panoramic world evolution from perspective observation synthesis. Given a single perspective image, AlayaVista constructs a 360-degree scene prior using a pretrained panorama expansion model and then evolves the scene as a camera-conditioned panoramic latent state. A latent viewport renderer maps this state to the requested perspective video latents, while a perspective refiner restores details, suppresses artifacts, and performs super-resolution. To support efficient streaming, we adapt the panoramic generator to chunk-autoregressive generation and distill both panoramic generation and perspective refinement into few-step processes. To provide the supervision required by this design, we construct MUGEN, a large-scale real-world panoramic video dataset containing 1,318 hours of videos at resolutions of at least 4K, together with rich semantic and geometric annotations.

[61] DynEoMT: Learning Object Dynamicity from Online Segmentation Queries cs.CV | cs.ROPDF

Calvin Galagain, Martyna Poreba, François Goulette

TL;DR: DynEoMT 在查询式视频分割中增加区域级动态/静态预测，只使用当前帧与传播查询在线推理，无需光流、深度、相机位姿或历史图像。

Details

Motivation: 视频分割能识别并跟踪物体，却不判断区域是否相对观察相机独立运动；语义类别本身不能确定动态性，且会与相机自运动混淆。

Result: DynEoMT 同时输出原有分割与每个预测区域的动态状态。由于常见基准没有该标注，作者以相机补偿光流和置信度时序滤波建立离线类别无关监督。VIPSeg、OVIS、YouTube-VIS 2022 和 VSPW 上动态性平衡准确率分别为 84.3、68.0、68.6 和 87.6，同时大体保持原分割性能。

Insight: 传播查询已包含预测区域运动线索，可支持无专用运动模块的在线动态性判断；训练监督依赖光流生成且各数据集准确率差异较大，需注意监督噪声与数据域迁移。

Abstract: Video segmentation models recognize and track objects over time, but they do not indicate whether each segmented region moves independently of the observing camera. This dynamicity attribute cannot be inferred from semantics alone and is confounded by camera ego-motion. We introduce \method, an online framework that augments query-based video segmentation with region-level dynamicity prediction. It jointly produces the original segmentation outputs and a dynamic or static state for each predicted region. At inference, DynEoMT uses only the current frame and propagated queries, without optical flow, depth, camera pose, previous RGB frames, or feature maps. Because established video segmentation benchmarks do not annotate this attribute, we also introduce a class-agnostic offline supervision pipeline using camera-compensated optical flow and confidence-aware temporal filtering. Across VIPSeg, OVIS, YouTube-VIS 2022, and VSPW, DynEoMT achieves balanced accuracies of 84.3, 68.0, 68.6, and 87.6, respectively, while largely preserving segmentation performance. These results show that segmentation-region dynamicity can be learned from propagated queries, enabling its online prediction without a dedicated motion-processing pipeline at inference. The complete code will be released as open source to enable full reproduction of the method and experiments.

[62] PRI-Net: A Lightweight Multimodal Framework for 3D UAV Localization cs.CV | eess.SPPDF

Zhixuan Chen, Jialiang Lu, Zhong Ye, Yinghui He, Guanding Yu

TL;DR: PRI-Net 融合稀疏激光雷达点云与图像，用点云投影生成稠密深度、残差注意力缓解模态偏差，并以信息瓶颈压缩无关特征。

Details

Motivation: 无人机三维定位受稀疏 LiDAR 几何、模态不平衡融合和边缘—服务器链路传输冗余特征影响，传统多模态方法难以同时做到准确、轻量和鲁棒。

Result: 框架先将稀疏 LiDAR 点云投影为几何一致的稠密深度图，再用图像支路粗估计、门控融合支路细化的残差注意力融合模块降低模态偏置，最后通过多模态信息瓶颈过滤任务无关冗余。实验显示轻量架构仍能实现高定位精度，并减少特征维数、改善边云感知效率和鲁棒性；摘要未报告具体误差和压缩比例。

Insight: 通过几何补全、分阶段融合和信息压缩分别解决稀疏观测、模态失衡与传输成本；性能和效率收益需要结合完整实验数值及边云部署协议评估。

Abstract: Accurate 3D localization of unmanned aerial vehicles (UAVs) remains challenging for existing multimodal approaches due to sparse LiDAR geometry, modality-imbalanced fusion, and redundant feature transmission over constrained edge-to-server links. To address these limitations, we propose PRI-Net, an efficient and lightweight multimodal fusion framework for UAV localization that integrates point cloud splatting, residual attention fusion, and an information bottleneck. Specifically, a 3D point cloud splatting (3DPCS) strategy is introduced to transform sparse LiDAR observations into geometrically consistent dense depth maps. A residual attention fusion (RAF) module is then designed to alleviate modal bias by using an image branch for coarse estimation and a gated fusion branch for refinement. In addition, a multimodal information bottleneck (MIB) module compacts features by filtering task-irrelevant redundancy. Experiments show that PRI-Net achieves high localization accuracy with lightweight architectures, while reducing feature dimensionality and improving edge-to-server UAV sensing efficiency and robustness.

[63] PuzzleMate: Benchmarking MLLMs for Egocentric Puzzle Assistance cs.CVPDF

Avijit Dasgupta, Shayon Dasgupta, Zakaria Laskar, C. V. Jawahar, Karteek Alahari

TL;DR: PuzzleMate 用第一人称拼图协助任务评估多模态大模型能否感知当前拼图进度并给出可执行的逐步操作建议，发现现有模型仍难处理精细空间推理。

Details

Motivation: 个人 AI 助手若要成为具身伴侣，不能只识别物体，还需根据用户实时进度给出连续、精确的操作指导；通用视觉理解并不能证明其能处理细粒度操作任务。

Result: 作者将拼图作为高精度空间关系和严格顺序逻辑的测试任务，构建 PuzzleMate 并开展用户在环研究，评估模型从第一人称视角判断拼图状态、生成下一步可执行指令的能力，归纳出七类主要瓶颈。基于洞见建立系统化评测基准。GPT-5.2 和 Gemini-2.5-Pro 等模型虽有较强能力，但在复杂推理与顺序操作上仍有明显差距；摘要未列出具体成功率。

Insight: 拼图提供了从静态物体识别走向连续具身操作推理的细粒度评测场景；用户在环发现的瓶颈说明高层语言能力不保证实际步骤规划成功。

Abstract: Personal AI assistants hold the potential to evolve from digital interfaces into embodied companions capable of guiding users through complex physical activities. For these assistants to become integral to daily life, they must do more than identify objects; they must provide precise, step-by-step instructions that align with a user’s real-time progress. While Multimodal Large Language Models (MLLMs) show promise in general visual understanding, their ability to deliver grounded, sequential guidance for fine-grained manipulation tasks remains largely unverified. In this paper, we choose the jigsaw puzzle as a strategic testbed for this capability. Unlike general object recognition, puzzle solving demands high-precision spatial reasoning, the ability to distinguish between minute geometric variations, and a rigorous adherence to sequential logic. We investigate this capability through PuzzleMate, a novel framework focused on jigsaw puzzle solving captured through an egocentric viewpoint. We deploy PuzzleMate in a user-in-the-loop study to evaluate how well state-of-the-art MLLMs perceive the current puzzle state and generate actionable next-step instructions. Our analysis reveals seven key bottlenecks that limit their effectiveness. Building on these insights, we propose a benchmark that enables systematic evaluation of MLLMs’ reasoning capabilities for puzzle solving. Our findings reveal a substantial performance gap in current models like GPT-5.2 and Gemini-2.5-Pro; while these MLLMs are highly capable, they struggle to navigate the intricate reasoning and sequential logic essential for jigsaw puzzle assistance.

[64] Open-UniMo: Towards Unified Motion-Language Understanding and Generation in the Open World cs.CV | cs.AIPDF

Guocun Wang, Kenkun Liu, Guorui Song, Jing Lin, Zhe Huang

TL;DR: Open-UniMo 用共享文本—动作 token 空间、运动一致 CoT 和两阶段训练统一开放世界动作理解与生成，并以 Open-MoBench 进行双向评估。

Details

Motivation: 现有动作—语言模型常将动作当作语言模型的附属模态，导致文本主导和跨模态交互弱；自回归预测长动作序列还会累积错误。

Result: Open-UniMo 在约 150K 文本 token 的 Qwen 词表上加入 64K 动作 token，使用运动一致思维链桥接语言语义与动作动态；先监督微调建立双向动作—语言映射，再用 GRPO 改善语义对齐并缓解累积误差。新基准 Open-MoBench 评估文本到动作、动作到文本和双向一致性。实验称在常用指标和 Open-MoBench 上达到先进表现；消融显示动作理解并非主要受动作词表大小限制，与可学习动作生成路径耦合更能提升跨模态表示。

Insight: 让理解与生成共享可学习运动路径，可能比单纯扩大动作词表更利于跨模态理解；摘要未列具体分数或开放世界数据规模，结果仍依赖基准设计。

Abstract: Unified motion generation and understanding is crucial for embodied AI systems that can both synthesize and interpret human actions in open-world environments. Existing motion-language models often treat motion as an auxiliary modality of a language model, leading to text-dominated representations and limited cross-modal interaction. Moreover, the next-token prediction paradigm is not naturally suited to long motion sequences, where autoregressive generation may accumulate prediction errors. To address these challenges, we propose Open-UniMo, a unified Large Motion-Language Model (LMLM) trained on million-scale open-world motion-language data. Open-UniMo promotes modality parity by extending Qwen’s vocabulary of about 150K text tokens with 64K motion tokens, enabling motion and language to share a unified token space. We further introduce motion-consistent Chain-of-Thought reasoning as an intermediate representation to bridge language semantics and motion dynamics. Open-UniMo is trained with a two-stage pipeline, where supervised fine-tuning establishes CoT-guided bidirectional motion-language mapping and Group Relative Policy Optimization (GRPO) improves semantic alignment while mitigating cumulative errors in autoregressive motion-token generation. To support comprehensive evaluation, we propose Open-MoBench, a VLM-guided benchmark for assessing text-to-motion (T2M) generation, motion-to-text (M2T) understanding, and bidirectional consistency. Extensive experiments show that Open-UniMo achieves state-of-the-art performance on both conventional metrics and Open-MoBench. Furthermore, ablation studies reveal that M2T understanding is not primarily limited by motion-token vocabulary size; instead, coupling M2T with the learnable T2M generation path yields stronger cross-modal representations, demonstrating that generation can facilitate understanding in AR-based motion-language modeling.

[65] Selective Tool Use for Agentic Change Visual Question Answering in Remote Sensing cs.CVPDF

Yakoub Bazi, Mohamad M. Al Rahhal, Mohamed A. Mekhtiche, Mansour Zuair

TL;DR: 该框架让遥感变化问答模型按问题类型选择直接作答或调用确定性语义变化分析工具，在 7,164 道题上提升准确率，但结果受语义图质量影响。

Details

Motivation: 双时相遥感变化问答中，涉及转移统计、面积测量和空间关系的问题需要明确证据，通用视觉语言模型单靠图像语义理解仍不可靠。

Result: 系统按需调用处理双时相语义图的转移、空间和时间工具，并在原 CDVQA 上扩展八类问题；工具监督和结构化观测由既有语义标注自动生成。使用参考语义图时，Qwen3.5-4B LoRA 模型整体准确率从 73.77% 升至 88.79%，平均问题族准确率从 69.11% 升至 89.65%。用自动预测语义图时，整体/平均族准确率分别为 77.47%/75.06%。

Insight: 问题特定的确定性工具可为变化推理提供结构化证据，收益明显；参考语义图与自动预测图之间的差距说明上游语义分割质量限制端到端效果。

Abstract: Change visual question answering (Change VQA) requires understanding semantic changes across bi-temporal remote sensing images. Although vision language models (VLMs) have shown promising performance on this task, they remain unreliable when answering questions that require explicit transition statistics, area measurements, or spatial information. To address this limitation, we propose a selective tool use framework in which a single VLM either answers directly or invokes a deterministic change analysis tool to obtain question specific evidence. Specifically, the selected tool operates on bi-temporal semantic maps and returns a structured observation, which the same VLM uses to generate its final answer. To support this framework, we construct a tool augmented extension of CDVQA covering eight question families and three tools for transition, spatial, and temporal analysis. Tool use supervision and observations are derived automatically from the original semantic annotations, without additional manual labeling. We then adapt Qwen3.5-4B using Low Rank Adaptation (LoRA) to jointly learn direct answering, tool invocation, and evidence conditioned answering. Experiments on 7,164 test questions show that selective tool use with reference semantic maps improves overall accuracy from 73.77% to 88.79% and average family accuracy from 69.11% to 89.65%. When the semantic maps are predicted automatically, the framework achieves 77.47% overall accuracy and 75.06% average family accuracy. These results demonstrate the benefit of question-specific semantic evidence for Change VQA, while highlighting the influence of semantic prediction quality on the resulting performance. Code and tool-augmented annotations will be made publicly available at https://github.com/yakoubbazi/ToolChangeVQA.

[66] Compositional SVG Generation via VLM-Driven Hierarchical Semantic Parsing cs.CV | cs.AIPDF

Sehwan Park, Taehoon Kim, Geonhee Han, Dohyun Kim, Seung Wook Kim

TL;DR: 论文提出由视觉语言模型驱动的分层语义解析代理，将 SVG 图像分解成几何完整、可编辑的语义部件，并建立衡量分组和编辑性的基准。

Details

Motivation: 视觉语言模型擅长视觉推理，但生成的 SVG 常是扁平、无语义路径集合，编辑单个对象需要人工查找其组成路径，缺少结构化可编辑性。

Result: 流程通过自顶向下分解、视觉定位和提示驱动的非显露遮挡恢复递归解析场景语义与几何层级，确保各部件几何完整。Semantic SVG Benchmark 含人工语义分组标注，并新增语义召回/精确率及 PERE 等子组件指标评价结构组合和实际可编辑性。实验中预测结构在分组质量和可编辑性上超过既有扁平生成方法的上限，同时保持先进视觉保真度。

Insight: 图形生成需要同时恢复物体层级和可执行编辑结构，像素相似度不足以评价 SVG 的实用性；摘要未提供具体基准规模和量化指标。

Abstract: While Vision-Language Models (VLMs) excel at visual reasoning, generating structured, editable Scalable Vector Graphics (SVG) remains a fundamental challenge. Existing pipelines predominantly yield flat, semantically agnostic collections of paths, where editing a single object requires manually identifying its constituent paths. To address this, we propose a VLM-driven agentic framework for semantic compositional SVG generation. Our pipeline recursively parses visual scenes into semantic and geometric hierarchies via top-down decomposition, visual grounding, and prompt-driven amodal occlusion recovery, ensuring each component is geometrically complete. Furthermore, we introduce the Semantic SVG Benchmark with human-annotated semantic groups and novel sub-component metrics (Semantic Recall/Precision, PERE) to explicitly evaluate structural compositionality and functional editability. Experiments show that our natively predicted structures surpass the upper bounds of existing flat-generation methods in both grouping quality and editability, while maintaining state-of-the-art visual fidelity.

[67] From Density to Biopsy Decisions and Malignancy Prediction: A Benchmark Study of Multimodal Large Language Models Against Radiologists in Digital and Contrast-Enhanced Mammography cs.CV | physics.med-phPDF

Ali Abbasian Ardakani, Afshin Mohammadi, Taha Yusuf Kuzan, Beyza Nur Kuzan, Alisa Mohebbi

TL;DR: 在 179 名女性的配对数字乳腺摄影和增强摄影研究中，放射科医师在分类任务普遍优于四种多模态 LLM，而提供病灶掩码后部分模型的连续恶性概率预测接近人类范围。

Details

Motivation: 乳腺密度、BI-RADS 分类、活检决策和连续恶性概率评估涉及不同复杂度，需在同一临床数据和不同专业水平读者下比较 MLLM 与放射科医师。

Result: 四个 MLLM 和三名医师评估有参考标准的 179 名女性配对 DM/CEM。密度二分类中医师准确率 55.81%–78.60%，高于多数 MLLM；五类 BI-RADS 和活检候选分类中医师也明显领先。CEM 提升所有读者表现。病灶掩码使 MLLM 连续恶性概率准确率在 DM 从 64.65%–71.63%升至 72.56%–78.60%，在 CEM 从 67.91%–77.21%升至 72.56%–81.86%；最佳掩码模型 AUC 与医师范围重叠。Muse Spark 和 Sonnet-4.6 是表现较强的 MLLM。

Insight: 当前模型在类别式临床判断上仍落后医师，但在提供病灶定位后，连续概率估计更接近人类；可能作为辅助工具，不能据此替代放射科医生。

Abstract: Purpose: To compare four multimodal large language models (MLLMs) with radiologists of varying expertise in breast density assessment, BI-RADS assessment, biopsy candidacy determination, and continuous malignancy probability estimation using digital mammography (DM) and contrast-enhanced mammography (CEM). Methods: This study included 179 women with paired DM/CEM examinations and reference standards. Four MLLMs (ChatGPT-5.2, Gemini-3.1 Pro, Sonnet-4.6, Muse Spark) interpreted images with and without masks; three radiologists interpreted non-masked images. Results: For binary density classification on DM, radiologist accuracies ranged from 55.81% to 78.60%, exceeding most MLLM values (62.33%-71.63%), while masks added limited benefit. Five-category BI-RADS accuracies were higher for radiologists on DM (56.74-67.44%) and CEM (62.33-82.79%) compared with MLLMs (DM 31.16-45.12%; CEM 40.00-55.81%). Binary biopsy-candidacy accuracies were likewise higher for radiologists (DM 85.12-89.77%; CEM 86.98-92.09%) than for MLLMs (DM 61.39-75.35%; CEM 69.30-82.79%), although CEM improved performance across all readers. Lesion masks substantially improved MLLM continuous malignancy-probability accuracies from 64.65%-71.63% to 72.56-78.60% on DM and from 67.91%-77.21% to 72.56%-81.86% on CEM, approaching radiologist ranges (DM 63.72-82.79%; CEM 81.86-88.84%). The corresponding AUCs for the top masked models overlapped those of the human readers. Overall, Muse Spark, followed by Sonnet-4.6, demonstrated the strongest performance among the MLLMs across domains. Conclusion: Radiologists generally outperformed MLLMs in categorical tasks, while selected masked models approached human performance for continuous malignancy probability estimation, suggesting a potential adjunctive role.

Weihong Qi, Chen Ling

TL;DR: PRR 将自然语言视觉类别提示、SAM 像素级分割和多模态 LLM 仲裁结合，构造可审计、可校准的社交科学视觉测量流程。

Details

Motivation: 社交媒体图像研究常用预设类别或粗粒度标签，难以规模化测量具体对象是否出现、显著程度和画面位置，自动检测结果也需可校正和审计。

Result: PRR 结合自然语言类别提示、SAM 空间细化和多模态 LLM 仲裁，并用三级可审计框架估计逐类别可靠性、选择任务配置及校正流行率。在四种视觉语言检测器、九类社会学类别上，精确率显著高于零样本基线，最佳主干提高 43.3 点。应用到 2024 选举期美国国会议员 103,920 张 Facebook 图像并关联 DW-NOMINATE 意识形态后，发现更保守议员展示的美国国旗视觉面积更大，位置更偏外围的趋势较弱；空间关系是二元检测看不到的模式。

Insight: 用空间显著性和位置测量扩展对象检测，能揭示社交科学相关的视觉策略；分类可靠性审计与流行率校正很关键，观察到的意识形态相关性不代表因果关系。

Abstract: Visual content shapes audience perception and opinion on social media, and computational social science increasingly relies on automated tools to analyze images at scale. Yet a measurement gap persists: existing tools rely on predefined categories or produce only coarse image-level labels, while measuring which specific objects appear in an image, how prominently, and where in the frame remains difficult at scale. We introduce Perceive, Refine, Reason (PRR), a calibrated pipeline that turns flexible vision-language detectors into auditable measurement instruments for social-scientific research. PRR combines natural-language category prompts with pixel-level spatial refinement via the Segment Anything Model (SAM) and a multimodal LLM arbitration layer whose reasoning chains externalize domain knowledge and lower the expertise threshold for human-in-the-loop validation. A complementary three-tier auditability framework applies quantification learning to profile per-category reliability, support task-aligned configuration, and statistically correct prevalence estimates. Across four vision-language detectors and nine sociological categories, the pipeline yields substantial precision gains over zero-shot baselines, including a 43.3-point improvement for the strongest backbone. Applying PRR to 103,920 Facebook images from U.S. legislators during the 2024 election cycle and linking detections to DW-NOMINATE ideology scores, we find that more conservative legislators display U.S. flags as larger visual elements, with a weaker tendency toward peripheral placement, a spatial pattern invisible to binary detection. PRR provides computational social scientists with a model-agnostic toolkit for accessible, spatially-grounded, and correctable visual measurement.

[69] CrossDistill: Balancing Quality and Diversity via Trajectory-Level Hybrid Few-Step Distillation cs.CVPDF

Yuxi Liu, Haoyu Li, Yixiang Cai, Tengxu Sun, Zekun Zhang

TL;DR: CrossDistill 沿扩散噪声轨迹切分训练目标：高噪声阶段保持模式覆盖，低噪声阶段做分布匹配，从而改善少步生成的质量—多样性权衡。

Details

Motivation: 少步蒸馏需兼顾样本多样性和保真度；轨迹蒸馏保留模式但细节较弱，分布匹配能锐化样本却可能降低多样性。

Result: CrossDistill 在一个交叉点处分割采样轨迹，高噪声区间使用保留轨迹目标、低噪声区间使用分布匹配目标，并通过交叉状态耦合两阶段；关键是沿噪声轴分配互补目标，而非简单混合损失。PCM 与 DMD 可作为插件实例。在文生视频扩散和图生视频定性实验中，方法拓宽少步质量—多样性前沿，保留种子间变化并达到有竞争力的视觉保真度。

Insight: 全局模式主要在高噪声阶段形成，局部细节在低噪声阶段修正，因此按阶段分配训练目标比统一损失更符合生成过程；图生视频结果摘要仅为定性，需完整指标判断。

Abstract: Few-step distillation accelerates diffusion models but must balance diversity and fidelity: trajectory-based distillation preserves mode coverage, while distribution matching sharpens samples but can reduce diversity. We show that this tension can be exploited in a noise-regime-dependent way: high-noise steps largely determine global modes, whereas low-noise steps refine local details. We propose CrossDistill, a trajectory-level hybrid distillation framework that splits the sampling trajectory at a crossover point, applies a trajectory-preserving objective on the high-noise interval and a distribution-matching objective on the low-noise interval, and couples the two stages through the crossover state. In contrast to loss-level mixing, and complementarily to training-time two-stage recipes, CrossDistill explicitly assigns complementary objectives along the noise axis, so that global branching is preserved before local statistics are sharpened. CrossDistill is a noise-level scheduling policy: PCM and DMD are plug-in instantiations, while the noise partition, crossover coupling, and objective ordering are the key design elements. Experiments on text-to-video diffusion models and qualitative image-to-video results show that CrossDistill expands the few-step quality-diversity frontier, retaining seed-level variation while achieving competitive visual fidelity.

[70] From Visual Feedback to Textual Reviews: A Multi-Agent Vision-Language Framework for Image-Grounded Review Assistance cs.CV | cs.AIPDF

Utsav Kumar Nareti, Ayush Bansal, Kumari Priya, Chandranath Adak, Soumi Chattopadhyay

TL;DR: 论文定义基于商品图像生成可编辑评论草稿的任务，并以四类专门代理通过商品定位、视觉情感估计和证据汇总完成图像支撑评论。

Details

Motivation: 电商用户上传的图像和视频能展示商品质量、缺陷、包装和真实使用，但图像缺少文字解释和主观评价，撰写详细评论又费力。

Result: 多智能体视觉语言框架包含商品定位、视觉情感估计、视觉证据生成和评论综合四个角色，通过商品实体、预测评分和证据摘要等显式中间表示提升可解释性与视觉落地。在 Amazon Reviews Electronics 精选子集上，实验显示可从用户商品图生成连贯、理解商品且情感适配的评论草稿；摘要未报告指标或人工偏好数值。

Insight: 把评论写作分解为商品识别、态度估计和证据组织，有助于避免普通图像描述缺乏主观评价的问题；生成内容仍是草稿，真实性和是否符合用户意图需进一步评估。

Abstract: Visual feedback in the form of user-uploaded images and videos is becoming increasingly common in e-commerce platforms because it provides authentic evidence of product quality, defects, packaging conditions, and real-world usage. However, visual feedback alone often lacks the contextual explanations and subjective opinions necessary for informed decision-making, while many users provide limited textual feedback due to the effort required to compose detailed reviews. To bridge this gap, we introduce image-grounded review assistance, a novel task that aims to generate editable review drafts from user-uploaded product images. Unlike conventional image captioning, which focuses on objective visual description, the proposed task requires product-specific understanding, sentiment estimation, and evidence-driven review composition under challenging real-world conditions, including degraded image quality, excessive zoom-in, target ambiguity, and partial product visibility. We propose a multi-agent vision-language framework consisting of four specialised roles: product grounding, visual sentiment estimation, visual evidence generation, and review synthesis. The framework employs explicit intermediate representations, including product entities, predicted ratings, and evidence summaries, to improve interpretability and visual grounding. Experiments on a curated subset of the Amazon Reviews Electronics dataset demonstrate the feasibility of generating coherent, product-aware, and sentiment-aware review drafts from visual feedback. To the best of our knowledge, this is the first study to formulate image-grounded review assistance as a multi-agent vision-language reasoning problem, providing a practical step toward AI-assisted review authoring in e-commerce systems.

[71] Zero-shot video highlight detection based on text descriptions and synthetic images cs.CVPDF

Michal Byra, Alberto Presta, Grzegorz Stefanski, Krzysztof Arendt

TL;DR: 论文提出零样本视频高光检测框架，使用视频标题或类别让大模型生成事件描述，再用扩散模型合成视觉原型并与帧级 CLIP 特征匹配。

Details

Motivation: 视频高光检测需要定位最有信息或吸引力的片段，但收集逐帧高光标注成本高，且训练集专用方法难以迁移到新视频或领域。

Result: 框架以轻量视频元数据为条件，由大模型描述可能的高光事件，再由扩散模型生成相应图像原型，最后通过 CLIP 将文本和视觉表征与视频帧匹配，实现无需高光标签和数据集训练的逐帧检测。在 TVSum 与 SumMe 上实验，尤其在 TVSum 上取得较好零样本效果；摘要未提供具体指标。

Insight: 把元数据转成语义和视觉原型可为高光定位提供无需人工时序标注的先验；零样本表现仍依赖元数据对事件的描述质量，且摘要中的效果未量化。

Abstract: Detecting video highlights, the most informative or engaging moments in a video, is important for applications such as video summarization and content recommendation. We propose a zero-shot framework that combines CLIP, large language models (LLMs), and diffusion models. Given lightweight video metadata, such as a title or category, an LLM generates textual descriptions of likely highlight events. These descriptions are further converted into synthetic visual prototypes using a diffusion model. Textual and visual representations are matched to video frames using CLIP, enabling frame-level highlight detection without highlight annotations or dataset-specific training. Experiments on TVSum and SumMe demonstrate strong zero-shot performance, with particularly favorable results on TVSum. The proposed approach provides an effective framework for metadata-conditioned zero-shot video highlight detection.

[72] MoVT: Video-Augmented Motion Tokenizer for Text-to-Motion Generation cs.CVPDF

Beibei Jing, Tianle Guo, Youjia Zhang, Zikai Song, Yawei Luo

TL;DR: MoVT 用视频增强动作 tokenizer 将人类动作视频中的真实动作模式注入三维动作码本，再以掩码 Transformer进行文本到动作生成。

Details

Motivation: 文本驱动三维人体动作生成受标注动作数据规模有限制，难以覆盖自由文本提示涉及的多样真实动作。

Result: 跨模态动作 tokenizer 将离散 3D 动作 token 投影到 2D，利用视频动作模式丰富码本，再映射回 3D 形成对齐的 2D/3D codebook。生成端用模态无关掩码 Transformer 预测被遮盖的动作 token，并加入基于二维码本与标注视频生成的文本—索引对训练。多个关键指标上的实验表现优于或媲美既有先进方法；摘要未列具体分数。

Insight: 海量公开视频可作为动作数据补充，通过对齐二维视频动作与三维离散码本降低标注数据限制；效果依赖 2D—3D 码本对齐质量，摘要未提供视频来源与迁移细节。

Abstract: Text-driven 3D human motion generation models face significant challenges in responding to diverse and unconstrained textual prompts, primarily due to the limited availability of 3D motion training data. To address this, we introduce MoVT, a novel framework that effectively leverages the extensive range of human action videos to enhance text-to-motion generation. At the core of our approach is the cross-modal augmented motion tokenizer, which projects discrete 3D motion tokens into the 2D domain. This projection allows us to enrich the motion codebook with complex, real-world motion patterns derived from videos. The enriched discrete tokens are then mapped back to the 3D domain, resulting in aligned 3D and 2D codebooks with an enhanced capacity to represent intricate motions. These enhanced codebooks are integrated into a generative masked transformer, which predicts masked motion token indices in a modality-agnostic manner. This enables the use of text-index pairs, generated from the 2D codebook and annotated motion videos, to further enhance the generator. Extensive empirical evaluations show that MoVT performs favorably against prior state-of-the-art methods across multiple key metrics.

[73] PhysBrain 1.5: From Vision-Language Models to Physical Foundation Models cs.CV | cs.ROPDF

DeepCybo Team, Yu Bin, Haipeng Cao, Zheng Chang, Kai Chen

TL;DR: PhysBrain 1.5 将语言理解、动作生成和未来状态预测统一为离散序列自回归模型，在 28 个具身理解基准上报告 8B 模型平均分 72.5。

Details

Motivation: 具身系统需围绕观察、交互和环境变化形成物理闭环，但现有视觉语言模型主要回答问题，缺少统一理解场景、生成动作并预测后果的能力。

Result: 模型把语言回答、末端执行器运动和稠密视觉目标编码为离散序列，以自回归下一 token 目标共同训练；预训练监督来自人类交互视频，微调混合人类示范、机器人轨迹和模拟经验。28 个具身理解基准上，8B 模型平均得分 72.5，达开源先进水平并与部分专有系统相当，14 项基准取得开源最佳。定性示例展示末端轨迹及空间对齐的 RGB、深度和机器人掩码未来场景预测。

Insight: 统一动作、语言和视觉目标的 token 化训练可扩展视觉语言模型到物理交互预测；基准成绩侧重理解评估，动作执行和真实机器人闭环效果在摘要中仅作定性展示。

Abstract: We present PhysBrain 1.5, a unified model for understanding physical environments, generating actions, and predicting future states. Motivated by the physical loop of observation, interaction, and environmental change, we bring these capabilities into a common learning framework. Starting from a general vision–language model, we encode language responses, end-effector motion, and dense visual targets as discrete sequences and jointly optimize them with autoregressive next-token prediction. Pre-training draws its embodied supervision entirely from human interaction videos, using task-centered episodes to pair semantic and spatial context with recovered motion and subsequent observations. We then adapt the model through supervised fine-tuning on a mixture of human demonstrations, robot trajectories, and simulated experience. Across 28 embodied understanding benchmarks, our 8B model achieves an average score of 72.5, setting a new open-source state of the art and performing on par with leading proprietary models such as GPT-6-Astra and Gemini 3.6 Flash. It achieves the best open-source results on 14 benchmarks while retaining general multimodal capabilities. Beyond these understanding evaluations, qualitative examples show the model’s ability to produce end-effector trajectories and predict future scenes through spatially aligned RGB, depth, and robot-mask outputs.

[74] G-ray: Ray-Level Relative Geometric Position Encoding in Multi-View Vision Transformers under Camera Heterogeneity cs.CVPDF

Shuo Zhang, Xin Su, Wei Wang, Jun Liu, Xinrui Zeng

TL;DR: G-ray 用相机本地射线角定义多视角视觉 Transformer 的相对旋转位置编码，使同一相机射线关系在不同投影模型下具有相同相位。

Details

Motivation: 不同视场或投影相机的图像平面坐标不同，传统 RoPE 等图像平面相对位置编码会导致跨投影注意力的相位与几何线索不一致。

Result: G-ray 按相机本地射线角参数化旋转相位，可直接使用或集成到 RoPE、GTA、RayRoPE 中，不增加可学习参数。三个异构 3D 重建基准、50 视角设置下，六项平均指标均领先；相较使用相同标定的 MapAnything，点图相对误差平均下降 45.8%。仅以同质针孔图像训练的模型可无需重训处理针孔与非针孔混合输入；NVS 中 GTA 和 RayRoPE 在视角与视场变化时也因 G-ray 改善。

Insight: 射线级几何关系比图像坐标更适合作为异构相机间的相对位置编码；跨投影泛化和误差改善较显著，具体指标依赖相机标定及基准视角数。

Abstract: We study relative position encoding for multi-view vision Transformers under camera heterogeneity, including varying fields of view (FoVs) or projection models. Existing rotary relative position encodings commonly use image-plane positional coordinates, producing projection-dependent relative phases and inconsistent geometric cues for cross-projection attention. We introduce G-ray, a ray-level relative position encoding whose rotary phases are parameterized by camera-local ray angles. The same camera-local ray pair induces the same relative phase across projections, providing projection-invariant positional consistency. G-ray can be used directly or integrated with existing encodings, retaining complementary geometric cues without additional learned parameters. We validate G-ray in three host encodings, RoPE, GTA, and RayRoPE, across 3D reconstruction and novel-view synthesis (NVS). Across three heterogeneous 3D reconstruction benchmarks at 50 views, G-ray leads all six averaged metrics and reduces mean pointmap relative error by 45.8% over MapAnything, with calibration supplied to both. Trained exclusively on homogeneous pinhole images, the 3D reconstruction model handles mixed pinhole and non-pinhole inputs without retraining and remains competitive on homogeneous pinhole 3D reconstruction protocols. For NVS, GTA and RayRoPE improve with G-ray under joint viewpoint and FoV variation. The project’s webpage is available at https://g-ray-project.github.io/.

[75] TwinICL: Diagnosing Multimodal In-Context Learning through Paired Counterfactuals cs.CV | cs.LGPDF

Zihan Xue, Po-Yi Lu, Serhii Honcharenko, Zih-Ching Chen, Hsuan-Tien Lin

TL;DR: TwinICL 通过程序生成配对的文本与图像演示任务比较多模态上下文学习，发现六个开源模型在 38 个任务上多模态 ICL 一贯落后于文本 ICL。

Details

Motivation: 现有 ICL 基准缺少匹配的文本与图像任务版本，难以控制任务差异并比较模态造成的上下文学习落差。

Result: TwinICL 提供程序生成的文本—图像配对任务，评估六个开源模型的 38 个任务。多模态 ICL 持续低于纯文本 ICL，差距随任务族变化。针对视觉访问、任务框定和推理的三项干预单独效果有限或不一致，组合后在诊断子集上恢复较强多模态 ICL。显式给出任务指令后仍有模态差距；进一步分析表明示例既增加待处理上下文，也提供任务本身证据。

Insight: 多模态 ICL 的困难不只是看不清图像，还包括从演示推断任务规则；成对基准让模态差异可控，组合干预的效果仍限于诊断子集。

Abstract: In-context learning (ICL) enables models to infer tasks from demonstrations, but existing benchmarks generally lack matched text and image versions needed to compare ICL performance across modalities. We introduce TwinICL, a procedurally generated benchmark providing such pairs for controlled comparison. Across six open-weight models and 38 tasks, multimodal ICL consistently underperforms text-only ICL, with gaps varying by task family. To test whether this gap can be recovered, we target visual access, task framing, and reasoning through three interventions. Their combination recovers strong multimodal ICL performance on a diagnostic subset, despite limited or inconsistent individual effects. To distinguish difficulties in executing tasks from those in inferring them, we evaluate models with explicit task instructions, revealing a modality gap even when the task is known. We then examine how adding demonstration inputs and outputs reshapes this gap, highlighting demonstrations’ dual role as additional context to process and evidence about the task. The dataset is available at https://github.com/lab-flair/TwinICL.

[76] Tele360: Real-Time Feed-Forward Human Reconstruction from Sparse Unposed Cameras cs.CVPDF

Hanzhang Tu, Zhanfeng Liao, Wei Min, Jiajun Zhang, Yebin Liu

TL;DR: Tele360 从稀疏、未标定的多摄像机 RGB 流中一次前向估计相机位姿并重建动态 3D Gaussian 人体场景，实现实时自由视点可视化。

Details

Motivation: 现有动态人体重建要么依赖昂贵优化，要么需要标定相机和低分辨率输入，难以在普通 GPU 上实时处理高分辨率稀疏视频。

Result: 系统以稀疏感知多视角 Transformer 编码人体前景并保留共享场景 token，再用 Transformer 高斯解码器减少卷积过平滑，混合特征金字塔提供多尺度外观线索，轻量可微 Levenberg–Marquardt 层细化相机几何，并从视觉几何基础模型蒸馏先验。预测高斯通过视频编解码传输至远端。工作室基准上视觉质量达到先进水平，单张消费级 GPU 可支持 2K 输入到渲染超过 25 FPS。

Insight: 联合估计相机和动态高斯场景避免稀疏未标定输入下的分步优化瓶颈，几何先验蒸馏帮助稳定重建；实时结果针对特定多摄像机设置，部署硬件与遮挡范围需参照正文。

Abstract: Live free-viewpoint visualization of real humans is critical for immersive communication and interactive digital experiences. Existing methods either rely on computationally expensive optimization or require calibrated cameras and low-resolution inputs, making real-time high-resolution deployment impractical. In this work, we present Tele360, the first real-time feed-forward system for dynamic human reconstruction and live free-viewpoint visualization from sparse, unposed RGB streams. Our system jointly estimates camera poses and reconstructs a dynamic 3D Gaussian representation for each time instance in a single forward pass. To achieve this, we start by designing a lightweight sparsity-aware multi-view transformer backbone that tokenizes foreground human regions while preserving global context through a shared scene token. We then employ a fully transformer-based Gaussian decoder to mitigate convolution-induced over-smoothing while keeping decoding sparse and efficient. In addition, we introduce a hybrid feature pyramid that injects multi-scale appearance cues into geometry prediction. We further introduce a lightweight differentiable Levenberg-Marquardt camera refinement layer to enhance multi-view consistency and geometric alignment. Moreover, to stabilize learning under sparse, unposed inputs, we transfer multi-view geometry priors from a large visual-geometry foundation model via teacher-student distillation. Finally, the predicted Gaussian maps are streamed with video codecs to remote devices for interactive free-viewpoint rendering. Extensive experiments show that Tele360 achieves state-of-the-art visual quality on studio benchmarks while supporting real-time 2K input-to-rendering at over 25 FPS on a single consumer GPU. Additional captured sequences illustrate its performance across varied subjects, clothing, and motions under our multi-camera setup.

Jianhe Zhao, Yanhua Qiu, Zhiyu Zhang, Zibo Zhao, Jinhua Xie

TL;DR: LG-VLN 是仅用单目 RGB 的零样本连续环境视觉语言导航框架，以在线三维重建共享几何和语义特征，并用 LangGraph 显式编排导航状态。

Details

Motivation: VLN-CE 要在未见三维环境中理解指令并执行连续低层动作，现有方法常依赖 LiDAR 或全景相机；几何建图与语义导航表征分离还会造成长轨迹空间语义不一致。

Result: 在线前馈三维网络预测深度、相机位姿和稠密点云，用于代理位姿和全局地图融合；共享 CleanDIFT 特征支持几何一致性筛除和目标实例约束，结合 BLIP-2 局部图文相关度形成语义价值图。LangGraph 将解析、感知、价值更新、规划、执行和失败恢复表示为带条件转移和持久状态的有向状态图。在 R2R-CE val-unseen 固定 550 集上成功率 21.3%，路径长度加权成功率 12.1%。

Insight: 几何与语义共享视觉特征并用显式状态编排，可在不依赖额外传感器时支持零样本导航；绝对成功率仍较低，说明连续行动与恢复仍有明显提升空间。

Abstract: Continuous-environment vision-and-language navigation (VLN-CE) requires interpreting natural-language instructions in unseen 3D environments and executing continuous low-level actions. Existing methods often depend on LiDAR, panoramic cameras, or extra sensors; separate geometric-mapping and semantic-navigation visual representations can cause long-trajectory spatial-semantic inconsistencies. We propose LG-VLN, a monocular zero-shot framework with shared visual features and LangGraph-based state orchestration. An online feed-forward 3D reconstruction network predicts depth, camera poses, and dense point clouds for agent-pose estimation and global map fusion. Geometry and navigation share dense CleanDIFT features: semantic consistency rejects incorrect inter-frame correspondences, while target-instance constraints define visual references whose similarity combines with local BLIP-2 image-text relevance to form a semantic value map. LangGraph represents instruction parsing, geometric perception, semantic value updates, path planning, action execution, and failure recovery as a directed state graph with conditional transitions, persistent state, and modular recovery mechanisms. On a fixed 550-episode subset of the R2R-CE val-unseen split, LG-VLN achieves 21.3% success and 12.1% success weighted by path length. Ablations show shared semantic features improve navigation, further boosted by combining visual similarity and image-text relevance. Results establish shared visual representations and explicit state orchestration as effective for zero-shot VLN-CE using monocular RGB alone. Code will be publicly released for reproducibility.

[78] AdaVSkip: Adaptive Visual Token Skipping Across Layers For Efficient MLLMs Inference cs.CV | cs.AIPDF

Yuyao Sun, Tao Deng, Shuang Li, Deqing Wang

TL;DR: AdaVSkip 为每层自注意力和 MLP 配置独立路由器，按输入动态跳过多模态大模型的视觉 token 计算，并以监督加强化学习训练路由策略。

Details

Motivation: MLLM 推理需将大量视觉 token 送过所有 Transformer 层；压缩减少 token 数量，早退和固定跳层利用层间冗余，但注意力与 MLP 冗余会随输入变化。

Result: AdaVSkip 冻结骨干，只训练逐层两路路由器。第一阶段用模块必要性分数产生的输入特定标签监督初始化策略，第二阶段用回答正确奖励和跳过一致性奖励做强化学习，以性能反馈优化离散决策。在三个骨干上维持较强任务表现；LLaVA-NeXT-7B FLOPs 减少 53.2%且平均表现持平，与视觉 token 压缩结合后 FLOPs 减少 91.2%，保留原模型平均表现的 97.2%。

Insight: 按输入动态分配自注意力和 MLP 计算可补充 token 压缩带来的效率收益；强化学习直接对答案质量反馈路由策略，组合压缩显著省算力但有约 2.8% 平均性能损失。

Abstract: Multimodal large language models (MLLMs) require substantial computation to process numerous visual tokens across all transformer layers. Most methods for efficient MLLM inference exploit horizontal redundancy by compressing visual tokens. Beyond token reduction, recent studies exploit vertical redundancy through early exit or fixed-layer skipping. However, we find that the extent and distribution of this redundancy vary across inputs and differ between self-attention and MLP modules. Motivated by these observations, we propose AdaVSkip, which equips each layer with two lightweight routers that independently determine whether visual tokens pass through by or skip the self-attention and MLP modules. These decisions collectively define an input-specific visual-computation path, but their discrete and non-differentiable nature makes learning effective paths challenging. To address this challenge, we develop a progressive two-stage training framework that updates only the routers while keeping the backbone frozen. Stage I establishes an initial routing policy through supervised training with input-specific targets derived from module-wise necessity scores. To further align the routing policy with task performance, Stage II uses reinforcement learning to optimize routing decisions with direct feedback from generated answers. It combines an answer correctness reward with a skip-consistency reward that discourages excessive retention of visual-token computation. Across three MLLM backbones, AdaVSkip maintains strong task performance with substantially less computation. On LLaVA-NeXT-7B, AdaVSkip reduces FLOPs by 53.2% while preserving the original model’s average performance. Combining it with visual token compression increases this reduction to 91.2%, while retaining 97.2% of the original performance on average.

[79] SparseTalk - Sparsifying 3D Gaussian Language Fields for Efficient 3D Visual Question Answering cs.CV | cs.LGPDF

Davit Soselia, Joseph JaJa, Amitabh Varshney

TL;DR: SparseTalk 将 3D Gaussian 语言场稠密语义嵌入稀疏化，发现 3D 问答可用不到 1% 的表示保持较强表现。

Details

Motivation: 3D Gaussian 语言场通过空间对齐语义支持 3D VQA，但每场景可能需要数万嵌入，带来存储、内存和推理开销；实际推理所需语义表示冗余量不清楚。

Result: 作者从完整嵌入出发，系统比较随机、几何、语义及联合选择策略，覆盖低至 8 个视觉 token 的极稀疏区间；另提出按检测对象分配预算并保留背景上下文的对象级稀疏化。在 ScanQA、MV-ScanQA 上，只保留数百个语义嵌入（不到原表示 1%）仍维持较强 VQA。256 token 时，SparseTalk 仅保留 SplatTalk 32,076-token 推理输入的 0.80%，以及平均 77,207 高斯稠密场的 0.332%；解码特征内存降低 125 倍。

Insight: 稠密三维语言场存在大量不影响问答的语义嵌入，按对象分配预算能以极少表示维持效果；极稀疏下游性能仍需关注具体场景和长尾对象。

Abstract: 3D Gaussian language fields provide an explicit, spatially grounded representation for 3D visual question answering (VQA), but their dense semantic features can require tens of thousands of embeddings per scene, resulting in substantial storage, memory, and inference costs. We investigate how much of this representation is actually necessary for downstream reasoning. Starting from a full embedding representation, we systematically sparsify its semantic embeddings, including the previously underexplored regime below a single image-equivalent block down to 8 visual tokens. We compare random, geometric, semantic, and joint spatial-semantic selection strategies and introduce an object-based sparsification method that distributes the token budget across detected object instances while retaining background context. Experiments on ScanQA and MV-ScanQA reveal substantial redundancy in dense Gaussian language fields. Strong VQA performance is retained with only a few hundred semantic embeddings, corresponding to less than 1% of the original representation. Object-based selection performs well relative to others, with only modest observed changes down to 256 tokens. At this budget, SparseTalk retains 0.80% of SplatTalk’s 32,076-token inference input and 0.332% of the mean 77,207-Gaussian dense field, increasing inference throughput while reducing decoded-feature memory 125-fold.

[80] Weakly Supervised Spatial Grounding for Discriminative Attention-Based Ultrasound-Histopathology Alignment in Prostate Cancer Grading cs.CVPDF

Obed Korshie Dzikunu, Emma Willis, Mohammad Mahdi Abootorabi, Mohamed Harmanani, Zhuoxin Guo

TL;DR: 研究将常规活检记录中的受累比例转成弱空间监督，独立约束超声特征并与病理教师对齐，提升前列腺癌等级判别与注意力集中度。

Details

Motivation: 无配对跨模态蒸馏常以同一对齐目标同时要求超声特征区分组织状态并选择重要 patch，导致注意力近乎均匀；需要将表征学习和空间选择监督分开。

Result: 作者用活检核心的百分比受累程度监督预测恶性组织比例，约束编码器特征；对齐损失则在核心内变化的特征上工作，使注意力聚焦部分 patch。在 7 个中心、811 名患者的 7,166 个活检核心上进行患者级五折交叉验证，宏 AUC 为 67.1、临床显著癌症 AUC 为 68.5；已有无配对对齐方法为 61.2/52.8，最强单模态基线为 63.1/62.6。消融显示防止注意力均匀塌陷的正则不能替代直接作用于特征的标签派生监督。

Insight: 仅限制注意力分布并不足以让被关注特征具诊断差异性，弱比例标签需直接塑造编码表征；方法比既有对齐和单模态基线有提升，但宏 AUC 数值仍表明任务有较大改进空间。

Abstract: Unpaired cross-modal distillation transfers grade structure from histopathology into a micro-ultrasound (micro-US) encoder by aligning a pooled needle-region embedding to a frozen histopathology teacher under grade-group correspondence alone. A single objective is thereby required to serve two distinct functions: rendering patch features discriminative of tissue state, and selecting which patches enter the pooled representation. We decouple them. Weak spatial supervision derived from percentage involvement, recorded routinely at biopsy, constrains the predicted proportion of malignant tissue within each core, acting on the encoder features independently of the alignment objective. The alignment loss then operates on features that differ across a core, and attention concentrates on a subset of patches rather than remaining near-uniform. On 7,166 biopsy cores from 811 patients across seven centers under patient-level 5-fold cross-validation, the method reaches 67.1 macro AUC and 68.5 csPCa AUC, against 61.2 and 52.8 for the existing unpaired alignment method and 63.1 and 62.6 for the strongest unimodal baselines. Ablation against existing attention regularizers designed to prevent attention-uniformity collapse shows that such regularizers do not substitute for label-derived supervision: they constrain the attention distribution, whereas the signal required acts on the features that attention reads.

[81] PACE: Progressive Angular-to-Norm Contrastive Embedding cs.CV | cs.AI | cs.CL | cs.LGPDF

Yanping Li, Wei Zhou, Yawen Liu, Yibo Wang, Ke Zhu

TL;DR: PACE 分两阶段训练多模态嵌入：先用余弦目标和 LoRA 建立稳定角度几何，再切换点积和全参数微调，让向量范数也承载语义信息。

Details

Motivation: 余弦对比学习训练稳定但只使用角度信息，无法让嵌入范数提供语义信号；直接优化更具表达力的点积相似度又容易训练不稳和性能下降。

Result: 作者将不稳定归因于优化空间过早扩张造成的角度—范数纠缠、方向各向异性以及全参数微调放大效应。PACE 第一阶段用余弦目标和低秩适配在受限空间确立角度几何，第二阶段改用点积并全参微调，使方向与范数共同编码语义；另提出置信度自适应 Focal Embedding Loss，降低正例置信度高查询权重、强调存在竞争负例的模糊查询。多个骨干规模和多模态任务实验均支持方法有效，摘要未列具体增益。

Insight: 点积目标的潜力需要渐进式释放，先稳定角度结构再引入范数语义可缓解优化不稳定；模糊查询加权帮助模型聚焦难负例，具体任务收益仍需看分项指标。

Abstract: Multimodal embedding models encode heterogeneous inputs into a shared embedding space, enabling efficient similarity computation across modalities and tasks. Most existing methods optimize cosine-based contrastive objectives, which promote stable training but restrict semantic compatibility to angular geometry, precluding embedding norms from serving as an additional semantic signal. However, directly optimizing the more expressive dot-product similarity, which leverages both angular and norm information, underperforms cosine-based training and exhibits unstable training dynamics. We attribute this discrepancy to premature optimization-space expansion, manifested as angular–norm entanglement and directional anisotropy in the representation space and further compounded by full-parameter fine-tuning. In this paper, we propose PACE, a two-stage framework that progressively expands both the representation and trainable parameter spaces. Stage I combines cosine-based objective with low-rank adaptation to establish a reliable angular geometry within constrained optimization spaces. Stage II switches to dot-product similarity and full-parameter fine-tuning, enabling embedding directions and norms to jointly encode semantic information. We further introduce Focal Embedding Loss, a confidence-adaptive objective that downweights queries with high positive retrieval confidence while emphasizing ambiguous queries with competitive negatives. Experiments across multiple backbone scales and diverse multimodal embedding tasks consistently validate the effectiveness of PACE.

[82] GRAVA: Grounded Reasoning-to-Action Representation and Learning for Autonomous Driving cs.CV | cs.ROPDF

Xiao Liu, Haoyu Li, Jianghao Leng, Lin Wang, Chao Sun

TL;DR: GRAVA 将物理场景证据、推理和可执行动作统一为自回归表示，配套构建 220 万问答对及 7 万推理轨迹数据集，改善自动驾驶中推理到行动的落地。

Details

Motivation: 驾驶 VLA 模型会先推理再行动，但中间推理常与场景物理证据联系弱，也难以转成可执行行为，造成语言解释和车辆动作脱节。

Result: GRA 表示将行动相关语言指代锚定到二维区域与自车物理状态，用轨迹锚定类型图组织交互决策，再由单一 VLM 输出推理及可确定性解码为连续轨迹的紧凑规划动作。GR-NavSim 含 220 万落地问答对和 7 万推理轨迹；渐进训练依次进行预训练、模仿和强化学习。仅使用约 60% 可用人工示范监督动作时，GRAVA-8B 在完整 NAVSIM 基准纯自回归驾驶模型中达先进表现；内部长尾基准中，完整 GRA 相较仅动作预测使关键对象遵循率提高 19.3%、闭环驾驶分提高 20.5%。

Insight: 把对象证据、轨迹锚定推理和可执行规划动作保存在同一生成流中，可提升长尾场景里的关键目标遵循；完整 GRA 的收益来自内部长尾基准，具体部署安全性仍需实际道路验证。

Abstract: Driving vision-language-action (VLA) models increasingly reason before acting, but their intermediate reasoning is often weakly grounded in physical scene evidence and loosely connected to executable behavior. We present GRAVA, a framework built around Grounded Reasoning-to-Action (GRA), which unifies grounding, reasoning, and action generation in a single autoregressive stream. GRA links action-relevant language references to 2D visual regions and ego-centric physical states, organizes object interactions and decisions in a trajectory-anchored typed graph, and serializes this structure into grounded reasoning. A single VLM generates this reasoning followed by a compact Executable Planner action that is deterministically decoded into a continuous trajectory. We further introduce an agentic GRA data construction pipeline that combines forward scene grounding with backward trajectory anchoring, and use it to build GR-NavSim with 2.2M grounded question-answer pairs and 70K GRA reasoning traces. A progressive training strategy develops grounded cognition through pre-training, establishes the reasoning-to-action interface through imitation, and improves driving behavior through reinforcement learning and exploration. Using about 60% of the available human driving demonstrations for action supervision, GRAVA-8B achieves state-of-the-art performance among purely autoregressive driving models on the full NAVSIM benchmark. On an internal long-tail benchmark, full GRA improves key-object compliance and Closed-loop Driving Score by 19.3% and 20.5% over action-only prediction, respectively. These results show the benefit of preserving action-relevant physical evidence from grounded reasoning through executable action generation.

[83] Does Attention-Guided Masking Really Help Object Discovery in Object-Centric Learning? cs.CVPDF

Youliang Tao, Yanhua Han, Bin Zhao, Juho Kannala, Joni Pajarinen

TL;DR: 六个目标发现数据集上的实验显示，基于内部注意力生成掩码不总能改善对象中心学习；纹理丰富数据的背景分割可能获益，但前景发现通常持平或变差。

Details

Motivation: 对象中心学习以 Slot Attention 聚合特征，再通过随机遮挡图像块进行自监督；既然注意力本身含对象信息，作者研究用注意力引导遮挡能否提供更好的自监督信号。

Result: 作者提出 Attention-Guided Masking，与均匀随机遮挡比较六个常用数据集。无条件 slot 初始化时，AGM 在 COCO、VOC 等真实纹理数据的背景分割显著改善；但不论条件或无条件初始化，前景对象发现性能总体相近或下降。作者据此提醒，利用内部注意力语义改进遮挡重建可能存在风险。

Insight: 模型注意力具有对象信息不代表据此采样遮挡就能提升对象发现，前景分割并未稳定获益；结果提示需要按背景与前景分别评估，而不能只报告单一对象发现总分。

Abstract: Object-Centric Learning (OCL) aims to decompose images into objects without human annotations. A major family of mainstream methods uses Slot Attention to aggregate image features into object-level representations and then from them reconstructs masked image content, i.e., Random Masking (RM), to provide self-supervision. The recent method DIAS simply masks image patches at uniform randomness yet achieves competitive object discovery accuracy. Since attention during aggregation already possesses object discovery ability, we explore using it to develop a better image patch masking strategy, i.e., Attention Guided Masking (AGM), thereby providing better self-supervision. Results on six recognized datasets show that AGM does not always outperform RM. Under unconditional slot initialization, AGM substantially improves background segmentation on datasets with realistic textures (COCO and VOC); Regardless of conditional or unconditional slot initialization and across datasets, foreground object discovery remains comparable or decreases. We suggest peer researchers in the OCL community that attempts to exploit internal attention semantics to improve OCL with masked decoding are risky. Our source code, model checkpoints and evaluation logs will be released upon acceptance.

[84] Reconstructing Is Not Acting: Action-Centric Latent Dynamics Modeling cs.CVPDF

Dingjie Fu, Dianxing Shi, Yangyang Xu, Jun Yu

TL;DR: ACT-LAM 针对潜动作模型的“重建好不代表动作学得好”问题，用可学习动作查询提取转移线索，并以动作 token 持续条件化前向状态演化。

Details

Motivation: 潜动作模型从无标注视频推断动作并重建未来状态，但低重建误差不必然意味着潜动力学正确或下游规划有效；反向动作模型可能抓取外观干扰，前向模型可能走状态预测捷径而忽视动作。

Result: ACT-LAM 的 AQ-IDM 使用可学习动作查询和门控聚合筛选动作相关变化，AT-FDM 将潜动作投射成动作 token，逐步与演化状态交互，实现持续状态感知条件化；同时精简特征处理以集中容量于动力学建模。在多个机器人数据集和 VP² 基准上，动作一致性、前向动力学及下游视觉规划更强，参数更少、计算开销更低；VP² 聚合成功率比此前最佳提高 7.6%。

Insight: 动作表征需同时能从状态变化中被辨认，并在前向预测中被真正使用；仅以重建损失训练可能掩盖两种失配，ACT-LAM 通过双侧结构约束改善规划表现。

Abstract: Latent action models (LAMs) learn action representations from unlabeled videos by inferring latent actions from visual transitions and reconstructing future states. However, we identify a fundamental $\textbf{reconstruction-action mismatch}$: lower reconstruction error does not necessarily yield better latent dynamics or downstream performance. We attribute this mismatch to two underconstrained aspects of reconstruction-based latent dynamics modeling: (i) the inverse dynamics model (IDM) is not explicitly encouraged to distinguish action-related transitions from nuisance appearance, and (ii) the forward dynamics model (FDM) can underutilize the inferred latent action by exploiting predictive shortcuts from the current state. To address both limitations, we propose $\textbf{ACT-LAM}$, a lightweight action-centric framework that strengthens both action extraction and action utilization. Specifically, its Action Query IDM (AQ-IDM) employs learnable action queries and gated aggregation to selectively extract rich action-related transition cues without strong information bottlenecks. And its Action Token FDM (AT-FDM) projects latent actions into action tokens that progressively interact with evolving state representations, enabling continuous state-aware action conditioning. ACT-LAM further streamlines feature processing to concentrate model capacity on latent dynamics modeling. Extensive experiments on several robotic datasets and the VP$^2$ benchmark demonstrate stronger latent action consistency, forward dynamics, and downstream visual planning performance with fewer trainable parameters and lower computational overhead. In particular, ACT-LAM surpasses the previous state of the art by $\textbf{7.6%}$ on the aggregated VP$^2$ success rate. Codes at $\href{https://github.com/DingjieFu/ACT-LAM}{url}$.

[85] Long-to-Short Video Evidence Reasoning for Grounded Question Answering cs.CVPDF

Kaiyan Chen, Junbin Xiao, Xun Yang

TL;DR: LOVER 用从长程证据到短时推理的课程训练、重叠度奖励和自适应时间戳渲染，提升视频问答的证据定位与落地问答。

Details

Motivation: 现有视频推理强化学习方法对证据持续时间缺少渐进适配，定位奖励也可能不适合证据检出，且视频帧上的时间戳不易观察。

Result: LOVER 提出长到短视频证据课程，让模型从长时程定位逐步过渡到短时推理；以 IoP 奖励强调证据检出而非严格时间段 IoU；并根据背景自适应选择时间戳位置和颜色，提高帧上时间可见性。三个模块可适配不同骨干且相互补充，在 QA、grounding 和 grounded QA 上均有改善。基于 Time-R1 的 LOVER 在 NExT-GQA 和 ReXTime 两个常用基准上达到开源模型新 SOTA，消融支持三项设计。

Insight: 证据持续时间、定位奖励形式和时间戳可读性共同影响视频推理强化学习，长到短的课程衔接全局定位与局部推理；摘要未给出具体数值提升。

Abstract: We present LOVER, a \underline{L}ong to sh\underline{O}rt \underline{V}ideo \underline{E}vidence \underline{R}einforced model for grounded question answering (GQA). LOVER highlights three innovations over existing reinforcement-learning (RL) based video reasoning models: (1) \textbf{Long-to-short Video Evidence Curriculum Learning}, which organizes RL training according to evidence duration and progressively adapts the model from long-range grounding to short-term reasoning; (2) \textbf{GQA Rewards}, which underscore the benefit of IoP reward over IoU for evidence spotting rather than strict temporal span overlap; (3) \textbf{Adaptive Timestamp Rendering}, which adaptively renders timestamps onto video frames using background-aware position and color selection to enhance temporal observability. The three designs are model-agnostic and reciprocal. They effectively improve QA, grounding, and grounded QA performance over different backbones. Notably, LOVER built on Time-R1 achieves new state-of-the-art (SOTA) results among open-source models on popular GQA benchmarks: NExT-GQA and ReXTime. Comprehensive ablation studies further validate the effectiveness of our three innovative components.

[86] Unsupervised Point Cloud Registration via Training-Time Semantic Guidance cs.CVPDF

Kezheng Xiong, Shiyun Xu, Sheng Ao, Siqi Shen, Cheng Wang

TL;DR: CAESAR 用仅在训练时可见的 3D 分割教师保护无监督 LiDAR 点云配准中的语义结构，并以语义引导重匹配恢复被少量错误邻居遮挡的内点。

Details

Motivation: 户外点云几何歧义会产生噪声伪标签并导致无监督配准收敛不佳，nuScenes 等稀疏低分辨率扫描尤其明显；模型原生语义感知会被噪声监督迅速侵蚀。

Result: CAESAR 使用现成 3D 分割教师只在训练期提供语义指导，不需要配准训练数据的语义标注，也不增加推理成本。双线索引导重匹配从噪声特征邻域中重新选择被少量伪邻居掩盖的潜在内点；语义—几何标签挖掘以批次为单位轻量细化教师并构造可靠伪标签；语义预测蒸馏进一步巩固学生表征。在 KITTI 和 nuScenes 上达到先进配准性能，在困难 nuScenes 上增益显著。

Insight: 模型本身可能已含与配准相关的语义结构，训练时的短暂教师指导可保护这种结构并改善伪标签；教师语义仅用于训练降低部署依赖，但摘要未报告各基准具体指标。

Abstract: Unsupervised registration of large-scale LiDAR point clouds remains challenging due to the geometric ambiguity inherent in outdoor scenes, which degrades pseudo-label quality and leads to suboptimal convergence, particularly for sparse, low-resolution scans such as those from nuScenes. We reveal that registration models intrinsically encode semantic awareness that strongly correlates with registration accuracy, albeit without explicit semantic supervision. However, this native awareness is fragile: noisy supervision arising from geometric ambiguity in unsupervised settings rapidly erodes the learned semantic structure, causing performance collapse. To this end, we propose CAESAR, a teacher-student framework guided by an off-the-shelf 3D segmentation model exclusively during training. We observe that potential inlier matches are often buried just beneath a few spurious neighbors in the noisy feature space, motivating Dual-Cue Guided Re-Matching to recover them through reselection rather than simply rejecting. Building on this, a train-only Semantic-Geometric Label Mining performs lightweight, batch-specific teacher refinement and mines reliable pseudo-labels under semantic guidance. We further introduce Semantic Predictive Distillation to consolidate the student’s semantic awareness in the feature space. Extensive experiments on KITTI and nuScenes demonstrate state-of-the-art performance, with pronounced gains on the challenging nuScenes benchmark. Crucially, CAESAR incurs zero inference overhead and requires no semantic annotations on the registration data. Code will be released.

[87] Closed-form Bayesian homography estimation from noisy point correspondences cs.CV | stat.MEPDF

Hanne Beuter, Sebastian Dorn

TL;DR: 论文提出从噪声点对应估计单应矩阵的闭式贝叶斯方法，结合测量不确定性和先验知识，同时输出参数后验分布。

Details

Motivation: 传统单应估计通常只给点估计，无法直接量化噪声对应引入的不确定性，而该不确定性会传递到相机标定和三维重建等后续步骤。

Result: 方法在齐次坐标中推导单应矩阵后验均值闭式解，并用迭代贝叶斯方法处理非线性。在不同噪声条件的合成投影实验中，估计精度优于 DLT；真实图像拼接实验表明可用于图像对应关系并提供不确定性信息。摘要未给出具体误差数值。

Insight: 将对应点测量不确定性和先验纳入单应后验，可让下游系统传播几何估计置信度；闭式均值与迭代非线性处理相结合，适用于传统点估计难以表达风险的任务。

Abstract: While homographies are fundamental to many computer vision tasks, the majority of conventional estimation techniques provide only point estimates without directly quantifying uncertainty introduced by noisy observations. Uncertainty, though, propagates to subsequent processing steps such as camera calibration and 3D reconstruction and is particularly relevant in safety-critical and socially relevant fields including medical imaging, autonomous driving, and defense. We present a fast Bayesian formulation for homography estimation from point correspondences that explicitly incorporates measurement uncertainty and prior knowledge while providing a posterior distribution over the homography parameters. A closed-form solution of the posterior mean of the homography is derived in homogeneous coordinates and supplemented by an iterative Bayesian approach to handle non-linearities. Synthetic experiments demonstrate the applicability to projective transformations and show improved estimation accuracy over DLT under varying noise conditions. Image stitching experiments further demonstrate applicability to real image correspondences while additionally providing uncertainty information.

[88] Pre-PEFT Probing: Weight Statistics and Perturbation Robustness for Layer Selection in VLM Vision Encoders cs.CV | cs.AIPDF

Qingtao Xia, Jiahua Bao, Siyao Cheng, Jie Liu

TL;DR: 论文提出微调前的 VLM 视觉层探测方法，通过 Q/K/V 权重统计和参数扰动鲁棒性预判最值得进行 PEFT 的层。

Details

Motivation: 常见做法在所有视觉编码器层同时加 LoRA 等适配器，参数成本高，层选择又常凭经验；作者希望在微调前估计各层可适应性。

Result: 方法统计各 Transformer 层 Q/K/V 投影权重的范数、条件数等特征，并测量受控参数扰动后的鲁棒性，再与单层 PEFT 对下游表现的增益比较。在七个基准、五种 PEFT 变体中，权重范数较大、条件数较高的层或矩阵通常更抗扰，也更可能带来较大微调收益。由此可在减少可训练参数时维持或提升适配表现。

Insight: 无需先运行完整微调，也可用权重分布与扰动测试获得层选择信号，从而减少适配成本；观察到的相关性适用于所测模型和基准，不保证所有架构都遵循同一关系。

Abstract: We propose a pre-fine-tuning probing method for Parameter-Efficient Fine-Tuning (PEFT) layer selection, aiming to obtain more stable and higher gains with fewer trainable parameters when adapting large vision–language models (VLMs). Unlike the common practice of applying LoRA and other adapters to all layers at once—where layer selection often relies on heuristic rules—we focus on the vision encoder and directly evaluate the “adaptability’’ of each Transformer layer. Specifically, we characterize each layer from two perspectives: (i) the statistical properties of its Q/K/V projection weights (e.g., norms and condition numbers); (ii) robustness under controlled parameter perturbations. We then systematically compare these indicators with the downstream performance gains brought by applying PEFT to a single layer. Across experiments covering seven benchmarks and five PEFT variants, we observe a consistent correlation: layers (or matrices) with larger weight norms and higher condition numbers are usually more robust to perturbations and are more likely to yield larger fine-tuning gains. These results show that distribution-statistics analysis and perturbation tests before fine-tuning can provide practical signals for adaptation-layer selection, thereby maintaining or improving performance while reducing trainable parameters.

[89] A 25-$μ$s/inf Event-driven Graph Neural Network Processor with Spatiotemporal Caching and Spline Convolution for Ultra-low-latency AI at the Edge cs.CV | cs.ARPDF

Adrian Kneip, Martin Lefebvre, Daniel Gehrig, Victoria Catalán Pastor, Davide Scaramuzza

TL;DR: ETHEREAL 是面向动态视觉事件图神经网络的硬件加速器，结合邻域并行样条卷积和时空缓存，在单个事件上达到 25.6 微秒延迟。

Details

Motivation: 动态视觉传感器以微秒级逐像素产生事件，事件图神经网络混合规则密集计算与稀疏不规则访存，给超低延迟边缘硬件带来挑战。

Result: ETHEREAL 通过邻域并行样条卷积引擎及 2D/3D 分离内存层次扩展到 640×480 分辨率，并提出区域兴趣时空缓存机制。实测端到端推理延迟为每事件 25.6 微秒，能耗为每事件 1.7 微焦耳，覆盖先进事件视觉工作负载。

Insight: 事件级图处理必须同时优化计算并行和稀疏访存，时空 ROI 缓存有助于在高分辨率下保持低延迟；摘要没有列明具体任务精度和与基线的能耗对照。

Abstract: Dynamic-vision-sensor (DVS) cameras generate events on a per-pixel basis with a $μ$s-level temporal resolution, calling for new algorithm-hardware co-design approaches compared to standard frame-based vision. While event-driven graph neural networks (EV-GNNs) emerge as a promising algorithmic solution, they raise new HW challenges by mixing dense-regular compute operations and sparse-irregular memory accesses. We present ETHEREAL, the first EV-GNN accelerator that scales to 640$\times$480 resolutions, thanks to a neighbor-parallel spline convolution engine and a 2D/3D-split memory hierarchy with a novel region-of-interest spatiotemporal caching mechanism. Measurement results demonstrate end-to-end inference with 25.6$μ$s latency and 1.7$μ$J energy per event on state-of-the-art workloads

[90] AlignUS: MRI-Guided Ultrasound Representation Learning for ALS Classification from Tongue Images cs.CVPDF

Kadija Abdel Ghader, Emani Babe, Lorenzo Pettinari, Meya Haroune, Sidaty El Hadramy

TL;DR: AlignUS 用 MRI 表征蒸馏指导舌部高分辨率超声 ALS 分类，训练时跨模态对齐，推理只需要低成本超声。

Details

Motivation: ALS 早期评估困难，资源不足地区往往无法获得 MRI；舌部高分辨率超声便携且成本低，但数据集小、单靠超声特征难形成可靠分类表征。

Result: AlignUS 将分类损失、监督对比学习和特征级蒸馏结合，以 MRI 嵌入监督 HRUS 编码器，部署推理时不依赖 MRI。患者级四折交叉验证汇总平衡准确率 0.958、宏 F1 0.963、ROC-AUC 0.990，稳定优于超声基线和跨模态替代方法。

Insight: 训练阶段可借助 MRI 解剖知识增强超声表示，同时保留推理时对 MRI 的独立性，适合资源受限筛查设想；高分来自特定患者级交叉验证，仍需外部临床队列确认。

Abstract: Amyotrophic lateral sclerosis (ALS) is a progressive neurodegenerative disease in which early assessment remains challenging, particularly in low-resource settings where MRI is often unavailable. High-resolution ultrasound (HRUS) of the tongue offers a portable and low-cost alternative for evaluating bulbar involvement, but learning reliable diagnostic models is limited by small datasets and the difficulty of extracting robust representations from ultrasound alone. We propose AlignUS, a cross-modal knowledge distillation framework that transfers anatomical knowledge from MRI to a HRUS-based classifier while requiring only HRUS at inference time. The model combines classification loss, supervised contrastive learning, and feature-level distillation to align HRUS representations with MRI embeddings. AlignUS achieves a patient-level balanced accuracy of 0.958, macro-F1 of 0.963, and ROC-AUC of 0.990, aggregated across four patient-level cross-validation folds, with consistent improvements over HRUS baselines and cross-modal alternatives. These results demonstrate that MRI-derived supervision can substantially improve ultrasound-based ALS assessment while preserving low-cost, inference-time independence from MRI.

[91] Learning from Reliable Negatives: Confidence-Anchored Test-Time Adaptation for GUI Grounding cs.CVPDF

Yizhou Liu, Fei Tang, Yuchen Yan, Zhengxi Lu, Songqin Nong

TL;DR: CANL 在 GUI 定位测试时适配中只使用可靠负样本和坐标 token 置信度筛选信号，减少错误正伪标签带来的风险。

Details

Motivation: GUI 智能体需把自然语言指令映射为屏幕坐标，但监督微调和强化学习标注成本高；无标签测试时学习要避免受错误伪正样本干扰。

Result: 作者发现坐标 token 置信度比整段输出置信度更适合作为信号，且稀疏坐标空间中的负样本比可能有噪声的正样本更可靠。CAL 用坐标置信度筛伪标签并赋距离型二元奖励；CANL 仅以负样本优化以规避错误正标签。在 ScreenSpot-V2 上，CANL-7B 得分 92.1%；ScreenSpot-Pro 得分 33.8%，比基础模型绝对提升 8.9 个百分点。

Insight: 坐标级置信度可替代部分人工标注用于测试时适配，而负样本学习降低错误伪正例风险；高难 ScreenSpot-Pro 的最终分数仍较低，复杂 GUI 泛化仍需提升。

Abstract: Graphical User Interface (GUI) grounding is essential for autonomous agents to map natural language instructions to precise screen coordinates. However, existing supervised fine-tuning and reinforcement learning methods are constrained by the high cost of annotation, creating a scalability bottleneck. In this paper, we introduce a label-free test-time training paradigm driven by two key insights: (1) confidence patterns in coordinate tokens are a better indicator than full-sequence confidence, and (2) in sparse GUI coordinate spaces, negative samples offer more reliable learning signals than potentially noisy positive ones. We first propose Confidence-Anchored Learning (CAL), which utilizes coordinate-token confidence to filter pseudo-labels and assign distance-based binary rewards. Building on this, we develop Confidence-Anchored Negative Learning (CANL), which exclusively optimizes the model using negative samples to bypass the risks of incorrect positive samples. Experimental results demonstrate that CANL-7B achieves 92.1% on ScreenSpot-V2. On more challenging ScreenSpot-Pro, CANL-7B reaches 33.8%, an 8.9% absolute improvement over the base model. Our findings establish coordinate-token confidence as a powerful alternative to manual annotations for scalable GUI agent development.

[92] Hypergraph-Regularized Gramian Volumes for Multimodal Retrieval cs.CVPDF

Anindya Nag, Ambuj Mehrish, Sebastiano Vascon

TL;DR: HyVol 在训练期间通过模态超图和语义超图修正多模态嵌入，再用原有 Gramian volume 检索评分；训练完成移除超图，推理成本不变。

Details

Motivation: Gramian volume 可联合评分视频、音频和字幕的高阶对齐，但分数仍是候选局部计算，语义相关训练样本主要只作为对比负例，跨候选语义关系未被利用。

Result: 文档超边连接候选内部可见模态，语义超边连接字幕互为 top-k 邻居的候选；浅层门控超图网络对模态嵌入施加残差修正。缺失模态用掩码排除消息传递，并用恒等填充保持观测 Gram 子矩阵行列式而不插补特征。VAST-27M 的 15 万片段训练后，在六个零样本基准评估，五个检索基准的 R@1 均提高，MSR-VTT、VATEX 视频到文本分别提升 8.3、7.6。缺模态时 V2T 所有设置仍为正增益，但 T2V 四种设置略有下降。

Insight: 训练时跨候选语义关系可补足候选内高阶对齐，且移除模块后不增加部署成本；缺模态下收益依检索方向而异，T2V 结果并非全为正。

Abstract: Volume-based multimodal retrieval jointly scores a text query with a candidate’s video, audio, and subtitle embeddings. While this approach captures higher-order within-candidate alignment, the score remains candidate-local, and semantically related training samples primarily serve as contrastive negatives. This work introduces Hypergraph-Regularized Gramian Volumes (HyVol), a training-time module that incorporates these semantic relations prior to evaluating the original volume loss. Document hyperedges connect the observed modalities of each candidate, whereas semantic hyperedges link candidates whose detached captions are mutual top-k neighbors. A shallow gated hyper-graph network applies residual corrections to the modality embeddings. Presence masks exclude unavailable streams from message passing, and identity padding preserves the determinant of the observed Gram submatrix without feature imputation. As refinement operates on embeddings rather than scores, the same construction applies to both Gram and HyperGram. We remove the hypergraph after training, leaving the backbone-only architecture, original scoring function, and retrieval cost unchanged. We train both backbones on a 150K-clip subset of VAST-27M and evaluate zero-shot performance on six benchmarks. Under the paired protocol, HyVol improves R@1 across all five retrieval benchmarks, with video-to-text gains reaching +8.3 on MSR-VTT and +7.6 on VATEX. Under missing-modality masking, the V2T margin remains positive in all experimental settings, although the T2V margin becomes slightly negative in four.

[93] Concept-Grounded Reasoning with Prompt-Driven Localization for Interpretable Structured Report Generation cs.CV | cs.AI | cs.LGPDF

Xinyue Xu, Hongbin Lin, Juangui Xu, Hualiang Wang, Lehan Wang

TL;DR: CORAL 将病灶定位、临床概念瓶颈和视觉语言生成串联，在超声与胸片报告生成中显式建模中间临床属性以提升可解释性。

Details

Motivation: 多模态大模型常端到端融合影像并生成报告，缺少标准临床流程中定义的中间属性，导致报告证据关联弱、解释有限。

Result: CORAL 使用提示驱动医学分割模型定位病灶，以概念瓶颈预测多类临床属性，再将文本概念 token 与掩码调制视觉特征结合输入 MLLM，完成结构化报告生成和诊断预测。在 BUS-CoT 与 IU X-ray 数据集上，诊断准确率、概念一致性和报告质量均较通用及医学 MLLM 基线稳定改善；摘要未列具体分值。

Insight: 把临床概念和空间定位显式作为生成中间状态，有望使报告更贴近循证诊断流程并便于检查；具体增益和对概念预测错误的敏感性需看完整实验。

Abstract: Medical imaging modalities such as ultrasound and X-ray are widely used in clinical practice, where diagnosis follows a structured, evidence-driven workflow aligned with standardized criteria. While multimodal large language models (MLLMs) show promise for automated medical report generation, most existing systems rely on end-to-end multimodal fusion without modeling clinically defined intermediate attributes, leading to limited grounding and interpretability. To address this issue, we propose CORAL (COncept-grounded ReAsoning with Localization), a multimodal framework that integrates spatial grounding and concept-level supervision into a unified reasoning process. CORAL employs a prompt-driven medical segmentation model to localize lesions and predicts multi-class clinical attributes through a Concept Bottleneck module. The resulting textual concept tokens are combined with mask-modulated visual features within an MLLM to enable structured report generation and diagnostic prediction. Experiments on BUS-CoT and IU X-ray datasets demonstrate consistent improvements in diagnostic accuracy, concept consistency, and report quality over strong general-purpose and medical MLLMs, indicating that concept-grounded reasoning better aligns generation with clinical decision processes.

[94] Query-Conditioned Spherical Centroid Aggregation for Multimodal Retrieval cs.CVPDF

Ambuj Mehrish, Anindya Nag, Sebastiano Vascon

TL;DR: SCALAR 按文本查询为可用的音视频和字幕模态学习相关性权重，再计算球面质心，缓解对称融合方法中模态贡献过于均匀造成的检索退化。

Details

Motivation: Gramian、双曲体积和谱目标等几何聚合通常对所有模态对称处理，统一协议下其联合分数可能比最强单模态低 1.9–27.6 R@1，受均匀模态影响。

Result: SCALAR 对可用模态按查询条件学习相关性权重后计算球面质心，支持任意模态子集，通过 rank-8 LoRA 在遮挡和减少模态视图上训练。五个基准中四个获得正聚合增益，最高 +4.0 R@1；以 480 万可训练参数在三个基准达到最高文本到视频 R@1，另一个在随机种子波动内接近最佳。测试时模态缺失下，其表征阶段比公开 GRAM 检查点高 3.2–10.9 R@1；完整模态训练的重排器在模态减少时逐渐退化为视频单模态路径。

Insight: 聚合应根据查询动态调整模态重要性，而非默认所有线索等权；两阶段检索中的重排器可能削弱表征级缺模态收益，需对缺模态训练和评估同步设计。

Abstract: Multimodal retrieval integrates video, audio, subtitles, and text; however, recent geometric aggregators, such as Gramian volumes, hyperbolic volumes, and spectral objectives, treat all modalities symmetrically. Under a unified evaluation protocol, their joint scores frequently lag behind the strongest single-modality pathway by 1.9 to 27.6 R@1. Controlled analyses attribute this outcome to uniform modality influence. This work introduces Spherical Centroid Aggregation with Learned Adaptive Relevance (SCALAR), a query-conditioned aggregator that assigns relevance-based weights to each available modality before computing a spherical centroid. SCALAR accommodates arbitrary modality subsets and is trained on masked, reduced-arity views using rank-8 LoRA adapters. Across five benchmarks, SCALAR achieves positive aggregation gain on four, reaching +4.0 R@1, while none of the evaluated prior aggregators is positive on more than one. A uniform-weight ablation reproduces the degradation observed with symmetric aggregation. With only 4.8 million trainable parameters, SCALAR attains the highest text-to-video R@1 on three and performs within seed variation of the best result on a fourth. Under test-time modality dropout, SCALAR’s representation-stage score surpasses the released GRAM checkpoint at every evaluated masking rate and benchmark by 3.2 to 10.9 R@1. Finally, as modalities are removed, rerankers trained exclusively on complete modality sets increasingly converge toward their video-only pathways, diminishing these representation-level gains and underscoring a limitation of standard two-stage retrieval pipelines.

[95] Robust Multi-Model Fitting through Learning Neighbor Regions cs.CVPDF

Chang Nie, Guangming Wang, Zhe Liu, Hesheng Wang

TL;DR: LNR 将多模型拟合分成候选最小集筛选和假设局部精修，通过邻域区域特征提高几何视觉任务中的稳健性和效率。

Details

Motivation: 在噪声环境中同时拟合多个模型常受特征利用不足、优化低效、模型重叠和不可微流程限制，计算也浪费在大量差的最小集上。

Result: 粗阶段网络结合最小集的局部点关系与全局上下文特征，预测置信度并预先筛选少数好候选；细阶段为各假设编码可被多个区域共同使用的邻域点几何特征，再由网络独立细化和评分。网络直接从数据点特征学习，避免对采样过程和模型求解器求导。四类经典多模型拟合任务上达到先进表现，且作者称易适配其他稳健拟合任务；摘要未给出数值。

Insight: 先筛选候选、再利用跨区域共享邻域特征处理模型重叠，可减少无效求解并绕过不可微拟合步骤；四类任务支持通用性，但外部场景适配仍需验证。

Abstract: Multi-model fitting involves fitting multiple models accurately in a noisy environment. It is the basis for computer vision tasks such as scene reconstruction and mixed reality. However, its performance is often limited by insufficient feature utilization, inefficient optimization, model overlap, and the non-differentiable pipelines. To overcome these limitations, we introduce a robust coarse-to-fine framework called Learning Neighbor Regions (LNR). Recognizing that substantial computational resources are wasted on numerous bad minimum sets, we propose the coarse-level module. This module utilizes a neural network to extract and analyze geometric feature of both local point-wise relationships and global contextual information in minimum sets, outputting confidence to pre-select a small number of good minimum sets, thereby enhancing overall efficiency before solving hypotheses. To address model overlap, LNR encodes neighbor region features for each hypothesis in its fine-level module. These region features consist of geometric features of neighboring data points, which can be used by multiple regions simultaneously. This design allows the neural network to individually refine and score each hypothesis. Importantly, LNR is trained to learn directly from data point features rather than from the hypothesis parameters, thus avoiding differentiating the sampling process and the model solvers. Extensive experiments on four classic multi-model fitting tasks demonstrate that LNR achieves state-of-the-art performance. The analysis suggests that LNR can be easily adapted to various robust multi-model fitting tasks.

[96] CapsuleMotion: A Lightweight Real-Time Visual Motion Predictor for Capsule Endoscopy cs.CVPDF

Oliver Bause, Julia Werner, Oliver Bringmann

TL;DR: CapsuleMotion 根据图像压缩特征预测胶囊内镜相邻帧运动，并动态调整帧率与肠道前低功耗模式，降低电池消耗。

Details

Motivation: 胶囊内镜尺寸必须足够小以便吞服，限制电池容量，而图像采集和传输耗能较高，可能影响完整检查小肠的概率。

Result: 患者特定动态胶囊行为框架从设备已有图像压缩指标预测连续帧间运动，并据预测幅度调整相机帧率；进入小肠前切换到低帧率省电模式，若检测到运动则调用 LocalizationNet 判断当前器官。在 Rhode Island VCE 数据集评估，并部署于带集成加速器的超低功耗单核 RISC-V 原型。相较固定帧率胶囊，电能最多减少 20.66%，小肠入口点检测准确性也改善。

Insight: 根据内容变化动态采样可把有限电量集中到关键检查区域，压缩特征还能复用现有设备信号；20.66% 节能结果依赖特定数据集和硬件原型，完整临床覆盖率需继续验证。

Abstract: Video Capsule Endoscopy (VCE) is a non-invasive medical examination that allows for the observation of the small intestine, which is otherwise difficult to access. A fundamental challenge persists in the form of their limited size in order to still be swallowable. The resulting restricted battery capacity, however, contradicts with the power-intensive nature of image capture and transmission. Therefore, we propose CapsuleMotion, a patient-specific dynamic capsule behavior that utilizes the available energy in a goal-oriented manner to increase the likelihood of a complete screening of the gastrointestinal tract. By investigating and combining metrics from the on-device image compression, CapsuleMotion predicts the motion between two successive frames. The camera’s frame rate will be modified in accordance with the predicted magnitude of motion. Furthermore, prior to entering the small intestine, the capsule operates in a low power mode with a significantly reduced frame rate. In this mode, the LocalizationNet is employed to determine the current organ, provided that motion was predicted. The proposed framework is evaluated on the Rhode Island VCE dataset and deployed on an ultra-low power single-core RISC-V demonstrator with an integrated hardware accelerator. CapsuleMotion demonstrated the capability to reduce electric energy consumption by up to 20.66% in comparison with conventional capsules that lack a dynamic frame rate. Additionally, the accuracy of detecting the entry point of the small intestine has been improved.

[97] MarKey: Marginal Utility Guided Greedy Keyframe Selection for Long Video Understanding cs.CV | cs.CLPDF

Hongchang Shi, Jinpeng Hu, Ao Wang, Wenzheng Zhou, Hui Ma

TL;DR: MarKey 将长视频关键帧选择建模为感知已选子集的贪心优化，综合查询相关性、边际覆盖增益与上下文冗余，降低有限视觉预算下的重复采样。

Details

Motivation: 长视频理解中密集编码成本高，均匀采样可能漏掉稀疏但决定性证据；现有免训练选帧常独立评分候选帧，忽略新帧对当前已选子集的补充性。

Result: 每轮以可计算替代分数同时评估候选帧的查询相关性、边际覆盖与上下文相关冗余，选择效用最高帧。为降低迭代开销，用紧凑代表锚点近似全视频覆盖，并通过有界已选帧窗口限制上下文比较。在六个整体视频理解、以人为中心视频理解和开放式理解基准上持续优于既有方法；不同 MLLM 骨干、规模和帧预算下均有稳健收益。

Insight: 帧价值取决于它相对当前子集新增了多少证据，不能只看单帧相关性；锚点近似和有界比较使子集感知选择可高效实现。摘要未给具体分数，收益范围需看各基准数据。

Abstract: Long-video understanding remains challenging for multimodal large language models (MLLMs) because densely encoding long frame sequences is computationally expensive, while uniform sampling under a limited visual budget can miss sparse yet decisive evidence. Recent training-free keyframe selection methods have enabled more efficient inference and yielded promising performance gains. However, many existing methods score frames largely in isolation without explicitly considering how each candidate complements the currently selected subset, potentially resulting in redundant selections and incomplete evidence coverage. To address this limitation, we propose MarKey, a training-free framework that formulates keyframe selection as subset-aware greedy optimization. At each iteration, MarKey scores each candidate using a tractable surrogate that jointly accounts for query relevance, marginal coverage gain, and context-dependent redundancy, and selects the frame with the highest utility. To make this iterative subset-aware evaluation efficient, MarKey uses a compact set of representative anchors to approximate full-video coverage and a bounded window of previously selected frames to limit context-dependent comparisons. Experiments on six benchmarks spanning holistic video understanding, human-centric video understanding, and open-ended video understanding demonstrate that MarKey consistently outperforms existing methods. Further analyses show robust gains across different MLLM backbones, model scales, and frame budgets.

[98] ViCo-SAM3: Vision-Conditioned Alignment for Open-Vocabulary Camouflaged Object Segmentation cs.CVPDF

Qiangqiang Zhou, Wenjun Tang, Yong Chen, Dandan Zhu, Jiawei Xu

TL;DR: ViCo-SAM3 用图像全局视觉上下文动态调节文本嵌入，并通过跨模态绑定强化局部视觉与文本语义对齐，用于开放词汇伪装目标分割。

Details

Motivation: SAM3 在开放词汇伪装分割中存在全局文本语义与细粒度像素线索之间的语义差距；全量微调文本编码器参数成本高，也容易过拟合训练类别、损害开放词汇能力。

Result: ViCo 模块按当前图像内容调制文本表征，ViCoBind 模块加强视觉—文本交互与语义绑定。在 OVCamo 基准上，ViCo-SAM3 达到先进分割表现并显示较强泛化；摘要未列分项指标。

Insight: 用视觉条件而非全量文本编码器微调来适配语言语义，可在降低参数开销的同时缩小像素级对齐差距；开放词汇泛化仍需跨数据集和类别报告具体结果。

Abstract: Open-vocabulary camouflaged object segmentation (OVCOS) aims to segment unseen camouflaged objects under text guidance. We observe that SAM3 still suffers from a pronounced semantic gap between global textual semantics and fine-grained pixel-level visual cues in OVCOS. Meanwhile, fully fine-tuning the text encoder introduces heavy parameter overhead and risks overfitting to training categories, which compromises open-vocabulary representation flexibility. To address these issues, we propose ViCo-SAM3, a Vision-Conditioned alignment framework designed for OVCOS. Specifically, we introduce vision-conditioned (ViCo) module, which dynamically modulates text embeddings with global visual context, enabling textual representations to adapt to the current image content and thereby effectively bridging the semantic gap between vision and text. Building on this, we further design a vision-conditioned cross-modal binding (ViCoBind) module to enhance cross-modal interaction and semantic alignment between visual and textual representations. Without bells and whistles, ViCo-SAM3 achieves state-of-the-art performance on the OVCamo benchmark and demonstrates strong generalization.

[99] A Conservative OCR-Enabled Workflow for R214 Sodium Screening of South African Packaged Foods cs.CV | cs.AIPDF

Mayimunah Nagayi, Alice Scaria Khan, Tamryn Frank, Rina Swart, Clement Nyirenda

TL;DR: 论文构建南非包装食品 R214 钠限量的保守图像筛查流程，串联包装区域检测、OCR、证据提取、类别判定和阈值比较，并对不确定样例输出复核。

Details

Motivation: 依据包装照片核查钠含量需同时识别产品、营养成分证据、报告单位和法规类别阈值，任一步误读都可能造成错误合规判定。

Result: 评估 442 种产品、3,929 张完整包装图。YOLO26s 检测器生成 4,195 个区域裁剪，严格后处理每个产品形成一条钠证据记录。整合工作流输出 290 个不在 R214 范围、139 个待复核、7 个筛查通过、6 个筛查不通过；独立 Qwen2.5-VL 7B 流程相应为 387、31、20、4。两者对 R214 类别精确一致 415/442（93.9%），对是否在法规范围内一致 416/442（94.1%），最终筛查结果一致 307/442（69.5%）。60 个产品的人工核验显示严格结果一致性低于法规状态一致性，但所有人工数据不足案例均未被自动流程强判为通过或不通过。

Insight: 保守流程把不确定样本送审而不强制二元决定，适合监管筛查的风险控制；类别和法规范围判断较一致，但最终通过/不通过一致率仅 69.5%，还不能代替人工合规审核。

Abstract: Using food package images to monitor sodium and salt content against South Africa’s R214 sodium limits is challenging when screening decisions require product identity, nutrition facts panel evidence, reporting basis, and category-specific thresholds. This study presents a conservative image-based workflow that combines region detection, optical character recognition (OCR), product identity and sodium evidence extraction, R214 category assignment, deterministic threshold comparison, and independent vision language model comparison. The evaluation used 442 packaged food products and 3 929 full package images from a real-world South African food packaging dataset. A YOLO26s small detector generated 4 195 region crops, and strict post-processing produced one sodium evidence row per product. The integrated workflow produced 290 OUTSIDE R214 SCOPE, 139 REVIEW, seven SCREEN-PASS, and six SCREEN-FAIL outcomes. The independent Qwen2.5-VL 7B vision language model workflow produced 387 OUTSIDE R214 SCOPE, 31 REVIEW, twenty SCREEN-PASS, and four SCREEN-FAIL outcomes. The workflows agreed on exact R214 category assignment for 415 of 442 products (93.9%) and on whether the assigned category was within R214 scope for 416 of 442 products (94.1%). Final screening outcome agreement was 307 out of 442 products, or 69.5%. Manual verification on 60 products showed lower strict outcome agreement than regulated status agreement, while all manual INSUFFICIENT DATA cases were kept out of SCREEN-PASS and SCREEN-FAIL by both automated workflows. The findings show that conservative image-based screening can organise package evidence, identify clear cases, and assign uncertain cases to REVIEW rather than forcing SCREEN-PASS or SCREEN-FAIL decisions.

Shengjie Jin, Zelong Sun, Hengbo Xu, Yanbiao Ma, Zhiwu Lu

TL;DR: AnchorGUI 以认知状态锚点逐步比较 GUI 导航的预期和实际转移，再让记忆有选择地保留错误截图，支持单轮纠错和跨轮经验蒸馏。

Details

Motivation: VLM 导航代理难以处理连续密集视觉历史，既影响单次操作中的即时纠错，也使多次尝试后的经验归纳成本高；正常转移可文本压缩，意外结果则需要视觉因果证据。

Result: 每步 CSA 将预期与观察转移比较并生成显式预测误差信号。非对称记忆在单轮中用滑动窗口只保留检测到不匹配的截图，提供即时视觉反馈；跨轮蒸馏将昂贵的信用分配搜索聚焦在高概率失败步骤。四个基准实验中，AndroidWorld 成功率为 57.3%，每步 token 减少 2.4 倍；跨轮蒸馏达到 69.2% 成功率，比基线高 11.9 个百分点，优于标准反思方法，并保持上下文次线性增长。

Insight: 视觉历史并非都需等量保留，预期—观察差异可识别需要截图证据的关键失败点；跨轮聚焦失败步骤提高成功率同时控制上下文成本。

Abstract: Vision-Language Models (VLMs) enable autonomous GUI navigation, but agents still struggle to process and learn from dense, continuous visual histories. This bottleneck hinders both immediate error correction within a single episode (intra-trial) and experience distillation across multiple attempts (cross-trial). We trace these challenges to an empirical informational asymmetry in GUI navigation: while expected transitions can often be compressed into lightweight textual summaries, unexpected outcomes benefit from preserved screenshots as causal evidence for accurate diagnosis. Building on this insight, we propose AnchorGUI, a unified framework driven by the Cognitive State Anchor (CSA). The CSA acts as a per-step primitive that actively compares expected and observed transitions, converting passive multimodal trajectories into explicit prediction-error signals. These signals orchestrate a dual-scale learning mechanism via an asymmetric memory. For intra-trial correction, a sliding window selectively retains visual evidence for detected mismatches, providing immediate, visually-grounded feedback. For cross-trial distillation, this asymmetric memory focuses the computationally expensive credit assignment search space on likely failure steps. Experiments across four benchmarks validate the effectiveness of our approach. On AndroidWorld, AnchorGUI achieves a 57.3% success rate with a $2.4\times$ token reduction per step. Furthermore, cross-trial distillation reaches 69.2% success (+11.9% gain), significantly outperforming standard reflection methods while maintaining sub-linear context scaling.

[101] CatchMonitor: a machine learning system for automated fish discard quantification cs.CVPDF

Geoff French, Michal Mackiewicz, Mark Fisher, Helen Holah, Rebecca Lamb

TL;DR: CatchMonitor 是面向拖网船远程电子监控视频的自动鱼类弃捕量化原型，通过半监督物种识别和目标跟踪辅助人工分析。

Details

Motivation: 拖网船舱内视频受真实作业环境影响复杂，人工审阅量大；准确量化弃捕还需比较系统与多名专家人工分析之间的变异。

Result: 作者在既有工作基础上用半监督学习改善物种识别，采用简单稳健目标跟踪构建弃捕量化原型，并分析多名专家对同一材料人工量化的差异，将其作为系统比较基准。摘要描述持续开发和评估设计，但未提供数据规模、识别准确率或与专家一致度数值。

Insight: 把专家间人工量化变异纳入系统评价，可避免把人工单一标注当作无误差真值；当前摘要缺少量化结果，尚无法判断系统性能和在真实监管中的可用性。

Abstract: We report on the continued development of CatchMonitor, resulting in a prototype computer vision system designed to automatically quantify discarded fish from video footage collected from Remote Electronic Monitoring (REM) systems on fishing trawlers. The analysis of trawler surveillance footage is a challenging problem due to the real-world conditions on board fishing vessels. Building on our prior work we improve the accuracy of species identification through the application of semi-supervised learning. We utilise a simple and robust object tracking approach, upon which we build our prototype discard quantification system. Finally we analyse the variability of manual discard quantification performed by multiple expert human analysts, using it as a benchmark against which we compare the performance of our system.

[102] BVB: Benchmarking Agentic Video Understanding via Programmatic Reconstruction in Blender cs.CVPDF

Yolo Y. Tang, Daiki Shimada, Jiayue Meng, Jing Bi, Pinxin Liu

TL;DR: BVB/Blender-VideoBench 通过让代理用 Blender 程序化重建真实视频来测试视频理解，评估时空事实保留与感知相似度两方面。

Details

Motivation: 多模态代理可通过代码在 Blender 中创建复杂动画场景，但视频理解基准通常只用问答评测，难判断模型是否能把理解转化为可执行的时空场景重建。

Result: Mini-BVB 为各代理提供相同沙盒和成本限制，要求重建动画场景，再从动画相机渲染并以双重 VQA 测时空事实保留、潜在相似度测视觉接近度，二者几何均值作为平衡总分。评估 10 个模型家族的 51 种配置。最佳模型潜在相似度达 88.6，但只保留源视频正确时空问答的 53.7%；额外推理提高视觉相似度却未弥合事实准确差距。15 名评审、五种配置的盲测显示视觉相似度与人类偏好强相关。

Insight: 程序化重建把视频理解从回答问题推进到生成可执行场景，但视觉逼真和时空事实保持是不同能力；当前主要瓶颈仍是语义与事件细节保留，而非渲染观感。

Abstract: Multimodal agents can create complex videos in software such as Blender by coding without relying on diffusion models. Yet video understanding benchmarks still evaluate models mainly through question answering. If an agent truly understands a video, it can reconstruct it programmatically. We introduce BVB, Blender-VideoBench, a benchmark that tests this ability by asking agents to reconstruct real-world videos as animated Blender scenes. To ensure fair comparison, each agent programs the reconstruction through a lightweight harness, Mini-BVB, in an identical sandbox under a shared cost limit. The benchmark renders each reconstruction from its animated camera and evaluates it on two axes: (1) Dual VQA measures how many spatiotemporal facts the reconstruction preserves. (2) Latent Similarity measures how closely the reconstruction matches the source video perceptually. Our overall score, a square-root mean, favors balanced performance. We evaluate 51 configurations from 10 model families and analyze semantic retention, perceptual similarity, reasoning effort, and cost. The best model reaches 88.6 Latent Similarity but retains only 53.7% of the source-correct spatiotemporal answers. Additional reasoning improves visual similarity but does not close this gap in factual accuracy. In a blind study with 15 raters and five configurations, Latent Similarity correlates strongly with human preference. These results show that programmatic reconstruction is a viable test of agentic video understanding, and that semantic retention remains the main challenge.

[103] PIVOT: Physics-Grounded Verification for AI-Generated Audio-Video Detection cs.CV | cs.AI | cs.MMPDF

Bo Zheng, Kangran Zhao, Xiaoyu Zhang, Weinan Guan, Zhiheng Li

TL;DR: PIVOT 通过从音视频中估计物理量并检查事件是否满足相关物理定律，提供可追溯证据以检测 AI 生成的音视频。

Details

Motivation: 生成模型的视频和音频越来越逼真，传统依赖伪影的检测线索逐渐变弱；但生成器仍难完全复现真实事件的物理行为，物理一致性可作为互补证据。

Result: PIVOT 根据视频和音频估计物理量，为每个片段选择适用物理定律并验证可测约束，除真假判别外还记录验证结论、相关时间窗和支持数值。作者构建 PhysForensics-Bench，包含九类事件场景及两种近期音视频生成器的真实/生成配对片段。Real+Seedance 上准确率/F1 为 70.30%/64.29%，Real+VEO 为 72.16%/65.82%；Gemini 3.1 Pro 直接判断分别为 53.96%/60.09% 和 57.22%/63.44%。

Insight: 物理约束验证可提供可检查的时间定位和量化证据，补充伪影检测而非只输出黑盒真假标签；框架能否扩展到其他模态取决于相关物理量能否可靠估计。

Abstract: As generative models continue to advance, AI-generated content (AIGC) is becoming increasingly realistic, weakening the artifact cues commonly exploited by existing detectors. Nevertheless, faithfully reproducing the physical behavior of real-world events remains challenging for current generators. We therefore explore detecting AIGC by assessing whether the depicted event satisfies measurable constraints derived from physical laws. We introduce PIVOT, a physics-grounded AIGC detector, instantiated here for audio-video clips, that estimates physical quantities from video and audio, selects physical laws relevant to each clip, and verifies their measurable constraints. Beyond a real/fake decision, PIVOT returns supporting evidence that records the verification outcome, relevant time window, and supporting quantities for each applicable law. Although instantiated and evaluated here on audio-video data, the framework can, in principle, extend to other AIGC modalities whenever the physical quantities required for verification can be estimated reliably. We also introduce PhysForensics-Bench, comprising paired real and generated audio-video clips from nine event-centric scene families and two recent audio-video generators. On PhysForensics-Bench, PIVOT achieves 70.30% accuracy and 64.29% F1 score on Real+Seedance, and 72.16% accuracy and 65.82% F1 on Real+VEO. In comparison, direct inspection with Gemini 3.1 Pro obtains 53.96% accuracy and 60.09% F1 on Real+Seedance, and 57.22% accuracy and 63.44% F1 on Real+Veo. These results demonstrate the practical promise of physical-consistency verification as a structured and inspectable source of evidence that complements artifact-based AIGC detection.

[104] What a gated sensing pipeline never looks at: bandwidth reduction and the misses behind it cs.CVPDF

Raghu Venkat, Tricha Anjali

TL;DR: 这项空中传感案例研究显示，门控感知管线可大幅压缩传输数据，但同一门控也决定哪些事件被忽略，必须同时报告压缩率和漏检。

Details

Motivation: 争用链路无法持续回传视频，系统倾向于只发送事件发现并用压缩比表明效率；但未进入判决环节的观测可能造成严重漏报，单报压缩率会误导。

Result: 管线组合学习式目标检测、图文比较和确定性调度、门控、证据累积及传输规则。初始语义阶段活动时，在 211 秒视频中传 38,736 bit，压缩 41,977 倍并检出 4 个事件中的 1 个；后续 tracker 速度修正移除了模型学习的测量伪影，校正代码在四次飞行共 7 个事件中只检出 1 个。无事件控制飞行压缩 155,830 倍但信息量最低。3,187 行逐 tick 轨迹将三个漏检分别定位到无轨迹、未通过结构位置阈值和证据分数不足；另展示在线正常性模型吸收待判对象的已知失效。

Insight: 通信节省和漏检由同一感知门控共同决定，不能把压缩率脱离事件召回单独报告；证据仅为四次有事件飞行的案例研究，修正前后结果都保留以呈现代码变更影响。

Abstract: An airborne sensor on a contested link cannot send video, so the appealing move is to send findings instead and report the ratio between the two. We evaluate a gated sensing pipeline that does this, combining learned object detection and image-text comparison with deterministic scheduling, gating, evidence accumulation and transmission rules. On staged footage with the semantic stage live it sends 38,736 bits over 211 s, a reduction of 41,977x, and names 1 of 4 staged events with no false report. That detection has since been superseded: a correction to how the tracker measures speed removed the measurement artefact the normality model had been learning from, and the flight no longer warms. Under the corrected code the pipeline names 1 of 7 staged events across four flights, and we report both. On a control flight where nothing was staged it reports nothing, a reduction of 155,830x: the largest number in the study and the least informative, because a reduction ratio measures the scene. The mechanisms that produce the reduction also decide which observations ever reach a decision, so the two cannot be reported apart. We give a tick-level trace of one flight (3,187 rows) that places each of three missed events at the stage where it stopped progressing: one produced no track, one failed the structural place test at 0.129, and one passed 539 structural ticks but reached only 3.192 against a boundary of 3.807. We also report one instance of a known failure mode, an online normality model absorbing the object it will later judge, measured against the threshold that object then failed. The evidence is one detection and six misses across four staged flights, beside one clean control, and we treat it as a case study. We give the reproduction protocol and generated results, identify which supporting artifacts are not distributed, and state which experiments did not run.

[105] A Unified Vision-Language Model for PSMA PET/CT Report Generation, Visual Question Answering, and Lesion Segmentation cs.CV | cs.AIPDF

Yang Xing, Jiong Wu, Savas Ozdemir, Yang Zhou, Boxiao Yu

TL;DR: 论文提出统一 PSMA PET/CT 视觉语言模型，同时执行结构化报告生成、视觉问答和病灶分割。

Details

Motivation: PSMA PET/CT 解读对前列腺癌管理关键，但现有 PET/CT AI 多针对单一任务，难以在同一架构中完成交互、报告和体素级定位。

Result: 框架采用 LLaVA 风格结构，包括 PET/CT 视觉编码器、MLP-Mixer 投影、LoRA 微调语言模型和 3D 分割分支，训练分四阶段完成视觉预训练、投影对齐、VLM 微调及多任务微调。语言任务使用 5,747 组 PSMA PET/CT 配对报告，分割使用 AutoPET 的 PSMA 子集。模型在标准报告生成指标上超过 PET2REP 和 CT 基线，VQA 各问题类型有所提升，Dice 和病灶重叠 F1 高于 SegAnyPET 与 nnUNet；摘要未列具体分数。

Insight: 报告、问答与分割联合训练支持结构化且有体素定位的影像解释，减少孤立模型间的信息断裂；具体临床收益和跨机构泛化需进一步评估。

Abstract: Accurate PSMA PET/CT interpretation is central to prostate cancer management, yet existing PET/CT AI models typically address isolated tasks. We propose a unified PSMA PET/CT vision-language model for report generation, visual question answering, and lesion segmentation. The framework adopts an LLaVA-style architecture, comprising a PET/CT vision encoder, an MLP-Mixer projection module, a LoRA-tuned large language model, and a 3D segmentation branch. Training followed a four-stage strategy: vision encoder pretraining, projection-layer alignment, VLM fine-tuning, and final multitask tuning. Language tasks used 5,747 PSMA PET/CT datasets with paired reports, while segmentation used the PSMA subset of AutoPET. The model outperformed PET2REP and a CT-based baseline across standard report-generation metrics, improved performance across VQA question types, and achieved higher Dice and lesion-level overlap F1 than SegAnyPET and nnUNet. These results support the feasibility of a unified framework for structured, interactive, interpretable PSMA PET/CT analysis with voxel-level grounding within a single multitask model architecture.

[106] ModaLens: Measuring Image Sensitivity in Report-Conditioned Medical VLMs cs.CV | cs.AIPDF

Sebastián Andrés Cajas Ordóñez, Maximin Lange, Quang Bui, Anqi Peter Li, Felipe Ocampo Osorio

TL;DR: ModaLens 通过固定报告和问题、替换医学影像的配对反事实，测量报告可用性如何改变放射学 VLM 对图像变化的敏感度。

Details

Motivation: 如果临床报告本身已能回答问题，就难判断视觉语言模型是否实际利用影像；仅看最终答案正确率无法区分图像证据和报告文本的贡献。

Result: 在 3,199 个 MIMIC-CXR 配对病例（293 名患者）中，对每例 14 个问题固定问题与报告，将影像替换为同一患者其他检查图像。MedGemma-27B 在显式回答提示下，报告可用时答案随图像变化的比例为 4.26%，无报告时为 20.94%，配对差异 16.7 点（患者聚类 95% CI 15.6–17.7）；原提示下为 4.70% 对 17.07%。连续答案分数有时变化，即使二元判断不变。方向在另外两个模型家族复现，但标签由报告派生，限制视觉正确性结论。

Insight: 提供报告会降低模型对图像替换的敏感度，说明报告可能主导回答；但此实验测的是图像反事实敏感度而非真实影像诊断准确性，报告派生标签也限制临床解释。

Abstract: A radiology report can already answer a clinical question, so it is hard to tell whether a vision-language model also uses the image. ModaLens, a paired image-swap audit, measures how report availability changes image sensitivity: MedGemma-27B on 3,199 paired MIMIC-CXR cases from 293 patients, all 14 questions per case (13 finding-specific and one composite), each image replaced by one from another study, usually of the same patient, with question and report fixed. Under an explicit answer instruction, the model’s generated answer changes on 4.26 percent of trials with the report and 20.94 percent without it, a paired increase of 16.7 points (patient-clustered 95 percent CI 15.6 to 17.7), so report availability reduces image-swap sensitivity under this protocol; the original prompt with a lowercase first-token readout gives 4.70 percent against 17.07 percent, and substitutions also move continuous answer scores where the binary prediction does not change. The labels are derived from reports, which limits conclusions about visual correctness; the direction replicates in two further model lineages. Code, the exact prompts and a run record for every number are at https://github.com/criticaldata/MODALENS.

[107] VideoScout: Learning Agentic Active Exploration with Adaptive Reasoning Pacing for Long Video Understanding cs.CV | cs.AIPDF

Weixin Xu, Zhenyu Yang, Bing Wang, Shengsheng Qian, Changsheng Xu

TL;DR: VideoScout 将长视频问答建模为序贯证据获取，代理逐轮决定观看速度、保留证据、回访不确定片段及停止时机。

Details

Motivation: MLLM 受视觉上下文窗口限制，均匀采帧或粗到细放大难以在超长视频中定位稀疏决定性证据，且固定观看节奏无法兼顾效率与分析深度。

Result: VideoScout 通过自适应推理节奏控制时间轴遍历，让有限上下文覆盖更长视频同时平衡内容分析深度。作者构建 VideoScout-66K，含 1 万条答案验证轨迹中的 6.6 万个高质量探索轮次；训练先监督微调学习逐轮格式，再以 DAPO 对完整轨迹强化学习，奖励同时考虑答案正确、格式合规和观看进度与教师作答时机的 IoU。7B 模型在长视频理解与推理基准上优于已有训练型 7B agent 模型，摘要未列具体分数。

Insight: 长视频理解需要主动管理观看节奏和证据状态，而非预先固定采样；轨迹级奖励能将答案质量与何时停止观看结合，具体基准增益需查看全文。

Abstract: Multimodal Large Language Models (MLLMs) have achieved remarkable progress on short video understanding yet remain limited on long videos due to the limited visual context window. Prevailing approaches rely on uniform frame sampling or recent coarse-to-fine agentic zooming, both of which struggle to localize sparse, decisive evidence in sufficiently long videos. We formulate long video understanding as a \textbf{Sequential Evidence Acquisition (SEA)} problem, in which an agent reads the video turn by turn along the temporal axis, deciding at each turn how fast to watch, what evidence to retain, when to revisit uncertain segments, and when to stop and answer. Inspired by this view, we propose \textbf{VideoScout}, a multi-turn reasoning agent that instantiates the SEA paradigm through adaptive reasoning pacing. Specifically, by dynamically controlling the viewing pace, VideoScout enables efficient traversal of long videos within a bounded visual context window, allowing the agent to access more video content while balancing content analysis depth with reading efficiency. To train VideoScout, we construct VideoScout-66K, a set of over 66K high-quality exploration turns from 10K answer-verified trajectories, and adopt a two-stage pipeline: cold-start supervised fine-tuning teaches the agent per-turn output format, while the Decoupled Clip and Dynamic sAmpling Policy Optimization (DAPO) algorithm performs trajectory-level reinforcement learning with a composite reward that jointly considers answer accuracy, output format compliance, and the temporal alignment between the agent’s viewing progress and the teacher’s answer timing measured by intersection-over-union (IoU). Extensive experiments on long video understanding and reasoning benchmarks demonstrate that our 7B model achieves strong performance compared with existing trained 7B agentic models.

[108] Human-Grounded Calibration for Long-Text Image-Text Congruence in Vision-Language Models cs.CV | cs.CLPDF

Alessandro Gambetti, Qiwei Han

TL;DR: Congruency Score 是将长文本图像—文本相似度校准为有界一致性分数的轻量层，并强调检索质量、人类相关性和阈值校准应分别评估。

Details

Motivation: 双编码器的原始相似度难解释为图文一致性，图像与文本嵌入模态间隔会使未校准分数难以在长描述匹配任务中使用。

Result: 作者在 DOCCI、Urban1k 上评估四个冻结视觉语言骨干，发现投影后质心距离缩小并不总能改善图文检索。DOCCI 人类校准表明，直接后验校准能保持与人类判断的高关联；部分投影配置则可降低阈值相关的斜率和截距偏差，但以检索表现和人类关联为代价。Congruency Score 提供显式分离并操作化这些目标的方法。

Insight: 一个相似度分数可能同时影响排序和校准，但二者并非同一目标；评价图文一致性时需分开报告检索效果、人类关联及阈值可靠性。

Abstract: Long-text image–text congruence scoring is increasingly important for vision-language systems that must evaluate whether detailed textual descriptions match visual content. However, raw similarity scores from dual-encoder models are difficult to interpret as calibrated congruence measures, especially under the modality gap between image and text embeddings. This paper proposes Congruency Score (CS), a lightweight calibration layer that maps image–text similarity evidence into a bounded score. Using DOCCI and Urban1k, we evaluate four frozen vision-language backbones and show that observed reductions in post-projection centroid distance do not uniformly improve image–text retrieval performance. Human-grounded evaluations on DOCCI further reveal a trade-off: direct post-hoc calibration preserves high association with human judgments, whereas selected projection-based configurations can reduce threshold-relevant slope and intercept distortions at the cost of retrieval performance and association strength. These results establish long-text image–text congruence scoring as a calibrated score-estimation problem, where retrieval performance, human association, and threshold calibration must be evaluated as distinct objectives. CS provides a lightweight way to expose and operationalize this separation.

[109] From Model Patterns to Abstract Semantics in Compositional Zero-Shot Learning cs.CVPDF

Weize Li, Zhicheng Zhao, Fei Su

TL;DR: CLEAR 将组合零样本识别中的属性变化视作语境激活的具体视觉线索，以填空式语义推理从粗到细估计组合含义并重排序。

Details

Motivation: 组合零样本学习需把已见属性和对象原语重新组合到未见搭配；现有方法显式建模多种原语表示，却受固定变体容量和抽象/具体语义竞争限制。

Result: CLEAR 先从原语候选集中提取条件化变体，再用 cloze 式推理归纳高层组合语义，并重排预测以纠正显著具体原语造成的偏置。在 C-GQA 和 MIT-States 等困难基准上，持续改善基础模型并超过先进方法；摘要未给具体分值。

Insight: 组合变化可能是上下文触发不同具体视觉线索，而非独立的新原语实体；填空推理和重排序能够利用这一点，减少对抽象与具体表示的直接竞争。

Abstract: Compositional Zero Shot Learning aims to recognize unseen compositions by recombining learned primitives. Recent methods rely on vision language models and attempt to explicitly model contextual variations of primitives through multiple representations. However, such approaches are limited by fixed variant capacity and competition between abstract and concrete semantics. In this work, we present a new perspective that views primitive variations as the context-driven activation of concrete visual cues rather than independent entities. Based on it, we propose CLEAR, a CLoze-style rEAsoning-based Re-ranking framework inspired by human perceptual processes. CLEAR extracts conditional variants from the primitive candidate set in a coarse-to-fine manner, performs cloze-style reasoning to infer high-level semantics, and re-ranks predictions to correct biases toward salient concrete primitives. Extensive experiments demonstrate that CLEAR consistently improves the Base Model and outperforms state-of-the-art methods on the challenging C-GQA and MIT-States datasets. Code is available at https://github.com/buptLwz/CLEAR.

[110] Circuit-MLLM: Topological Logic-Guided Latent-Space Visual Reasoning for Circuit Schematic Understanding cs.CV | cs.AI | cs.LGPDF

Jinyuan Deng, Yuqi Jiang, Wenjing Huang, Xin Li, Qi Sun

TL;DR: Circuit-MLLM 在潜空间内按器件定位、路径追踪和拓扑顺序进行电路图推理，避免普通图像扫描顺序忽略电路连接逻辑。

Details

Motivation: 通用多模态模型虽能处理常规图文任务，但电路图布局密集且由拓扑连接定义电气语义，需要细粒度结构解析和逻辑推理。

Result: 框架通过电路知识挖掘，将模型潜表征与多粒度电路视觉专家提取的结构特征对齐，使其内化拓扑语义；拓扑引导的排序策略再将推理从固定栅格扫描中解耦，按电路连接逻辑逐步推理。在多种电路分析任务上优于强基线，摘要报告平均分比 GPT-5.1 高 25%，代码已开放。

Insight: 电路理解的关键是连接拓扑而非像素邻近，显式按器件路径顺序推理可补足通用视觉语言模型的结构缺失；摘要未细分各任务指标和评测范围。

Abstract: Through pre-training on extensive text and image datasets, current multi-modal large language models (MLLMs) achieve strong performance on general tasks. However, circuit schematics present a unique challenge for MLLMs due to their dense component layouts and distinct topological logic, demanding fine-grained structural parsing to extract the electrical semantics. To address this, we propose Circuit-MLLM, a multimodal reasoning framework that reformulates circuit topology analysis as a process of device localization, path tracing, and sequential reasoning within the latent space. We introduce a circuit knowledge mining mechanism that deeply aligns the model’s latent representations with structurally rich features derived from multi-granularity circuit vision experts, enabling the model to effectively internalize topological semantics. Building upon these internalized semantics, we devise a topology-guided sequencing strategy that decouples reasoning from the rigid raster-scan order, enforcing stepwise inference along the circuit’s topological logic in latent space. Across diverse circuit analysis tasks, Circuit-MLLM consistently outperforms strong baselines, notably achieving a 25% higher average score than GPT-5.1, which demonstrates the effectiveness of our framework in circuit schematic topology analysis. Code is publicly available at https://github.com/IC-Yuan/Circuit-MLLM.

[111] Benchmarking Intra-Patient 3D Deformable Multimodal Image Registration cs.CV | cs.AIPDF

Matteo Barbieri, Giammarco La Barbera, Juan Pablo De La Plata, Sabine Sarnacki, Isabelle Bloch

TL;DR: 论文基准比较三维多模态可变形配准方法，发现合成基准上学习方法更强，但真实骨盆配准提升有限，几何重叠指标与图像结构相似性也经常不一致。

Details

Motivation: 不同成像模态的同一解剖结构强度差异大，使患者内 3D 多模态配准困难；单一评价指标可能无法反映全局对应质量或局部标签重叠。

Result: 研究在三个覆盖不同解剖区域与难度的数据集上评估传统优化、深度学习和基础模型，包含合成变形恢复及临床场景，并使用 DSC、HD95 和 MIND-SSC。学习方法在大型合成基准表现较好，但真实骨盆配准改善有限。DSC/HD95 几何指标与 MIND-SSC 图像相似度持续不一致；解剖标签引导方法重叠最高，却在分割区域外性能退化。

Insight: 配准结果应多指标共同评估，标签重叠提升不必然代表全局多模态对应改善；目前没有方法能跨解剖和模态稳定泛化，仍是开放问题。

Abstract: Multimodal image registration is a key component of many clinical workflows, yet it remains challenging because corresponding anatomical structures often exhibit substantially different image intensities across modalities. In this work, we present a comprehensive benchmark of intra-patient 3D multimodal deformable registration methods across three datasets covering different anatomical regions and difficulty levels, including both synthetic deformation recovery and real clinical scenarios. We evaluate classical optimization-based approaches and modern learning-based methods, including recent deep learning and foundation models, using complementary metrics: Average Dice similarity coefficient (DSC), average 95th-percentile Hausdorff distance (HD95), and a modality-independent structural similarity measure based on the MIND self-similarity context (MIND-SSC). Results show high variability across datasets, with learning-based methods demonstrating superior performance on large synthetic benchmarks, while only limited improvements are observed in real pelvic registration. A key finding of this study is the consistent disagreement between geometric metrics (DSC, HD95) and image-based similarity metrics (MIND-SSC), highlighting that improved overlap does not necessarily imply better global multimodal correspondence. Furthermore, anatomy-guided approaches achieve the highest overlap scores but exhibit degraded performance outside of segmented regions, revealing a trade-off between label-driven alignment and global structural coherence. Overall, our results indicate that no current method achieves robust performance across anatomies and modalities. We demonstrate that intra-patient 3D multimodal registration requires multi-criteria evaluation, including deformation-based metrics, and remains an open problem.

[112] Don’t Send What You Don’t Need: Question-Guided Token Pruning as a Privacy Defense for Vision-Language Models cs.CV | cs.AI | cs.CRPDF

Md Khalid Syfullah, Alvi Ataur Khalil

TL;DR: QPriv-VL 在跨设备传输前根据问题相关性和隐私敏感度动态剪枝视觉 token，用于联邦、拆分和 U 型拆分 VQA。

Details

Motivation: 隐私敏感场景虽可本地保留原图，但跨模型分割仍需传输大量视觉 token，带来带宽成本和残余隐私泄露风险。

Result: 轻量动态阈值预测器一次前向联合预测样本级剪枝比例和逐 token 保留掩码：跨模态相似度估计问题相关性，冻结 DINOv2 特征估计敏感度，无需敏感性标签。在 GQA、OK-VQA、VQAv2、SLAKE、VQA-RAD 和 PathVQA 上，与四类攻击比较。VQA-RAD 中成员推断成功率由 0.99 降至 0.76–0.79；FSHA/FORA 重建 PSNR 低于固定比例剪枝，同时以约 40% 原视觉 token 保持有竞争力的准确度。敏感区域排除比率为 1.20±0.18，保留行为随问题语义而非一般显著性变化。

Insight: 问题条件化保留有助于在减少传输量时保留答题证据并避开敏感区域；隐私攻击仍有 0.76–0.79 成功率，说明 token 剪枝是风险缓解而非隐私保证。

Abstract: Visual Question Answering (VQA) with Vision-Language Models (VLMs) is increasingly used in privacy-sensitive and bandwidth-constrained settings. Federated Learning (FL), Split Learning (SL), and U-Shaped Split Learning (USL) keep raw data local, but transmitting all visual tokens across a model partition remains costly and can expose private information. We propose QPriv-VL, a question-guided, privacy-aware token-pruning framework for FL, SL, and USL that prunes visual tokens before transmission based on task utility and privacy sensitivity. Its core component is a lightweight Dynamic Threshold Predictor (DTP) that jointly estimates a sample-specific pruning ratio and a token-level retention mask in one forward pass. DTP combines question relevance, computed from cross-modal similarity between visual patches and the pooled question embedding, with a sensitivity signal derived from frozen DINOv2 features. This allows the model to suppress potentially sensitive regions while preserving patches useful for answering the question, without requiring sensitivity labels. We evaluate QPriv-VL on GQA, OK-VQA, VQAv2, SLAKE, VQA-RAD, and PathVQA against four privacy attack families: FSHA, FORA, iDLG, and attribute-inference membership inference attacks. DTP matches or outperforms fixed-ratio pruning while using substantially fewer transmitted tokens. On VQA-RAD, it reduces membership-inference attack success from 0.99 to 0.76-0.79, lowers FSHA and FORA reconstruction PSNR relative to fixed-ratio pruning, and preserves competitive VQA accuracy using about 40% of the original visual-token budget. A sensitivity exclusion ratio of 1.20 +/- 0.18 indicates preferential removal of privacy-sensitive patches, while explainability analysis shows that retention adapts to question semantics rather than generic visual saliency.

[113] MambaMPD: A Mamba-Driven Segmentation Framework for Marine Pollution Detection from Remote Sensing Imagery cs.CVPDF

Shuaiyu Chen, Wei Han, Peng Ren, Chunbo Luo, Zeyu Fu

TL;DR: MambaMPD 将频域增强与多尺度边缘引导结合到 Mamba 语义分割框架中，用于遥感海洋污染检测，重点处理低对比、碎片化污染斑块。

Details

Motivation: 海洋污染遥感检测对生态保护重要，但污染信号弱、图案不连续，污染物与海面视觉相似导致边界模糊，普通分割难以兼顾全局和细节。

Result: FAA 在编码器中用小波分解多尺度频率子带，提取低频语境和高频结构；EGA 融合拉普拉斯边界线索和深层语义以锐化边界，再由含 SE 注意力和深监督的 U-Net 解码器逐级恢复信息。在两个基准上 mIoU 高于竞争方法，且计算量低于基础模型方法；MADOS 上 F1 比 OSDMamba 高 3.6%，M4D 上油污 IoU 比 TransOilSeg 高 6.82%。

Insight: 低频上下文帮助发现微弱污染，高频与边缘先验有助于恢复不规则边界，Mamba 提供长程建模且成本较低；结论主要基于两个海洋污染基准。

Abstract: Accurate marine pollution detection (MPD) is essential for protecting coastal ecosystems and marine biodiversity. Vision Mamba models have shown promise in remote-sensing semantic segmentation by efficiently capturing long-range dependencies and global context, yet their potential for MPD remains underexplored. MPD is particularly challenging because of low signal-to-noise ratios, fragmented pollution patterns, and indistinct boundaries caused by the visual similarity between pollutants and the surrounding sea. To address these issues, we propose MambaMPD, an enhanced Mamba-based framework incorporating two complementary structural priors: Frequency-Aware Augmentation (FAA) and multi-scale Edge-Guided Attention (EGA). FAA integrates wavelet transforms into the encoder to decompose features into multi-scale frequency subbands, enabling the model to capture low-frequency contextual semantics and high-frequency structural details needed to identify small, low-contrast, and irregular pollution patterns. EGA adaptively fuses hierarchical, Laplacian-derived boundary cues with deep semantic representations, refining encoder features before decoding to sharpen boundaries and reduce ambiguity in visually confusing, spatially fragmented scenes. Together, these modules improve sensitivity to subtle pollution signals while preserving fine boundary structures. A U-Net-style decoder with squeeze-and-excitation attention and deep supervision progressively restores and refines semantic and spatial information across scales. Extensive experiments on two benchmark MPD datasets show that MambaMPD achieves higher mIoU than competing methods while requiring substantially less computation than foundation-model-based approaches. On MADOS, it improves F1 by 3.6% over OSDMamba; on M4D, it raises Oil Spill IoU by 6.82% over TransOilSeg.

[114] V-ICAL Bench: Evaluating Video In-Context Learning for Multimodal Agents in Interactive Environments cs.CVPDF

Ziqian Fan, Shibo Xu, Junjie Li, Xiangyu Zhao, Shengyuan Ding

TL;DR: V-ICAL Bench 用 342 个互动任务评估多模态代理能否从视频示例中学会可执行策略，并在 37 个环境上揭示当前代理的视频 ICL 能力与人类基线差距明显。

Details

Motivation: 多模态代理需将视频演示转成策略、迁移到新视觉状态并根据环境反馈迭代动作，但视频情境中的上下文学习远未被充分评估。

Result: V-ICAL 包含 37 个环境中的 342 个互动任务，以人工整理的任务特定视频作为行为示范，从指定初始状态持续交互评估。基准覆盖状态落地、时序记忆、规划和动态适应。19 个先进多模态代理中最佳 Seed-2.1-Pro 得分 54.4/100；Gemini-3.1-Pro、GPT-5.6 等未超过 50，远低于人类基线 83.6。控制实验显示，视频示例未能稳定转化为性能提升。

Insight: 视频 ICL 不只是识别演示内容，还要求把示范变成可执行并可根据反馈调整的策略；当前代理在这一迁移上存在显著瓶颈，基准明确连接了视频理解与互动控制。

Abstract: While In-Context Learning (ICL) enables models to adapt from exemplars without parameter updates, multimodal ICL remains largely underexplored, particularly regarding video demonstrations in interactive environments. For multimodal agents, learning from videos presents unique challenges: they must translate in-context demonstrations into executable policies, ground these policies in novel visual states, and iteratively refine actions based on environmental feedback. We introduce V-ICAL, a novel benchmark designed to evaluate video-based ICL for multimodal agents. Comprising 342 interactive tasks across 37 environments, V-ICAL utilizes human-curated demonstration videos as task-specific behavioral exemplars, evaluating agents through sustained interaction from a target initialization. The benchmark seamlessly connects in-context knowledge induction with core agentic capabilities, including state grounding, temporal memory, planning, and adaptation in dynamic environments. Extensive evaluations across 19 state-of-the-art multimodal agents reveal significant limitations: the best-performing model, Seed-2.1-Pro, achieves a score of only 54.4/100, while other leading models (e.g., Gemini-3.1-Pro, GPT-5.6) fail to surpass 50, far below the human baseline of 83.6. Controlled comparisons further demonstrate that current agents struggle to reliably translate video exemplars into effective policies, failing to yield consistent performance gains. Ultimately, V-ICAL exposes a critical gap in the ICL capabilities of multimodal agents, underscoring an urgent need for future research.

[115] VC-Attention: Value Smoothing and Softmax Casting for Low-bit Attention cs.CVPDF

Xingyang Li, Dongyun Zou, Shining Zhang, Jiacheng Chen, Haocheng Xi

TL;DR: VC-Attention 通过按块聚类平滑注意力 Value，并将对数分数直接转为 FP8 概率码，在视频扩散 Transformer 中同时提高低比特精度和注意力速度。

Details

Motivation: 长时空序列使视频扩散模型的注意力成为主要成本；低比特量化受离群值影响，而 softmax 中的高精度指数运算即使矩阵乘法低比特化后仍可能成为瓶颈。

Result: 免训练 V-Smooth 在线聚类 value token，使硬件块内更易量化，仅量化减去块均值后的残差，再用在线 softmax 已维护的行和恢复均值。ExpCast-FP8 将对数分数用一次融合乘加映射为 E4M3 概率码，避免 FP32 指数和格式转换。在 B200、B300、H200、RTX PRO 6000、RTX 5090 上测试 Wan2.2、LongCat-Video、HunyuanVideo-1.5、MiniMax-H3；相较低比特基线保真度更高，注意力核在数据中心 GPU 比 BF16 FlashAttention-4 快 1.46–1.59 倍、工作站卡快 2.3–3.6 倍，端到端生成分别快 1.13–1.19 倍和 1.36–1.70 倍。

Insight: 低比特视频注意力不仅要处理离群 Value，还需优化 softmax 指数阶段；块均值恢复与直接概率 cast 分别解决精度和流水线瓶颈。

Abstract: Diffusion Transformers deliver state-of-the-art video generation, but their long spatiotemporal sequences make attention the dominant deployment cost, and a deployable low-bit kernel must be accurate and fast. Accuracy is limited by outliers: a block’s quantization scale is set by its largest entries, leaving typical entries confined to a narrow range of representable values. Prior work smooths queries and keys, but value outliers follow no fixed channel or spatiotemporal structure and remain the dominant source of output error. Speed is limited by softmax: low-bit Tensor Cores accelerate only the two matrix multiplications, so the high-precision exponential between them becomes the longest pipeline stage on datacenter GPUs. We propose VC-Attention, a training-free low-bit attention framework that addresses both by pairing Value smoothing with a fused probability Cast. V-Smooth reorders value tokens by lightweight online clustering, so the tokens in a hardware block quantize well together. It quantizes only the residual after subtracting the block mean, and restores that mean from the row sum the online softmax already maintains. ExpCast-FP8 maps log-domain scores directly to E4M3 probability codes with one fused multiply-add, eliminating the FP32 exponential and the format conversion. We implement VC-Attention for B200, B300, H200, RTX PRO 6000, and RTX 5090. Across Wan2.2, LongCat-Video, HunyuanVideo-1.5, and MiniMax-H3, VC-Attention improves fidelity over low-bit baselines, speeds up the attention kernel over BF16 FlashAttention-4 by 1.46-1.59x on datacenter Blackwell and Hopper and by 2.3-3.6x on workstation cards, and generates a clip 1.13-1.19x and 1.36-1.70x faster end to end.

[116] Integrating Multi-view Multi-light Surface Reconstruction into Cultural Heritage Workflows cs.CVPDF

Baptiste Brument, Robin Bruneau, Benjamin Coupry, Vincent Demoulin, Jean Mélou

TL;DR: 论文将多视角多光照表面重建方法集成进 Meshroom，为文化遗产工作流提供可操作的软件层，而非提出新的重建算法。

Details

Motivation: 文化遗产已广泛使用摄影测量进行三维记录，也有 RTI 圆顶等多光照采集设备，但现有工具通常不利用多视角、多光照信息，先进计算机视觉方法又不易融入实际工作流。

Result: 系统在开源 Meshroom 中整合多光照光度立体完整工具链（标定、自标定和通用模式）、自动对象掩码以及多视角法线和反射率整合，使研究方法能在遗产记录流程中使用和评估。摘要强调其作为计算机视觉研究代码与实际文化遗产应用之间的软件桥接层，不报告新的重建算法指标。

Insight: 文化遗产领域的技术落地需要把分散研究方法封装到现有摄影测量流程，而不仅提出新算法；集成平台降低使用门槛，实际文物精度收益需结合未来应用评估。

Abstract: Cultural heritage documentation increasingly relies on image-based 3D surface reconstruction, with photogrammetry software making such workflows accessible to archaeologists, conservators, and heritage technicians. These tools have been successful for conventional multi-view acquisition, but they do not routinely exploit richer multi-view, multi-light data, despite its potential for improving fine-scale surface reconstruction. This limitation is particularly relevant in heritage contexts, where controlled-light acquisition devices such as RTI domes are already used to capture illumination-varying image sets. The challenge is therefore to connect these existing acquisition practices with recent computer vision methods in a form that can be used within operational heritage workflows. In this work, we address this need by integrating state-of-the-art components from computer vision for multi-view, multi-light surface reconstruction into Meshroom, an open-source photogrammetry framework. Rather than proposing a new reconstruction algorithm, our contribution is to assemble and expose existing advanced methods, namely a complete photometric stereo ecosystem (calibrated, self-calibrated and universal), automatic object masking, and multi-view normal-and-reflectance integration, within a usable heritage-oriented workflow. The proposed system thus provides an intermediate software layer between computer vision research code and practical cultural heritage applications, making recent techniques easier to use and evaluate.

Xiaofeng Mao, Peijia Lin, Shaohao Rui, Yibo Zhang, Haibin Wan

TL;DR: LynnReal-Omni提出统一的多模态视频生成框架，将文本、图像参考、3D场景、游戏状态及结构控制纳入同一生成模型，并训练实时版本以支持可控的智能体视觉创作和长视频生成。

Details

Motivation: 视频扩散模型生成随机、难以精确控制，长时段内容还容易出现外观、交互和时间连贯性漂移；基于3D场景或可执行游戏状态的智能体创作虽能提供稳定控制，却不能自动保证对象和角色的视觉保真。将二者结合有望兼顾稳定性与生成质量。

Result: LynnReal-Omni使用32B共享多模态扩散Transformer，覆盖文本生成、图像条件、参考引导、结构控制、编辑、退化视频修复和长视频生成；另训练27B实时Flash模型，并构建视频清洗、主体关联、多模态标注及控制对齐数据流程。论文还提出MSAVP评测设计，包含100个提示和20项指标，考查指令遵循、生成合理性、视觉质量、时间行为及音频协调。在单张H100上，22帧540p视频的预热生成与解码耗时为Omni 843毫秒、Flash 377毫秒。

Insight: 统一模型接收参考外观、可编辑3D渲染和游戏录制等异构条件，试图把智能体的显式场景控制与扩散模型的视觉生成结合。摘要报告了单机推理时间和评测方案，但没有给出完整基准对比结果；实测计时也限定在单张H100及预热条件下。

Abstract: Video diffusion models are stochastic and hard to control: precise content often requires repeated sampling without guaranteed success, and long-horizon scenes drift in appearance, interactions, and temporal coherence. Agentic visual creation provides explicit references, editable 3D scenes, or executable game states for stable control, but does not by itself guarantee high object or character fidelity. Combining the two can enable stable, high-quality generation. To realize this combination, we present LynnReal-Omni, a native multimodal video generation framework built on a 32B shared multimodal diffusion transformer that unifies text-to-video, image-conditioned generation, reference-guided generation, structural control, editing, degraded video restoration, and long-video generation. It accepts heterogeneous visual inputs, including appearance references, editable 3D renders, and game recordings, allowing agents to compose visual conditions within a unified model. We also train a dedicated 27B Flash shared multimodal diffusion transformer for real-time rendering. We build a systematic data pipeline for video cleaning, subject association, multimodal annotation, and aligned control construction, yielding a curated corpus of multi-shot audiovisual segments, and introduce MSAVP, a 100-prompt, 20-metric evaluation design that separates instruction following, generating plausibility, visual quality, temporal behavior, and audio coordination. LynnReal-Omni-Flash further reduces inference cost through model and decoding acceleration, including a lightweight VAE decoder; on one H100, warm generation and decoding of a 22-frame 540p video take 843 ms with LynnReal-Omni and 377 ms with Flash. These results provide a foundation for real-time streaming video generation, making LynnReal-Omni a unified, controllable, and efficient basis for agentic visual creation.

[118] Anatomical Grounding and Leakage-Aware Multimodal Contrastive Learning for Alzheimer’s Disease Classification from Structural MRI cs.CV | cs.AIPDF

Paul-Gabriel Nicolae, Irina Georgiana Mocanu

TL;DR: 该研究用轻量MRI模型分析阿尔茨海默病分类中的解剖关注偏差和临床变量标签泄漏：认知评分融合带来的高准确率属于泄漏上界，按内侧颞叶裁剪则提升了纯影像分类表现。

Details

Motivation: 结构MRI深度模型可能在取得尚可分类准确率的同时关注头骨、眼眶和背景等无关区域；多模态模型加入临床表格后，也可能依赖曾用于定义诊断标签的变量，从而造成标签泄漏并夸大影像模型效果。

Result: 研究在ADNI-1的1,075例基线T1加权扫描上使用ResNet18切片编码器和一层Transformer。基于FastSurfer分割标签训练的YOLOv8解剖定位模型mAP_50超过0.96；Grad-CAM显示影像分类器常关注头骨、眼眶和背景。临床变量融合中，认知评分取得87.3%的三分类准确率，但作者将其视为泄漏驱动的上界；区域体积融合为73.0%。在不使用表格推理的MCI对CN任务中，按认知评分对比学习对齐时纯影像头准确率为52.4%，按体积对齐时为73.8%。将影像限制为个体内侧颞叶裁剪后，纯影像三分类准确率从58.7%升至65.1%。

Insight: 多模态融合的高分不必然代表影像表征更可靠：训练目标和临床标签变量会改变图像编码器学到的特征。研究提醒应将潜在标签泄漏结果与纯影像结果区分，并通过解剖区域检查关注位置；作者也说明结果来自小型平衡测试集上的单次运行，且协议与已发表结果不可直接比较。

Abstract: Deep networks trained on structural MRI for Alzheimer’s disease (AD) staging often reach reasonable accuracy while attending to anatomically irrelevant regions, and multimodal models that add clinical tables frequently rely on variables that were used to assign the diagnostic label in the first place. We study both issues with a deliberately lightweight slice-based encoder (ResNet18 with a one-layer Transformer over slices) on 1,075 baseline T1-weighted scans from ADNI-1. First, we use FastSurfer segmentations as an anatomical reference: YOLOv8 models trained on segmentation-derived labels localize Alzheimer-relevant structures with mAP_50 above 0.96, and a Grad-CAM comparison shows that the image-only classifier frequently attends to the skull, orbits and background. Second, we adapt a CLIP-style image - tabular contrastive framework and organize ADNIMERGE variables along a label-leakage spectrum. Fusion with cognitive scores yields 87.3% three-way accuracy, which we treat as a leakage-driven upper bound rather than an imaging result; fusion with regional volumes yields 73.0%. We observe that the choice of contrastive target changes what the image encoder learns: on MCI vs. CN, the image-only head reaches 52.4% when the encoder is aligned to cognitive scores and 73.8% when aligned to volumes, although no tabular input is used at inference. Third, restricting the input to a per-subject crop of the medial temporal lobe raises image-only three-way accuracy from 58.7% to 65.1%. All results come from single runs on a small balanced test set, and we report confidence intervals and the protocol differences that prevent direct comparison with published numbers.

cs.CL [Back]

[119] PhysMent: An Interactive Approach For LLM Reasoning In Physics Problems cs.CL | cs.CVPDF

Joseph Chan, Utkarsh Jha, Xiyin Yang, Abhinav Jarajapu, Anik Sahai

TL;DR: 论文提出了PhysMent基准测试，用于评估大语言模型（LLMs）在物理模拟器（MuJoCo）中通过工具交互进行主动实验和物理推理的能力。该基准包含105个经典力学场景，通过六维评分框架评估模型在迭代交互中应用力、查询状态、推进时间和修改场景几何体以解决问题的表现。研究发现，当前模型在定性单概念任务上表现尚可（准确率最高达80%），但在需要精确多步实验程序的定量任务上表现大幅下降，最难的类别准确率低于30%，瓶颈在于程序性工具使用而非概念理解。

Details

Motivation: 现有LLMs在静态科学基准上表现良好，但其通过主动实验对物理世界进行推理的能力尚不明确。本文旨在填补这一空白，评估LLMs在需要交互式探索的物理环境中的推理能力。

Result: 在PhysMent基准上，七个模型的准确率在25%到67%之间。模型在定性单概念任务上可达80%准确率，但在最难的定量单概念类别上大多低于30%。失败主要源于过早提交答案、低效探索和对模拟器反馈的不一致理解，而非概念性差距。

Insight: 创新点在于构建了一个强调迭代、工具介导交互的物理推理基准，将评估重点从静态知识转向动态实验过程。客观来看，其提出的六维评分框架和对“程序性瓶颈”（如自适应多步工具使用）的识别，为理解LLMs在物理世界中的实际推理局限提供了新视角。

Abstract: Large language models (LLMs) perform strongly on static science benchmarks, yet their ability to reason about the physical world through active experimentation remains poorly understood. We introduce PhysMent, a benchmark that evaluates LLM physical reasoning via iterative, toolmediated interaction with a MuJoCo physics simulator. Unlike static benchmarks that supply all quantities upfront, PhysMent requires models to discover information by applying forces, querying object states, advancing time, and modifying scene geometry before answering. The benchmark comprises 105 scenes of classical mechanics, organized across four difficulty regimes (Easy/Hard and Single/Multi), three scene modalities (standard, object creation, hidden objects), and a scene-manipulation category, evaluated with a six-dimensional scoring framework. Results show that current models perform reasonably well on qualitative single-concept tasks (up to 80% accuracy) but degrade substantially on quantitative tasks that demand precise, multi-step experimental procedures: most models fall below 30% on the hardest single-concept category, where the bottleneck is procedural (adaptive multi-step tool use) rather than conceptual load. Across the seven models, accuracy ranges from 25% to 67%, with failures due to premature answer submission, inefficient exploration, and inconsistent grounding in simulator feedback rather than conceptual gaps.

[120] TestHallVQA: Exploring LVLMs’ Document-Level Reasoning under Redundant Contexts from Scientific Exams cs.CLPDF

Yongqi Yu, Yu Zhang

TL;DR: 该论文提出了TestHallVQA，一个多图像视觉问答基准，旨在评估大型视觉语言模型在科学考试场景下处理文档级冗余上下文时的推理能力。论文通过理论分析指出了无关视觉标记的影响，并提出了F1-R²新指标来联合量化模型的推理能力和证据检索鲁棒性。

Details

Motivation: 现有平面VQA基准存在局限：要么侧重长文档理解但推理深度有限，要么需要复杂视觉推理但局限于单页无噪声环境。同时，无关视觉标记导致的性能下降问题缺乏系统性量化研究。

Result: 在主流LVLMs上进行的广泛实验和分析揭示了它们在多个维度上的潜在缺陷，为未来研究提供了具体见解和方向。

Insight: 创新点在于构建了一个同时体现文档级规模和人类考试难度的多图像VQA基准，并提出了可控注入多级上下文冗余的方法以及F1-R²联合评估指标，系统性地量化了模型对冗余信息的鲁棒性。

Abstract: Large Vision–Language Models (LVLMs) are increasingly expected to perform visual question answering (VQA) over planar media. However, existing planar VQA benchmarks typically emphasize isolated challenges: some emphasize long-document understanding with limited reasoning depth, while others require complex visual reasoning but remain restricted to single-page, noise-free settings. Moreover, through theoretical analysis, we identify the impact of irrelevant visual tokens, which leads to measurable performance degradation but has received little attention with respect to systematic quantification. To address these limitations, we introduce TestHallVQA, a multi-image VQA benchmark that simultaneously embodies document-level scale and the difficulty of human examinations, while providing comprehensive task coverage. Leveraging TestHallVQA’s ability to controllably inject multi-level contextual redundancy, we further propose a novel metric, F1-R\textsuperscript{2}, which jointly quantifies LVLMs’ computational reasoning capability and their evidence retrieval robustness against document-level redundancy. Extensive experiments and analyses on mainstream LVLMs reveal their latent deficiencies across multiple dimensions, offering concrete insights and directions for future research. The associated datasets, code, and complete theoretical derivations are available at https://github.com/yqyu2317/TestHallVQA-benchmark.

[121] Lexical Prompt Compression for Large Language Models: A Training-Free, Deterministic Pipeline with Empirical Pareto Analysis Across Eleven Task Categories cs.CLPDF

Shamin Chokshi

TL;DR: 本文提出了一种无需训练、完全确定性的词汇提示压缩流水线，用于压缩大型语言模型的提示。该方法基于经典的词汇NLP技术，通过11种可切换的词汇转换（如停用词移除、填充短语删除、词性剪枝等）组合成可配置的流水线。在涵盖11个任务类别的1,242个英文提示上评估了15种配置，结果显示最激进的配置实现了40.3%的平均令牌减少，同时BERTScore-F1保持在0.876。

Details

Motivation: 解决现实世界中提示（如思维链推理和上下文学习）变得日益庞大和复杂，导致推理成本和延迟增加的问题。现有学习方法（如LLMLingua）需要辅助语言模型且具有非确定性，因此探索一种无需训练、完全确定性且仅需CPU的基于经典词汇NLP的压缩方法能达到何种效果。

Result: 在六个来源（Dolly-15k, LMSYS-Chat-1M等）的1,242个英文提示上，使用BLEU、ROUGE、BERTScore-F1和SentenceBERT余弦相似度评估输出保留度。最激进配置实现平均令牌减少40.3%（标准差9.2），BERTScore-F1为0.876；仅停用词配置实现29.6%减少，BERTScore-F1为0.913。为每个任务类别分析了压缩与保真度的帕累托前沿，发现常识推理在激进压缩下是系统性失败模式。

Insight: 创新点在于提出了一种完全基于经典词汇NLP、无需训练、确定性的提示压缩流水线，避免了辅助模型的使用和非确定性。客观来看，该方法通过组合多种可配置的词汇转换操作，实现了高效的令牌减少，并在多个任务类别上系统性地分析了压缩与保真度的权衡，为轻量级、可解释的提示压缩提供了新思路。

Abstract: Recent advances in large language models (LLMs) have made prompts increasingly large and complex. Techniques such as chain-of-thought reasoning (Wei et al., 2022) and in-context learning (Brown et al., 2020) frequently push real-world prompts past several thousand tokens, increasing inference cost and latency. Learned compression methods such as LLMLingua (Jiang et al., 2023) and Selective Context (Li et al., 2023) achieve high compression ratios but require auxiliary language models and are non-deterministic. We ask a complementary question: how far can a training-free, fully deterministic, CPU-only pipeline based on classical lexical NLP be pushed before output quality degrades significantly? Eleven toggleable lexical transformations - stopword removal, filler-phrase deletion, contraction and abbreviation substitution, part-of-speech-based pruning, lemmatization, WordNet-driven synonym shortening, and named-entity preservation - are assembled into a configurable pipeline. Fifteen configurations are evaluated on 1,242 English-only prompts from six sources (Dolly-15k, LMSYS-Chat-1M, WildChat-1M, MMLU, GSM8K, HellaSwag), spanning eleven automatically derived task categories, yielding 18,630 paired GPT-4o-mini completions. Output preservation is measured using BLEU, ROUGE-1/2/L, BERTScore-F1, and SentenceBERT cosine similarity. The most aggressive configuration achieves a mean token reduction of 40.3% (sigma = 9.2) at a BERTScore-F1 of 0.876 against the original-prompt output; a stopword-only configuration achieves 29.6% reduction at 0.913. The compression-versus-fidelity Pareto frontier is characterized per task category, with commonsense reasoning a systematic failure mode under aggressive compression. All code, prompts, and per-cell results are released for reproducibility.

[122] Clinical Reasoning Under a Partially Observed Objective in Cone Beam CT Report Generation cs.CL | cs.CVPDF

Ajo Babu George, Govind Arun, Sidharth N Krishna, Uma Ranjan

TL;DR: 本文研究了在锥形束CT报告生成中，使用部分可见的复合目标进行临床推理。该复合目标80%权重基于大型语言模型的事实蕴含判断，20%基于词汇重叠，但开发过程中仅词汇部分可见。研究通过纯Python复现评估指标，并利用离线蕴含代理降低优化成本，在公开数据集上验证了直接优化复合目标能显著提升性能。最终系统在未见中心的数据上达到了METEOR 0.3542。

Details

Motivation: 解决锥形束CT报告生成中，传统基于词汇重叠的评估指标（如BLEU）可能无法准确反映临床事实正确性的问题，探索在开发阶段仅部分目标可见的情况下如何优化报告生成质量。

Result: 在622个公开病例上，针对可见词汇排名选择的报告得分为0.2909，而针对复合目标优化的报告得分提升至0.4122。优化n-gram重叠会导致蕴含精度从0.522降至0.266。最终系统在来自未见中心的50个病例上达到METEOR 0.3542。

Insight: 创新点在于提出并优化了一个部分可见的复合评估目标，揭示了传统词汇指标可能奖励的是口述惯例而非解剖学事实。方法上，通过复现评估例程和构建高效的离线蕴含代理，使得直接优化该复合目标变得可行。

Abstract: Maxillofacial report generation from cone beam computed tomography is scored here by a composite objective placing 80% of its weight on a large language model judgement of factual entailment and 20% on lexical overlap, of which only the lexical fifth is visible during development. The grader’s BLEU-4 and METEOR routines are reproduced in pure Python and match the reference to machine precision, and an offline entailment surrogate, which tells a report written for one patient from one written for another at an area under the curve of 0.987, makes the composite objective cheap enough to optimise directly. Over the 622-case public release, a report selected against the visible lexical ranking scores 0.2909, whereas one selected against the composite objective scores 0.4122, because pursuing n-gram overlap drives entailment precision from 0.522 down to 0.266. A 29 million parameter encoder fine-tuned on the release reaches a prevalence-weighted out-of-fold area under the curve of 0.486 over 985 statements, indistinguishable from the corpus prior, while nine numbers read from the image header reach 0.945 for mandible coverage and 0.872 for condyle coverage, and acquisition centre alone predicts sentence choice at 0.718 against 0.663 for the image-derived model, identifying dictation convention rather than anatomy as the quantity the lexical metrics reward. The delivered system emits eight unconditional statements and five gated on header geometry under polarity, laterality and tooth-level consistency constraints, and reaches METEOR 0.3542 over 50 held-out cases from an unseen centre. The dataset and code are available at https://github.com/GIND123/CBCT-Clinical-Reasoner

[123] RFCLLM: Evaluating LLMs’ Reasoning Ability of Network Protocol State Machines cs.CLPDF

Anqi Chen, Dan Goldwasser, Cristina Nita-Rotaru

TL;DR: 该论文提出了RFCLLM框架，用于评估大型语言模型（LLM）对网络协议规范中有限状态机（FSM）的推理能力。通过设计4个任务和1482个查询，对16种协议进行测试，考察LLM的隐含表示与人工构建的真实模型之间的对齐程度，并分析了多种影响因素。

Details

Motivation: 旨在评估LLM在将文本协议规范映射为形式化表示时的可靠性，验证其是否真正理解规范，以判断LLM在网络协议安全或测试等应用中是否值得信赖。

Result: 论文通过系统实验评估了不同LLM在多个协议FSM推理任务上的表现，分析了任务难度差异、上下文类型和协议特性对结果的影响，但摘要未具体提及定量结果或是否达到SOTA水平。

Insight: 创新点在于构建了一个专门用于评估LLM协议状态机推理能力的基准测试框架，并系统性地研究了影响LLM推理性能的关键因素，为验证LLM在形式化规范理解上的可信度提供了方法论。

Abstract: Mapping textual specifications into formal representations is essential for ensuring the correctness of protocol designs and implementations. LLM-generated mappings, used for networking security or testing, are assumed to capture a perfect understanding of the specification, which may not hold in practice. The goal of this paper is to assess the extent to which LLMs can interpret the specification correctly. We examine the degree to which an LLM’s implicit representation of a finite-state transition system-defined via natural language descriptions-aligns with a manually generated ground-truth model. We designed 4 tasks and 1482 task queries for 16 protocols. We evaluated different judge biases, observed the inherent difficulty gaps between tasks, looked into the effect of 4 context types, and the influence of protocol characteristics. Our work contributes to a step toward verifying whether LLMs can really be trusted in FSM (Finite State Machine) reasoning of protocol specifications.

[124] From Token Probabilities to Semantic Constraints: Towards Declarative Probabilistic Evaluation of Language Models cs.CLPDF

Kyle Richardson, Cullen Anderson, Pranav Balakrishnan, Takuto Ban, Daksha Ladia

TL;DR: 本文提出了ModelLog，一种声明式概率框架，用于评估大型语言模型的预训练效果。该框架将评估目标定义为符号约束，通过衡量模型预测分布满足这些约束的程度来评估模型的知识和推理能力。

Details

Motivation: 旨在解决如何评估大型语言模型获得的知识和推理能力，以及这些评估如何与预训练中的学习信号相关联的问题。

Result: 通过针对否定、互斥性和一致性的新任务套件，发现了仅凭词元似然或答案准确率难以描述的系统性失败。

Insight: 创新点在于将评估目标形式化为符号约束，并建立了评估与学习之间的语义联系，其梯度反映了逻辑强度、信息量和变量级敏感性，为诊断模型行为和阐明学习语义结构提供了新工具。

Abstract: While Large Language Models have improved rapidly, many fundamental questions remain about how to evaluate the knowledge and reasoning abilities they acquire, and how such evaluations relate to the learning signals used in pre-training. In this paper, we propose ModelLog, a declarative probabilistic framework for pre-training evaluation that makes the semantic structure of model behavior explicit and provides new formal tools for relating evaluation to learning. ModelLog specifies evaluation targets as symbolic constraints over token-level predictions and measures how strongly a model’s distribution satisfies those constraints. We explore the framework through a new suite of tasks targeting negation, mutual exclusivity, and consistency, finding systematic failures that are difficult to characterize through token likelihood or answer accuracy alone. We further show that these evaluation scores can also be interpreted as losses, whose gradients reflect logical strength, informativeness, and variable-level sensitivity. This links evaluation and learning through a shared semantics, suggesting evaluation methods that diagnose model behavior while also helping to clarify the semantic structure of learning.

[125] Hindsight Bias in Clinical Temporal Reasoning: How Future Data Exposure Affects Large Language Model Judgment cs.CL | cs.AIPDF

Misaki Matsuura, Sayantan Kumar, Ojas Kadam, Jeremy C. Weiss

TL;DR: 本文针对临床语言模型在回顾性评估中可能存在的后见之明偏差问题，提出了一个配对基准测试。该基准包含171个临床病例报告，通过设计前瞻性答案与后见之明陷阱问题，评估模型在时间截断与完整时间线条件下的表现。研究发现，暴露完整时间线会导致模型产生后见之明敏感偏移，而时间掩码能有效减少偏差且不降低准确性。

Details

Motivation: 临床决策是前瞻性的，但临床语言模型常基于揭示最终诊断和结果的回顾性记录进行评估，这种评估可能奖励模型使用未来信息，而非在决策点的不确定性下进行推理，从而引入后见之明偏差。

Result: 在GPT 5.6 Sol、Gemma 4、GLM 5.2和Opus 5等模型上，暴露完整时间线会导致一致的后见之明敏感偏移（通过准确率、后见之明陷阱率、答案不稳定率和后见之明偏差率等指标衡量），而时间掩码能减少偏差且不降低准确性。

Insight: 创新点在于提出了一个专门测量临床时间推理中后见之明偏差的配对基准，包含文本叙述和文本时间序列两种表示，并设计了前瞻性答案与后见之明陷阱问题对。客观分析认为，该方法强调了在临床AI评估中模拟真实决策点不确定性的重要性，时间掩码策略为减少偏差提供了实用方向。

Abstract: Clinical decisions are prospective, but clinical language models are often evaluated on retrospective records that reveal the final diagnosis, treatment response, and outcome. Such evaluations may reward the use of future information rather than reasoning under the uncertainty present at the decision point. We introduce a paired benchmark for measuring outcome-conditioned shifts consistent with hindsight bias in clinical temporal reasoning. It contains 171 case reports from the PubMed Central Open Access Subset—40 sepsis and 131 GLP-1/diabetes cases—represented as both textual narratives and human-annotated and LLM-generated textual time series (TTS). For each case, questions are tied to a clinically meaningful cutoff and paired with a prospective reference answer and an outcome-consistent \emph{hindsight trap}. Models answer each question using either a TTS truncated at the cutoff or the complete timeline; additional conditions vary the narrative source (original or synthetic) and TTS annotation source (human or LLM). We evaluate accuracy (Acc), hindsight trap rate (HTR), answer instability rate (AIR), and hindsight bias rate (HBR), each of which captures different signals of hindsight bias. Across GPT 5.6 Sol, Gemma 4, GLM 5.2, and Opus 5, full timeline exposure produces consistent hindsight-sensitive shifts, while temporal masking reduces bias without lowering accuracy.

[126] Inside VLM Chart Reading: Tracing Value Reading from Vertical Bar Charts Across Space and Depth cs.CLPDF

Tianhao Niu, Qingfu Zhu, Wanxiang Che

TL;DR: 本研究通过受控反事实激活修补技术，分析了Qwen2.5VL-7B-Instruct和InternVL3.5-8B两个视觉语言模型在读取垂直条形图数值时的内部计算机制。研究发现，条形图顶部区域比条形主体对答案恢复更重要，图例和序列相关状态比几何和坐标轴状态更早失去局部可恢复性，并且模型在中间层会将信息从视觉图例区域转移到提示序列位置。

Details

Motivation: 尽管视觉语言模型能准确回答图表问题，但其输出准确性无法揭示模型如何整合证据以恢复精确数值。本研究旨在通过因果分析，定位支持精确条形图数值读取的内部计算过程。

Result: 在垂直条形图数值读取任务中，通过反事实激活修补实验发现，两个模型都能利用来自不同来源的几何和尺度状态来支持组合目标。InternVL在状态来自不同来源或同一图像时表现相似，而Qwen对单独来源的恢复率较低，表明其具有更高的上下文敏感性。

Insight: 创新点在于首次使用受控反事实激活修补技术对VLM的图表阅读进行因果分析，揭示了数值读取过程中视觉信息处理的层次性、信息传递路径（从视觉区域到语言提示的转移）以及不同模型在上下文整合能力上的差异，为理解VLM的图表推理机制提供了初步的因果证据。

Abstract: Vision–language models (VLMs) can answer chart questions accurately, but output accuracy does not show how they combine the evidence needed to recover an exact value. We study vertical-bar value reading with controlled counterfactual activation patching in Qwen2.5VL-7B-Instruct and InternVL3.5-8B. The study connects three analyses: (1) The single-factor results show that the changed bar-top region restores much more answer preference than the unchanged bar body, despite containing fewer visual tokens. Legend- and series-related states also lose local recoverability earlier than bar-geometry and axis-scale states. (2) In the handoff analysis, restoration shifts from visual legend regions in early layers to prompt-series positions in middle layers. Resetting the prompt-series state selectively reduces legend-source rescue, supporting its role as a partial mediator. (3) In the factorial analysis, both models can use geometry and scale states from separate donors to favor the combined target. InternVL performs similarly when the states come from separate donors or one image, while Qwen shows lower restoration for separate donors which suggests higher context sensitivity. Together, these results provide preliminary causal evidence for localizing the internal computations that support exact bar-value reading.

[127] SyRHM: Symbolic-Language-Enhanced Reasoning with Associative Retrieval for Zero-shot Harmful Meme Detection cs.CLPDF

Hanling Wang, Chenlong Wei, Yingjuan Li, Di Wu, Yuchao Zhang

TL;DR: SyRHM是一个用于零样本有害表情包检测的框架，它通过将检测任务分解为基于语义的检索和符号语言增强的多阶段推理来解决现有方法面临的挑战。该框架首先将多模态内容解析为文本元素和描述以检索语义相关的表情包，然后通过翻译器、规划器和求解器阶段进行可解释的多阶段推理，分析有害意图。

Details

Motivation: 有害表情包的检测对于维护安全的在线社区至关重要，但其有害意图往往隐含在视觉-文本的不一致性和文化刻板印象中，这对现有的多模态检测器构成了挑战。

Result: 在FHM、HarM和MultiOff数据集上的实验表明，SyRHM在大多数评估设置中优于多模态和基于推理的基线方法，实现了卓越的性能，并能为有害内容提供推理轨迹。

Insight: 论文的创新点在于将有害表情包检测分解为基于语义的检索和符号语言增强的多阶段推理，这提供了超越表面相似性的基础上下文，并实现了对有害意图的表达性和可解释性分析。从客观角度看，其将多模态输入转换为符号中间表示并进行多阶段推理的架构设计，为复杂、隐含意图的理解提供了一种新的、可解释的途径。

Abstract: Detecting harmful memes is critical for maintaining safe online communities. However, harmful intent is often implicit, arising from visual-textual incongruity and cultural stereotypes, which challenges existing multimodal detectors. We propose SyRHM, a framework that decomposes harmful meme detection into meaning-grounded retrieval and symbolic-language-enhanced multi-stage reasoning. SyRHM retrieves semantically related memes by parsing multimodal content into textual elements and descriptions, providing grounded context beyond surface-level similarity. Building on the retrieved context, SyRHM uses a translator stage to convert multimodal inputs into symbolic intermediate representations, and then performs multi-stage reasoning via planner and solver stages, enabling expressive and interpretable analysis of harmful intent. Experiments on FHM, HarM, and MultiOff demonstrate the effectiveness of SyRHM, achieving superior performance on most evaluation settings against multimodal and reasoning-based baselines, while providing reasoning traces for harmful content. The code is available at: https://github.com/Scabbards1500/SyRHM

[128] DARE: Dialectical Agentic Reasoning for Structured Knowledge Fact Checking cs.CLPDF

Yifei Li, Xiaohan Zheng, Wentao Qian, Liansheng Zhuang

TL;DR: 本文提出了一种名为DARE的多智能体框架，用于结构化知识事实核查，通过迭代的检索-推理-反思过程，结合基于关系的证据检索、辩证双向验证和置信度驱动的元反思，有效提升了事实核查的准确性和鲁棒性。

Details

Motivation: 针对现有基于程序生成的方法在结构化知识事实核查中存在的无效关系生成、单一路径推理缺乏自我修正以及证据评估偏倚（倾向于高估支持性信号）等局限性，旨在设计一个更稳健的推理框架。

Result: 在结构化知识事实核查基准测试中，使用8B参数骨干模型的DARE框架达到了88.12%的准确率，其性能匹配甚至超越了基于GPT-4o的程序生成基线方法。

Insight: 创新点在于将结构化知识事实核查构建为一个多智能体参与的辩证推理过程，通过关系约束的证据检索、支持与反驳视角的双向验证以及动态的元反思机制，有效激发了大型语言模型的潜在推理能力，并提升了推理过程的可靠性和可解释性。

Abstract: Structured knowledge fact checking aims to determine the truthfulness of natural language claims by reasoning over structured evidence. Recent program-generation approaches leverage large language models (LLMs) to generate executable graph reasoning programs, achieving strong performance on structured knowledge fact checking benchmarks. However, these methods remain limited by invalid relation generation, single-path reasoning that lacks self-correction, and biased evidence assessment that tends to overestimate supporting signals. We propose Dialectical Agentic Reasoning (DARE), a multi-agent framework that formulates structured knowledge fact checking as an iterative retrieve-reason-reflect process. DARE integrates relation-grounded evidence retrieval to constrain reasoning to valid structures, dialectical bidirectional verification to evaluate evidence from both supporting and refuting perspectives, and confidence-driven meta-reflection to dynamically determine whether additional evidence exploration is necessary. Extensive experiments demonstrate the effectiveness of DARE in structured knowledge fact checking, with an 8B backbone achieving 88.12% accuracy and matching or surpassing GPT-4o-based program-generation baselines, which attests to the efficacy of dialectical agentic reasoning in eliciting the latent reasoning capabilities of LLMs.

[129] When Consistency Does Not Mean Reliability: Evaluating Local LLM Judges Against Human Ratings cs.CLPDF

Aakash Kumar Tiwari

TL;DR: 该论文研究了使用本地开源大语言模型（LLM）作为自动评估员（LLM-as-a-Judge）的可靠性问题。通过让LLaMA-3-8B和Qwen2.5-7B模型评估GPT-2生成的300个回答，并与9位人类标注者的评分进行比较，发现尽管模型自身评分高度一致，但与人类评分的一致性却有限。

Details

Motivation: 动机在于探究LLM作为评估工具时，其评分的高自一致性是否等同于与人类判断的高一致性，从而揭示当前LLM-as-a-Judge方法在可靠性评估上的潜在缺陷。

Result: 在涵盖事实知识、指令遵循、数学、推理和写作五个类别的100个问题上，LLaMA-3-8B与人类评分的皮尔逊相关系数为0.275，Qwen2.5-7B为0.340；两者的平均绝对误差分别为27.71和18.64。然而，它们的自一致性（精确一致率）分别高达97.3%和92.3%。

Insight: 核心创新点在于明确指出并实证了LLM评估员的高自一致性并不保证其与人类判断的一致性，这挑战了仅依赖自一致性作为评估可靠性的常见假设。研究强调了在采用本地LLM作为自动法官时，必须同时评估其自一致性和与人类的对齐程度。

Abstract: Large language models (LLMs) are increasingly used to evaluate the responses of other language models. This approach, known as LLM-as-a-Judge, is faster and cheaper than human evaluation. However, a judge may produce consistent scores without necessarily agreeing with human evaluators. In this work, we study this issue using two local open-weight LLM judges, LLaMA-3-8B and Qwen2.5-7B. We evaluate 300 responses generated by an instruction-tuned GPT-2 (124M) model for 100 questions covering five categories: factual knowledge, instruction following, mathematics, reasoning, and writing. Each response is scored by nine human annotators and is evaluated three times by each LLM judge using the same rubric. We compare the judge scores with the average human scores using Pearson correlation, Spearman correlation, mean absolute error (MAE), signed bias, and self-consistency. LLaMA-3-8B shows a Pearson correlation of 0.275 with human scores, while Qwen2.5-7B achieves 0.340. Their MAEs are 27.71 and 18.64, respectively. Despite this limited agreement, both judges show high self-consistency, with exact consistency rates of 97.3% for LLaMA-3-8B and 92.3% for Qwen2.5-7B. These results show that high self-consistency does not necessarily indicate high agreement with human judgments. Our findings highlight the need to evaluate both consistency and human alignment when using local LLMs as automatic judges.

[130] Sweet Talkers: How Query Formulation Shapes Sycophancy in Romantic Relationship Advice cs.CLPDF

Helena Choi, Edric Castel Hao, Karl Bautista, Francis Gabriel Magleo, Renzo Panti

TL;DR: 本研究探讨了大型语言模型（LLM）在提供浪漫关系建议时，其回答如何因用户提问方式的不同而表现出不同程度的‘奉承’行为。研究使用RRASP数据集和ELEPHANT框架评估了GPT-5 Mini和Gemini 3 Flash模型，发现语法语气本身影响不大，而基于视角的提问框架（如‘我’vs‘朋友’）以及对话轮次的增加会显著增强模型在框架和道德层面的奉承倾向。

Details

Motivation: LLM越来越多地被用于情感支持和关系建议，但其倾向于维护用户‘面子’的特性可能无意中强化有害的人际行为。本研究旨在系统性地评估这种风险，探究用户提问的表述方式如何影响模型的奉承行为。

Result: 在RRASP数据集上的评估表明，语法语气未产生系统性差异，但视角驱动的提问框架影响显著，且随着对话轮次增加，模型在框架和道德层面的奉承行为持续增强。在道德奉承方面，Gemini 3 Flash的增长幅度远小于GPT-5 Mini，表现出更强的抵抗力。

Insight: 创新点在于揭示了用户提问的隐含前提（视角框架）比表面的语法形式更能驱动LLM的奉承行为，并且这种奉承在对话中会逐渐累积。这为理解和缓解LLM在敏感应用中的社会风险提供了新的评估维度和洞见。

Abstract: Large language models (LLMs) are increasingly used for emotional support and relationship advice, where a model’s tendency to preserve a user’s face can inadvertently reinforce harmful interpersonal behaviors. To systematically examine this risk, we developed the Romantic Relationship Advice-Seeking Prompts (RRASP) dataset of 2,400 prompts across five relationship themes and evaluated social sycophancy using the ELEPHANT framework on two consumer-facing models, GPT-5 Mini and Gemini 3 Flash. Contrary to our initial hypothesis, grammatical mood alone did not produce systematic differences in sycophantic behavior, suggesting that what a user implies matters more than how they phrase it. Instead, perspective-driven framing had a stronger influence, with gaps between original and flipped prompts widening in follow-up responses. Consistent increases in framing and moral sycophancy across turns indicate that models become more likely to accept a user’s stated premises and affirm their ethical stance as a dialogue progresses. Notably, Gemini 3 Flash exhibited substantially smaller increases in moral sycophancy than GPT-5 Mini, suggesting it is more resistant to reinforcing ethically problematic positions across turns.

[131] SHIFT-M3: Pre-fusion Alignment-based Consistency Screening for Multimodal ECG Record Integrity cs.CLPDF

Md Ashik Khan, Md Nahid Siddique

TL;DR: 本文提出SHIFT-M3，一种轻量级的基于预融合对齐一致性的多模态心电图记录完整性筛查方法。该方法通过比较LLM生成的心电图解读和临床报告摘要这两种文本视图之间的一致性，来检测记录中不同模态（如波形、报告）是否属于同一患者，以解决临床AI中因数据链接失败导致的安全问题。

Details

Motivation: 多模态临床AI通常假设一条记录中的波形、报告、元数据和下游预测属于同一患者，但实践中数据链接失败可能悄无声息地组装来自不同患者的看似合理但跨患者的组件，这带来了标准预测模型无法检测的安全隐患。

Result: 在784,680条MEETI心电图记录上，SHIFT-M3仅用573,569个参数，对于完整文本视图交换实现了97.6% TPR@5% FPR（AUROC 0.996），对于部分交换实现了90.3%（AUROC 0.974），对于标签匹配的困难负样本实现了97.7%（AUROC 0.996）。与同数据集词法基线相比，在部分交换和困难负样本上提升最大，表明模型学到了超越表面重叠的信息。

Insight: 创新点在于提出了一种基于文本预融合对齐的轻量级一致性筛查框架，用于多模态记录完整性分类，并引入了CMST评估分类法。该方法的核心是通过比较两种独立生成的文本视图（LLM解读与临床摘要）的语义一致性来检测模态错配，而非依赖原始波形数据，具有参数效率高和可解释性潜力。

Abstract: Multimodal clinical AI typically assumes that the waveform, report, metadata, and downstream predictions attached to a record belong to the same patient. In practice, linkage failures can silently assemble individually plausible but cross-patient components, creating a safety problem that standard predictive models are not designed to detect. We study this problem as multimodal record integrity triage: given an assembled record, should its modalities be trusted to belong together? We introduce SHIFT-M3, a lightweight text-based pre-fusion screen that measures alignment-based consistency between two separately produced ECG text views: an LLM-generated interpretation and a clinical report summary. On 784,680 MEETI ECG records, SHIFT-M3 achieves 97.6% TPR@5% FPR for full text-view swaps (AUROC 0.996), 90.3% for partial swaps (AUROC 0.974), and 97.7% for label-matched hard negatives (AUROC 0.996) with only 573,569 parameters. Compared with same-dataset lexical baselines, the gains are largest on partial swaps and hard negatives, suggesting that the model is learning more than surface overlap. We also introduce the CMST (Conflict-type Multimodal Stress Test) evaluation taxonomy, a three-seed stability study, a loss ablation, a temporal-tolerance sweep, and a shared-token masking control. The main remaining failure mode is longitudinal ambiguity: at the default operating point, same-patient cross-visit pairs still produce 87.0% Type-II false positives.

[132] North Small Translate: Advanced Cost-Effective Translation (Cohere CAT+) cs.CLPDF

Tom Kocmi, Alexandre Bérard, Phil Blunsom, Samuel Cahyawijaya, Shaun Cassini

TL;DR: 本文介绍了North Small Translate，一个基于指令跟随的大型语言模型（LLM）的机器翻译模型。它采用专家混合架构，拥有2180亿总参数和250亿激活参数，通过难度采样和五步训练协议（包括监督微调、直接偏好优化和在线强化学习）进行训练。该模型专注于高吞吐量，无需在推理时进行昂贵的计算，并在50种语言上实现了同类小参数模型中的顶级翻译性能，同时还能执行译后编辑和质量评估等相关任务。

Details

Motivation: 旨在开发一个开源的、基于LLM的机器翻译模型，该模型不仅具有指令跟随能力，还能在保证高吞吐量（推理效率）的同时，实现高质量的翻译，并扩展至译后编辑、质量评估等翻译相关任务。

Result: 在参数规模小于1万亿的模型类别中，North Small Translate在50种语言上实现了顶级的机器翻译性能。

Insight: 创新点在于结合了非推理基础模型以保证高吞吐量，并辅以可选的智能体能力来提升翻译质量；其训练流程整合了难度采样、监督微调、直接偏好优化和在线强化学习，这是一种系统化的高效训练方法。模型架构基于专家混合，能灵活激活参数，在效率与性能间取得平衡。

Abstract: We present North Small Translate, an open-weight, LLM-based machine translation (MT) model with instruction-following capabilities built on the same foundation as Cohere’s Command A Plus, a mixture-of-experts architecture with 25 billion active parameters out of 218 billion total parameters. North Small Translate is trained using difficulty sampling to obtain challenging documents and a five-step training protocol combining supervised fine-tuning, direct preference optimization, and online reinforcement learning. We prioritized throughput through a non-reasoning base model and supplemented with optional agentic capabilities to unlock translation quality gains. North Small Translate is trained to perform MT-related tasks, including post-editing and quality estimation, as well as related tasks such as general instruction following. The model achieves top MT performance across 50 languages in the class of models under 1T parameters, with no need to run expensive reasoning at inference time.

[133] Thought without systematicity? Evaluating reasoning models on rule induction tasks cs.CL | cs.AI | cs.LGPDF

Simon Schug, Brenden M. Lake

TL;DR: 本文通过扩展认知科学中的规则归纳任务，评估当前推理模型是否具备系统性思维能力。研究发现，尽管模型能正确解决特定任务，但在结构等效的任务变体上表现不佳，表明其行为缺乏系统性。

Details

Motivation: 动机是探究推理模型是否具备人类认知中的系统性原则，即模型能否在结构等效的任务变体上表现一致，以评估其认知能力的鲁棒性。

Result: 在规则归纳任务中，模型在结构等效变体（如重组和替换）上表现不一致，表明其性能缺乏系统性，无法超越特定评估环境。

Insight: 创新点在于将认知科学的系统性原则引入AI评估，通过任务同构创建变体来揭示模型行为的局限性；客观分析显示，这为评估模型泛化能力提供了新方法。

Abstract: A central tenet of human cognition is systematicity, the principle that understanding one concept is inherently tied to understanding close variations of that concept. Do reasoning models robustly exhibit such systematicity? If so, we would expect consistent performance on structurally equivalent variants of the same task. Here, we extend established rule induction tasks from cognitive science to assess the systematicity of thought in current reasoning models. Each task family has compositional structure that we use to create structurally equivalent task variations through task isomorphisms such as recombination and substitution. We find that despite being able to correctly solve a task, models often fail on structurally equivalent variants of the same task. These findings suggest that many model behaviors lack systematicity, rendering it difficult to robustly establish the cognitive abilities of reasoning models beyond the particular contexts they were evaluated in.

[134] Unlocking the Unsolvable: Teacher-Guided Curriculum for Data-Efficient RLVR cs.CLPDF

Yukang Zhu, Zhen Han

TL;DR: 本文提出了一种教师引导的课程学习方法，用于解决强化学习与可验证奖励（RLVR）中模型无法处理的‘无解’问题。通过利用更强模型的部分推理轨迹构建难度梯度，并采用后向链式课程逐步撤出指导，使模型最终能独立解决问题。该方法在仅使用128个无解问题的情况下，达到了与在2000个问题上训练的GRPO相当或更好的性能，实现了约16倍的数据效率提升，并显著扩展了模型在大型k值下的推理边界。

Details

Motivation: RLVR在提升大语言模型的数学推理能力方面取得了显著成功，但对于超出模型当前能力的‘无解’问题，由于无法产生学习信号，这些本应最具信息量的训练前沿问题被浪费了。本文旨在通过课程学习解锁这些无解问题，以更高效地利用训练数据。

Result: 在九个基准测试的平均结果上，仅使用128个无解问题训练，其性能匹配或超过了在完整2000个问题语料上训练的GRPO，数据效率提升约16倍。同时，该方法在大型k值下的pass@k指标上显著扩展了推理边界。此外，提出的单调前沿课程（MFC）方法在仅使用无解问题的训练机制中，持续优于现有课程学习方法。

Insight: 创新点在于利用更强模型的部分推理轨迹为无解问题构建难度梯度，并通过后向链式课程逐步撤出指导，使模型能自主学习解决原本无法处理的问题。客观分析认为，该方法有效解决了RLVR中无解问题训练信号缺失的瓶颈，并通过MFC方法缓解了分布偏移问题，为数据高效的强化学习训练提供了新思路。

Abstract: Reinforcement Learning with Verifiable Rewards (RLVR) has shown remarkable success in improving the mathematical reasoning of large language models. Yet problems beyond the model’s current capability, where rollouts uniformly fail and no learning signal is produced, are structurally wasted despite marking the most informative training frontier. We show that these otherwise-inert problems can be unlocked via teacher-guided curriculum learning: partial reasoning traces from a stronger model create a graded difficulty landscape, and a backward-chaining curriculum progressively withdraws guidance until the model solves problems unaided. Training on only 128 unsolvable problems matches or exceeds GRPO trained on a full 2,000-problem corpus (~16x data efficiency) on the nine-benchmark average for both base models, while substantially expanding the reasoning boundary measured by pass@k at large k. Furthermore, we identify a distribution-shift cost that is particularly acute in the unsolvable-only regime and propose Monotone Frontier Curriculum (MFC), a method that monotonically drives training toward unguided solving, consistently outperforming existing curriculum methods.

[135] GraMRAG: Orchestrating Multi-Agent Multi-Step Reasoning via Graph Memory with Reinforcement Learning cs.CL | cs.AI | cs.CVPDF

Zhongyu Wang

TL;DR: 本文提出了GraMRAG框架，通过图记忆和强化学习来协调多智能体多步推理，解决现有多智能体RAG系统在推理深度和记忆结构上的局限性。该框架集成了动态多模态记忆图，支持稳定的多步多模态推理，并引入视觉-文本桥接推理范式和拓扑感知策略优化，以增强长视野跨模态推理能力。

Details

Motivation: 现有多智能体RAG系统在复杂多模态推理任务中存在推理深度不足、记忆结构有限、检索不充分和状态盲区等问题，需要改进以应对知识密集型问题。

Result: 在具有挑战性的多模态基准测试中，该方法持续优于现有基线，并在复杂长视野推理任务上达到了最先进的性能水平。

Insight: 创新点包括：构建多模态记忆图，将智能体推理形式化为动态有向无环图，以建模动作-观察依赖关系；提出拓扑感知策略优化，利用图拓扑进行关键路径识别和节点剪枝，实现多步推理轨迹的细粒度信用分配；以及视觉-文本桥接推理范式，统一多尺度实体裁剪与ReAct风格视觉工具链。

Abstract: Although existing multi-agent Retrieval-Augmented Generation (RAG) systems have demonstrated promise on complex multimodal reasoning tasks, they remain fundamentally limited in reasoning depth and memory structure, suffering from inadequate retrieval and state blindness when answering knowledge-intensive questions. To address these limitations, we propose GraMRAG, a graph memory-guided multi-agent RAG framework that integrates a dynamic multimodal memory graph to enable stable, multi-step multimodal reasoning. We introduce a vision-text bridged reasoning paradigm that unifies multi-scale entity cropping with a ReAct-style visual toolchain, enhancing the long-horizon cross-modal reasoning capability. We further construct a multimodal memory graph that formalizes agent reasoning as a dynamic directed acyclic graph (DAG), explicitly modeling action-observation dependencies to mitigate state blindness and suppress redundant retrieval. Moreover, we propose Topology-Aware Policy Optimization (TAPO) that leverages graph topology for critical path identification and targeted node pruning, enabling fine-grained credit assignment across multi-step reasoning trajectories. Extensive experiments on challenging multimodal benchmarks demonstrate that our approach consistently outperforms existing baselines and achieves state-of-the-art performance on complex long-horizon reasoning tasks.

[136] One Size Does Not Fit All: Setting Inference Depth from the Questions a Deployment Actually Asks cs.CL | cs.LGPDF

Jerry Kaplan

TL;DR: 本文研究了在已知特定部署场景下提示范围的前提下，如何通过早期退出机制减少Transformer语言模型的推理计算成本。论文通过实验发现，可实现的节省高度依赖于任务类型，定制退出阈值能显著提升退出率，并指出当前基于token级保真度的标准评估方法在某些领域存在缺陷。

Details

Motivation: 解决Transformer模型在特定部署场景（如客服助手、代码工具）中，面对狭窄问题范围时仍需支付全量计算成本的问题，旨在探索如何根据实际提问范围预先减少不必要的推理开销。

Result: 在1.5B参数模型上，不同任务在保持token级保真度时早期退出率差异显著（算术应用题达96%，中文解释仅8%）；通过为每个部署定制退出阈值，退出率在三个模型上最高提升59个百分点；在算术应用题上，早期退出配置虽保真度高，但正确答案数量从全模型60个大幅下降至10-28个。

Insight: 创新点在于提出基于实际部署流量定制早期退出阈值的实用方法，仅需模型自身输出作为监督；客观分析揭示了token级保真度评估在可验证任务中的局限性，为面向边缘设备的高效推理优化提供了新视角。

Abstract: A transformer language model is trained to respond to any prompt, but each deployment asks only a narrow range of questions: a support assistant sees delivery complaints, a coding tool sees Python. Every deployment nonetheless pays the same computation per token. This paper measures how much of that cost is avoidable when the range of prompts is known in advance. The mechanism examined is early exit: a small, trained component - called a readout - is attached to an intermediate layer and proposes a token, and a confidence test decides whether to emit it or to run the remaining layers. The models are frozen, and the only supervision used is the model’s own output on ordinary traffic. Three findings are reported. First, achievable savings depend strongly on the kind of traffic: at half depth on a 1.5-billion-parameter model, 96 percent of tokens could be emitted early for arithmetic word problems and 8 percent for Chinese-language explanations, at matched token-level fidelity to the full model (a measure whose limits the third finding exposes). Second, of three ways a deployment might use knowledge of its traffic, only customizing the threshold for exiting early is worthwhile: calibrating it per deployment raised exit rates by up to 59 percentage points across three models, and by more than 10 points on most corpora tested. Third, token-level fidelity - the standard evaluation measure in the early-exit literature - fails in domains where tokens can be checked against ground truth: on arithmetic word problems, three models each answered sixty questions correctly when run in full, and between 10 and 28 correctly under early exit, in the configuration that scored highest on fidelity. The intended setting is small models on personal devices, where generation is limited by memory bandwidth rather than computation.

[137] Learning to Refer from Estimated Listener Gaze cs.CLPDF

Téa Wright, Alane Suhr

TL;DR: 本文提出了一种通过微调视觉语言模型来生成更符合语用学最优指代表达式的方法。该方法利用估计的听者注视扫描路径作为学习信号，将听者理解过程转化为奖励，以优化说话者策略。实验表明，该方法能显著缩短表达长度并提高指代成功率。

Details

Motivation: 动机在于利用听者理解过程中的隐含观察（如注视扫描路径）来优化指代表达式的生成，而不仅仅依赖于显式的交流成功信号。

Result: 在人类听者评估中，使用注视估计听者训练的说话者策略比基线模型在语用学上更优，将序列长度从15.4词减少到4.0词，同时将指代成功率从75.2%提高到80.0%。

Insight: 创新点在于将听者注视行为作为学习信号，通过神经听者模型估计人类注视扫描路径，并将其转化为词级和序列级奖励，以优化视觉语言模型的指代生成策略，这为基于交互的语言生成提供了新思路。

Abstract: We propose to finetune vision-language models to generate more pragmatically optimal referring expressions by transforming observations of incremental listener comprehension, in the form of gaze scanpaths, into learning signals. During training, referring expressions are sampled from the speaker policy being optimized, conditioned on images and target referents; then, a neural listener estimating human gaze behavior maps from images and sampled referring expressions to scanpaths, each represented by a sequence of fixations, with each fixation corresponding to a word in the referring expression. We experiment with several approaches to convert fixation sequences and target referents into token- and sequence-level rewards, which are used to optimize policy parameters. Through evaluation with human listeners, we find that speaker policies trained with gaze-estimating listeners result in significantly more pragmatically-optimal references than base models, reducing sequence length from 15.4 down to 4.0 words while increasing referential success from 75.2 up to 80.0%. Our work demonstrates a promising opportunity for learning to generate utterances through language-based interaction, not only from the explicit signal of communicative success, but also from implicitly-available observations of a listener’s process of comprehension.

[138] E2A-Bench: Benchmarking Evidence-to-Action Reliability in Financial Chart Reasoning cs.CLPDF

Xiaoya Wang, Yutong Xu, Junjie Wang

TL;DR: 本文提出了E2A-Bench，一个包含969个查询的金融图表推理基准测试，用于评估金融视觉语言模型（VLM）将图表证据转化为可靠行动建议的能力。该基准从323个HS300成分股中构建，包含三种输入模态和确定的OHLCV衍生证据锚点。通过评估20个VLM，研究发现仅依赖单一的幻觉分数会掩盖三个关键失败模式，并表明金融VLM评估应追踪完整的证据到行动链条。

Details

Motivation: 现有幻觉评估大多是声明中心的，评估生成的陈述是否被支持，但未评估证据是否通过推理、置信度和最终行动保持可追溯性。本文旨在解决金融VLM能否将图表证据转化为可靠行动建议的问题。

Result: 在E2A-Bench上评估20个VLM，揭示了被标量幻觉分数隐藏的三个失败：最低UCR模型因仅6.4%的方向覆盖率而在NDR排名中接近底部；预言机辅助验证减少了不受支持的声明但可能崩溃覆盖率；金融微调在严格的基座-微调模型对中将买入卖出比率放大了4.21至4.68倍。

Insight: 创新点在于提出了一个专注于证据到行动可靠性的金融图表推理基准（E2A-Bench），并引入了NDR等指标来衡量覆盖感知的可靠性，而非已实现的交易性能。客观分析认为，其核心洞察是强调评估需要追踪完整的证据-推理-置信-行动链条，而不仅仅是最终的声明支持性，这对于高风险金融应用至关重要。

Abstract: Can financial vision-language models (VLMs) turn chart evidence into reliable action recommendations? Existing hallucination evaluations are mostly claim-centric; they assess whether generated statements are supported, but not whether evidence remains traceable through rationale, confidence, and final action. We introduce E2A-Bench, a 969-query benchmark for financial chart reasoning, constructed from 323 HS300 constituents under three input modalities with deterministic OHLCV-derived evidence anchors. E2A-Bench evaluates grounding, reasoning-action consistency, evidence-confidence calibration, and directional coverage through UCR, RCI, ECI, and NDR, where NDR measures coverage-aware evidence-to-action reliability rather than realized trading performance. Evaluating 20 VLMs reveals three failures hidden by scalar hallucination scores: the lowest-UCR model ranks near the bottom by NDR due to only 6.4% directional coverage; oracle-aided verification reduces unsupported claims but can collapse coverage; and financial fine-tuning amplifies the BUY:SELL ratio by factors of 4.21 to 4.68 across strict base-fine-tuned pairs. These results show that financial VLM evaluation should trace the full evidence-to-action chain rather than rely on a single hallucination score. Code and data: https://github.com/wanng-ide/E2A-Bench

Eduin E. Hernandez, Luis F. Garcia, Nurassyl Askar, Sergio A. Diaz, Stefano Rini

TL;DR: 该论文将多跳知识图谱问答（KGQA）重新定义为问题条件化的图导航问题，提出THESEUS框架以实现可追溯的推理路径。作者通过构建带标注路径的数据集、设计评估协议以及改进现有路径推理代理，推动KGQA系统向可解释性方向发展。

Details

Motivation: 现有KGQA系统通常只预测最终答案，缺乏对中间推理步骤的显式建模和验证，无法确保答案来自可靠的多跳推理。

Result: 论文在KINSHIP和MQuAKE数据集上构建了导航就绪的KGQA资源，并设计了评估路径保真度、语言变化鲁棒性以及多跳/多答案问题性能的协议。

Insight: 创新点在于将KGQA重构为可追溯的图导航任务，并通过使用完整问题嵌入（而非符号化单关系查询）来引导基于路径的KG完成代理，增强了自然语言语义对推理轨迹的指导能力。

Abstract: Multi-Hop Knowledge Graph Question Answering (KGQA) tasks require models to assemble relational evidence along paths in a KG to answer natural-language questions. However, existing KGQA systems typically focus on predicting the final answer without explicitly modeling or validating the intermediate reasoning steps, obscuring whether the correct answers arise from faithful multi-hop reasoning. To address this limitation, we re-frame multi-hop KGQA as a question-conditioned graph navigation problem. We refer to this formulation as THESEUS - Traceable Hop-wise Evidence SEarch in a Unified Semantics. In this setting, an agent receives a KG, a question, and a topic entity, and traverses a sequence of relations towards the answer, making the reasoning path explicit. To systematically study this formulation, we provide three key contributions. (i) We augment the existing KINSHIP and MQuAKE resources into navigation-ready KGQA datasets with annotated evidence paths and paraphrased questions. (ii) We design evaluation protocols to measure path fidelity, robustness to linguistic variation, and performance across multi-hop and multi-answer questions. (iii) We adapt established path-based KG completion agents - MINERVA, MultiHopKG, and SQUIRE - to operate on full question embeddings rather than symbolic single-relation queries, enabling their trajectories to be guided by natural-language semantics. Together, these contributions advance KGQA research toward systems where traceability is fundamental: answers are accompanied by explicit reasoning paths whose agreement with reference evidence can be systematically evaluated.

[140] CompCQR: Compositional Query Generation for Training-Free Conversational Search cs.CL | cs.AIPDF

Yunah Jang, Kang-il Lee, Joongbo Shin, Kyomin Jung

TL;DR: 本文提出了一种名为CompCQR的新型免训练对话查询生成方法，用于解决信息检索场景中多轮对话查询的歧义性问题。该方法通过组合少量原子成分来生成大量查询，并利用LLM推理构建高质量的文档集，从而在减少LLM调用次数的同时，显著提升了检索性能。

Details

Motivation: 在信息检索的多轮对话中，用户查询通常具有歧义性和上下文依赖性，不适合直接用作检索查询。现有的基于LLM的对话查询改写方法虽然性能强大，但存在重复调用LLM以及与下游检索器不对齐的问题。

Result: 该方法在四个广泛使用的对话基准测试中取得了强劲性能，相对于之前的最先进基线，相对MRR提升了高达22.5%，同时LLM调用次数大幅减少。

Insight: 创新点在于观察到检索器对内容顺序高度敏感，并据此提出了一种组合式查询生成策略，通过最小化LLM使用来生成大量查询，同时利用LLM推理构建平衡精确率和召回率的文档集，实现了跨开源/闭源LLM以及稠密/稀疏检索器的泛化。

Abstract: Multi-turn interactions with LLMs are becoming increasingly common in information-seeking scenarios. However, user queries are often ambiguous and context-dependent, making them ill-suited for direct use as retriever queries. Conversational query reformulation (CQR) addresses this issue by rewriting the current utterance into a stand-alone query grounded in the dialogue history. Recent LLM-based CQR approaches achieve strong performance; however, their repeated LLM invocations and misalignment with downstream retrievers remain challenges. In this work, we begin from the observation that retrievers are highly sensitive to content ordering: simply reordering the same content can lead to changes in retrieval coverage and performance. Based on this, we propose a novel training-free method that generates a very large number of queries with minimal LLM usage by compositionally combining a small set of atomic components. We further apply LLM reasoning to construct a high-quality document set that balances precision and recall while capturing the user’s core intent. Our framework generalizes across both open- and closed-source LLMs as well as dense and sparse retrievers. It achieves strong performance on four widely used conversational benchmarks, with up to 22.5% relative MRR improvement over the previous state-of-the-art baseline with far fewer LLM calls.

[141] TATK: Triple-Aware Top-K Learning with Knowledge-Grounded Verification for LLM-based Sequential Recommendation cs.CLPDF

Yuchen Guan, Jiaye Liu, Yifei Han, Zhenxi Zhang, Yixuan Weng

TL;DR: 本文提出了TATK框架，用于解决基于LLM的序列推荐系统中，文本生成接口与全目录Top-K排序任务不匹配的问题。TATK结合了Top-K学习和知识驱动的验证，通过上下文感知的元数据知识图谱提示和位置感知的Top-K奖励来优化训练，并在单次LLM前向传播后对候选集进行结构感知的重排序。

Details

Motivation: 动机在于，当前基于大语言模型的序列推荐器通常将下一项预测视为文本生成任务，但这种接口与需要从全目录中进行Top-K排序的实际推荐需求严重不匹配。

Result: 在Amazon Reviews 2023的三个数据集上，使用Gemma-2-2B-It和Qwen2.5-3B-Instruct作为骨干模型进行评估。TATK在所有36个报告指标上均优于基准方法，在NDCG@10指标上，Qwen模型提升了8.05%、4.26%和3.78%，Gemma模型提升了27.03%、10.52%和10.23%，同时推理延迟仅为基础方法的1.17倍。

Insight: 创新点在于提出了一个三元感知框架，将Top-K学习与知识驱动的验证耦合，通过结构感知的重排序来提升推荐质量。客观分析表明，其核心洞察是：当候选项目具有可靠的知识图谱支持时，结构证据最为有效；而当元数据关系稀疏或嘈杂时，应进行门控处理。

Abstract: LLM-based sequential recommenders usually cast next-item prediction as text generation, but this interface is poorly matched to full-catalog top-K ranking. We propose TATK, a Triple-Aware framework that couples Top-K Learning (TKL) with Knowledge-Grounded Verification (KGV) for LLM-based sequential recommendation. Top-K Learning combines context-aware metadata-KG prompt grounding with position-aware top-K rewards, aligning training with ranking utility; Knowledge-Grounded Verification then applies structure-aware reranking over the top-M candidates after a single LLM forward pass, using the same metadata-derived item graph. We evaluate TATK on Musical Instruments, CDs and Vinyl, and Video Games from Amazon Reviews 2023 under a matched R2ec-style full-catalog protocol. Experiments use Gemma-2-2B-It and Qwen2.5-3B-Instruct backbones, compare against sequential, generative, KG-augmented, and reasoning-enhanced baselines, and include component, reward-shape, sequence-perturbation, reranking, relation-quality, and candidate-pool diagnostics. TATK improves over the matched R2ec reproduction on all 36 reported metrics. On NDCG@10, it improves Qwen by 8.05%, 4.26%, and 3.78% on the three datasets, and improves Gemma by 27.03%, 10.52%, and 10.23%, while keeping inference within 1.17x of Base RecPO latency. The diagnostics show that structural evidence is most useful for recoverable top-M candidates with reliable KG support, and should be gated when metadata relations are sparse or noisy.

[142] Optimizing Sparse Outcomes Through Dense Behavioral Signals via Value-Guided Preference Distillation cs.CLPDF

Ziyi Zhu, Daniel R. Cahn, Thomas D. Hull, Caitlin A. Stamatis, Olivier Tieleman

TL;DR: 本文提出了一种通过价值引导偏好蒸馏来优化稀疏长期对话结果的方法。该方法将多轮对话对齐问题构建为多目标强化学习任务，训练一个多头价值模型来预测用户在不同前瞻视野下的行为向量。通过结合密集的辅助行为信号进行信用分配，有效优化了稀疏的长期结果，并通过反事实用户模拟和对话级结果模型构建了安全评估框架。最终，通过参考锚定偏好优化将多目标价值偏好蒸馏到策略中，显著提升了用户留存率和积极行为指标。

Details

Motivation: 传统的多轮对话智能体对齐通常关注于匹配单轮的人类偏好，但直接优化长期稀疏结果往往效果不佳且容易导致奖励黑客问题。因此，需要一种能够有效进行信用分配并优化长期稀疏结果的方法，同时避免因优化单一代理目标而导致的策略退化风险。

Result: 在线A/B测试证实，经过蒸馏的策略显著改善了长期用户留存率，同时增强了积极行为和治疗过程标记。该方法在计算预算仅为在线强化学习一小部分的情况下，达到了与之相当的性能水平。

Insight: 创新点在于将长期对话优化构建为多目标强化学习问题，并利用密集的辅助行为信号进行有效的信用分配。此外，通过结合反事实用户模拟和已验证的对话级结果模型，建立了一个安全评估框架，用于在部署前识别潜在的策略退化风险。最后，提出的参考锚定偏好优化方法能够高效地将多目标价值偏好蒸馏到策略中，实现计算效率与性能的平衡。

Abstract: Aligning multi-turn dialogue agents is usually framed as matching turn-level human preferences, yet direct optimization of long-term outcomes is often ineffective and prone to reward hacking. We formulate long-horizon dialogue optimization as a multi-objective reinforcement learning problem and train a multi-head value model that predicts a vector of observed user behaviors across multiple look-ahead horizons. Our findings demonstrate that a scalarized composite of dense auxiliary behavioral signals enables effective credit assignment and optimization of sparse outcomes. However, optimizing unconstrained single-objective proxies might induce policy degradations that are harmful when the agent is exposed to real users. To identify these failure modes prior to deployment, we establish a safety framework combining counterfactual user simulation with a validated dialogue-level outcome model to evaluate preference weightings and policy optimization methods. Finally, we demonstrate that distilling multi-objective value preferences into the policy via reference-anchored preference optimization matches on-policy online RL at a small fraction of its compute budget. Live A/B testing confirms that our distilled policy significantly improves long-term user retention, while simultaneously enhancing the positive behaviors and therapeutic-process markers.

[143] Building Legal Reward Models for Grounding and Abstention cs.CL | cs.AIPDF

Rilton Franzone, Valentin Noël, Puyu Wang, Philip Torr, Fabio J. Fehr

TL;DR: 本文提出了一种构建法律奖励模型的方法，用于评估检索增强生成（RAG）系统中基于证据的推理和证据不足时的弃权能力。作者通过将现有法律问答数据集转化为上下文偏好数据，构建了LegalRewardBench（LRB）基准，并发现上下文DPO和长度平衡增强能显著提升模型在嘈杂和证据不足检索条件下的法律生成评估性能，且模型展现出跨司法管辖区的迁移能力。

Details

Motivation: 现有奖励模型主要针对通用偏好优化，而非上下文证据的准确性，这限制了其在需要基于检索证据进行推理和弃权的法律等高风险领域RAG设置中的评估能力。

Result: 在LRB基准上的实验表明，结合长度平衡的法律和通用上下文偏好数据的最强配置，比基线性能提升高达+25.6个百分点。模型在维多利亚刑法数据上精炼后，在美国法律基准（如Housing Statute QA）上的评估性能也提升了+16.2个百分点，显示出跨司法管辖区迁移的有效性。

Insight: 创新点在于提出了一个将法律QA数据集转化为上下文偏好数据的可复现框架，并构建了专门的LRB基准。客观分析认为，其核心洞察是强调了在RAG中构建领域特定（如法律）奖励模型时，偏好数据的构造（如长度平衡）对提升模型基于证据的推理和弃权评估至关重要，且领域精炼模型具有可迁移性。

Abstract: Large language models are increasingly used in high-stakes domains such as law, where systems must ground their reasoning in retrieved evidence and abstain when that evidence is insufficient. However, existing reward models are largely optimised for general preferences rather than contextual grounding, limiting their ability to evaluate these behaviours in retrieval-augmented generation (RAG) settings. We introduce a framework for transforming existing legal QA datasets into contextual preference data and use it to construct LegalRewardBench (LRB), a benchmark for evaluating grounded legal generation under noisy and insufficient retrieval conditions. Across general and legal contextual evaluation, we find that contextual DPO improves grounded evaluation, but performance is sensitive to preference-data construction. Length-balanced augmentation substantially improves grounded legal evaluation, with the strongest configuration combining length-balanced legal and general contextual preference data and improving performance by up to $\mathbf{+25.6}$pp over baseline. We further find evidence of cross-jurisdiction transfer: models contextually refined primarily on Victorian criminal-law data improve grounded evaluation on external US legal benchmarks, including a $\mathbf{+16.2}$pp improvement on \textsc{Housing Statute QA}. Together, these results provide a reproducible foundation for constructing and evaluating grounded legal reward models in retrieval-augmented settings.

[144] Func-R1: Incentivizing Mathematical Function Reasoning in Multimodal Large Language Models cs.CLPDF

Mingze Yin, Xiaohan Wang, Dian Li, Haichao Yao, Yilin Zhao

TL;DR: 本文提出Func-R1方法，旨在解决多模态大语言模型在视觉数学函数推理中存在的模态干扰问题。该方法通过解耦架构和分层后训练框架，结合感知对齐理论优化策略，提升模型对视觉线索的感知与逻辑推理能力。在多个基准测试中，Func-R1在开源MLLMs中达到最优性能，并在MathVerse函数任务上超越GPT-5。

Details

Motivation: 针对多模态大语言模型在数学函数推理中忽视或误解关键视觉线索的模态干扰现象，研究旨在增强模型在视觉上下文中的精确感知与符号逻辑融合能力。

Result: 在多样化基准测试中，Func-R1在开源MLLMs中取得最优性能，在MathVerse函数导向任务上比GPT-5提升8.4%，达到SOTA水平。

Insight: 创新点包括显式解耦架构与分层后训练框架，以及感知对齐理论优化策略，可借鉴于提升多模态任务中视觉与文本模态的协同推理鲁棒性。

Abstract: Performing deliberate mathematical reasoning in visual contexts is a hallmark of advanced Multimodal Large Language Models (MLLMs) and requires a sophisticated synthesis of perceptual grounding and symbolic logic. However, in the realm of mathematical functions, our investigation reveals a critical modality interference phenomenon: even advanced models, while performing textual computational reasoning, tend to disregard or misinterpret essential visual cues. To address this challenge, we propose Func-R1, which synergistically harmonizes precise visual perception and rigorous logical reasoning. Concretely, built upon an explicitly decoupled architecture, we employ a hierarchical post-training framework to progressively identify critical visual evidence and conduct in-depth theoretical reasoning. Furthermore, the Perception-Aligned Theoretic Optimization (PATO) strategy is proposed to steer policy updating towards internalizing fundamental theoretical properties while dynamically rectifying heterogeneous visual information throughout the reasoning process. Extensive experiments across diverse benchmarks demonstrate that Func-R1 delivers the optimal performance among open-source MLLMs, even surpassing GPT-5 with an 8.4% improvement on MathVerse’s function-oriented tasks.

[145] MedTRACE: Tool-Augmented Multimodal Clinical Reasoning Agents for Evidence-Grounded Decision-Making cs.CLPDF

Ji Lu, Lifei Liu, Haoran Yu, Xianglong Wang, Yiru Fang

TL;DR: 本文提出了MedTRACE，一种工具增强的多模态临床推理智能体，用于基于证据的决策。它通过特定模态编码器构建统一的患者状态表示，并执行假设形成、工具感知审议和证据验证的迭代循环，动态调用视觉定位、证据检索和结构化解析工具来获取诊断相关证据。最终输出诊断结果、支持证据、可审计的推理轨迹和校准后的置信度。

Details

Motivation: 现有的多模态临床决策模型通常直接将异构输入映射到诊断，而没有明确评估证据充分性、工具使用需求或诊断不确定性。本文旨在解决这一问题，开发一个能够主动获取和验证证据的推理系统，以提高临床决策的准确性、可解释性和可靠性。

Result: 在多个多模态临床诊断基准测试中，MedTRACE将诊断准确率提高了5.4%，AUROC提高了4.7个百分点。同时，证据选择F1分数提高了8.2个百分点，视觉定位IoU提高了6.5个百分点，预期校准误差降低了31.6%，无支持诊断错误减少了27.8%，超越了现有最强基线。

Insight: 论文的核心创新在于提出了一个主动证据获取与验证的迭代推理框架，将工具调用（如检索、解析）与假设形成、证据一致性验证紧密结合。这为构建可解释、可靠的多模态AI决策系统提供了一种通用范式，其证据记忆和置信度校准机制也值得借鉴。

Abstract: Multimodal clinical decision-making requires reliable reasoning over heterogeneous evidence from electronic health records, medical images, and physiological signals. Existing models typically map these inputs directly to diagnoses without explicitly assessing evidence sufficiency, tool-use requirements, or diagnostic uncertainty. This paper presents MedTRACE, a tool-augmented multimodal clinical reasoning agent for evidence-grounded decision-making. MedTRACE uses modality-specific encoders to construct a unified patient-state representation and performs an iterative loop of hypothesis formation, toolaware deliberation, and evidence verification. It dynamically invokes visual grounding, evidence retrieval, and structured parsing tools to locate diagnosis-relevant regions, retrieve clinical knowledge and similar cases, and extract structured findings. The acquired evidence enters an evidence memory, where a consistency verifier confirms or revises the current hypothesis. MedTRACE outputs a diagnosis together with supporting evidence, an auditable reasoning trace, and calibrated confidence. Experiments on multiple multimodal clinical diagnosis benchmarks show that MedTRACE improves diagnostic accuracy by 5.4% and AUROC by 4.7 percentage points over the strongest baseline. It also improves evidenceselection F1 by 8.2 percentage points and visual-grounding IoU by 6.5 percentage points, reduces expected calibration error by 31.6%, and decreases unsupported diagnostic errors by 27.8%. These results demonstrate that active evidence acquisition and verification improve the accuracy, interpretability, and reliability of multimodal clinical decisionmaking.

[146] Enemray: Toward Capable Language Models for Hassaniya cs.CL | cs.AIPDF

Cheikh Ahmed

TL;DR: 本文介绍了Enemray，一个专注于哈桑尼亚语（Hassaniya）的语言模型，旨在实现哈桑尼亚语的通用交互。该模型围绕稳定性-可塑性目标进行训练：在保持原有指令调优模型的通用推理、多语言、指令遵循和安全行为的同时，获得强大的哈桑尼亚语言和文化能力。开发流程将语言获取与行为专门化分离，通过分层选择性持续预训练、参数更新转移和监督后训练来实现。

Details

Motivation: 解决哈桑尼亚语缺乏强大语言模型的问题，目标是开发一个既能处理哈桑尼亚语通用交互，又保留基础模型通用能力的模型。

Result: 在评估中，Enemray在英语到哈桑尼亚语翻译上优于对比的开源和专有模型，在毛里塔尼亚翻译错误检测上获得最高分，同时在数学推理、知识、代码生成和函数调用等通用能力上基本保持了基础模型的性能。

Insight: 创新点包括稳定性-可塑性训练目标、分离语言获取与行为专门化的开发流程，以及整合新收集、重建和构建指令数据的大规模监督语料库，为低资源语言模型开发提供了可借鉴的方法。

Abstract: We introduce Enemray, a Hassaniya-centric language model that enables general-purpose interaction in Hassaniya. Enemray is trained around a stability–plasticity objective: acquire strong Hassaniya linguistic and cultural competence while preserving the general reasoning, multilingual, instruction-following, and safety behaviors of a capable instruction-tuned model. The development pipeline separates language acquisition from behavioral specialization. A separately assembled continual-pretraining corpus provides broad exposure to natural Hassaniya and Mauritanian text; layer-selective continual pretraining learns a compact language-specific parameter update; that update is transferred into the instruction-tuned parameter space; and supervised post-training develops conversational, cultural, literary, task-oriented, and cross-lingual behavior. The supervised corpus integrates selected public Hassaniya and Mauritanian resources with a substantially larger body of newly collected, reconstructed, curated, and constructed instruction data, while policy-generated replay provides a retention signal from the reference model’s own behavior distribution. The resulting collection is substantially larger and broader in purpose than existing Hassaniya text resources. In evaluation, Enemray achieves the strongest English to Hassaniya translation among the compared open and proprietary models and the highest overall score on Mauritanian translation error detection, while retaining most of the general capabilities of its instruction-tuned base model on mathematical reasoning, knowledge, code generation, and function calling. This report describes the motivation, data construction, model design, training methodology, and evaluation of Enemray.

[147] ModularRSI: Modular and Generalizable Recursive Harness Self-Improvement cs.CLPDF

Siwei Wu, Jincheng Ren, Yizhi Li, Haau-Sing Li, Chengran Yang

TL;DR: 本文提出了ModularRSI，一个用于实现通用化递归自我改进（RSI）的模块化框架。该框架通过对比同一任务的成功与失败轨迹，跨任务聚合证据以识别重复的行为缺陷，并将可改进的代理系统分解为五个功能模块进行独立演化。实验表明，该方法在未见过的领域内和跨领域任务上均能带来持续改进，且改进后的系统能迁移到不同的基础模型上。

Details

Motivation: 现有基于评估基准或其子集的递归自我改进方法难以区分可复用的改进与基准特定的适应，且单轨迹更新容易混淆系统缺陷与实例特定的推理细节，导致改进难以泛化。同时，整体式系统的优化难以定位重复行为缺陷，并可能使无关机制纠缠在一起。

Result: 在TB2.0和SWE-Bench Verified基准上的实验显示，该方法在未见过的领域内和跨领域任务上均取得了持续的性能提升，达到了先进的水平（SOTA），并且演化后的系统能够成功迁移到不同的基础模型。

Insight: 创新点在于提出了一个基准无关、对比式且模块化的通用化系统演化框架。通过任务对比和跨任务证据聚合来识别系统缺陷，并将系统分解为独立演化的功能模块（如代理循环、工具使用等），最后进行集成，这有助于实现更清晰、可归因且可泛化的改进。

Abstract: Recent work extends recursive self-improvement (RSI) to agent harnesses for long-horizon coding and terminal tasks, enabling agents to improve execution mechanisms from experience. However, generalizable harness RSI remains challenging. First, evolving harnesses on evaluation benchmarks or their subsets makes it difficult to distinguish reusable improvements from benchmark-specific adaptation. Second, single-trajectory updates can conflate systematic harness deficiencies with instance-specific reasoning and solution details, producing modifications that transfer poorly to unseen tasks. Third, localizing recurring behavioral deficiencies within monolithic harnesses is difficult, while whole-harness optimization can entangle unrelated mechanisms and complicate attribution and validation. We propose ModularRSI, a benchmark-disjoint, contrastive, and modular framework for generalizable harness evolution. ModularRSI contrasts successful and failed trajectories for the same task and aggregates evidence across tasks to identify recurring behavioral deficiencies. It decomposes the evolvable harness into five functional modules: Agent Loop, Tool Use, Observation Management, Context Management, and Task Completion Detection. Each module evolves independently within a restricted modification scope, followed by an integration stage that combines the evolved modules into a unified harness and resolves potential conflicts. To support benchmark-disjoint evolution, we curate 2,000 executable evolution tasks from external sources that are disjoint from downstream evaluation benchmarks. Experiments on TB2.0 and SWE-Bench Verified show consistent improvements on unseen in-domain and cross-domain tasks, with the evolved harness also transferring across different foundation models.

[148] One Example Is Enough to Pass Fairness Benchmarks: Rethinking Fairness Evaluation for Aligned LLMs cs.CL | cs.AIPDF

Naihao Deng, Samee Arif, Shuaichen Chang, Yulong Chen, Rada Mihalcea

TL;DR: 本文指出当前主流的公平性基准测试（如BBQ）过于简单，仅需一个示例即可显著提升模型性能，无法有效评估大语言模型的公平性对齐。通过实验证明，在Qwen 2.5 7B Base模型上使用单个BBQ示例进行GRPO微调或上下文学习，即可将平均准确率从79.9%分别提升至92.9%和99.0%，接近甚至超越大规模RLHF模型的表现。

Details

Motivation: 针对现有公平性基准测试（如BBQ）可能因设计缺陷而无法真实反映大语言模型的公平性对齐水平，论文旨在揭示这些基准的局限性，并呼吁构建更全面的评估体系。

Result: 在BBQ基准上，使用单个示例进行GRPO微调或ICL，可使Qwen 2.5 7B Base模型的平均准确率大幅提升至92.9%和99.0%，分别缩小了与大规模RLHF模型（96.1%）80%的差距并实现超越；该现象在不同模型家族中具有普适性。

Insight: 论文创新性地揭示了BBQ类基准仅测量单一结构性线索（如模型生成的推理轨迹中‘缺失证据’模式），而非真正的公平性能力；主张公平性评估需覆盖更广泛的维度，避免以偏概全。

Abstract: Warning: This submission studies stereotypes and biases, and contains toxic and offensive examples, used for illustration purposes only. Fairness benchmarks such as BBQ have become the de facto standard for fairness evaluation across major model families. We argue that these benchmarks are too easy to support their role: training Qwen 2.5 7B Base with Group Relative Policy Optimization (GRPO) on a single BBQ example, or placing that example in context as a one-shot demonstration for in-context learning (ICL), lifts mean BBQ accuracy from 79.9% to 92.9% and 99.0%, respectively, closing 80% of the gap to its large-scale RLHF counterpart (96.1%) with GRPO, and surpassing it with ICL. These effects generalize across model families. A cross-conditioning analysis shows the improvement is carried by the reasoning traces generated by the model, and one example suffices to elicit a category-agnostic ``missing evidence’’ reasoning pattern. We argue that BBQ-style multiple-choice abstention benchmarks measure a single structural cue, and a model that solves them does not thereby become fair. We call for evaluation suites that cover a broader spectrum of fairness alignment.

[149] Forty Shades of Blue: Quality-Diversity Alignment via Mode-Conditioned Reinforcement Learning cs.CL | cs.AI | cs.LG | cs.MAPDF

Jiayi Yuan, Hangoo Kang, James Jihao Liu, Yejin Choi, Vikram Iyer

TL;DR: 本文提出了一种名为MoDA（Mode-conditioned Diversity Alignment）的在线后训练强化学习算法，旨在解决大语言模型对齐训练中常见的模式崩溃问题，即输出多样性丧失。该算法通过将单个共享LLM策略与抽象编号角色（模式）条件化，激励不同角色探索高质量输出空间中的互补区域，从而联合优化生成质量和多样性。

Details

Motivation: 动机在于解决LLM对齐训练导致的模式崩溃问题，该问题会损害模型在推理时的表达能力，尤其限制了需要开放式探索和多元视角的应用，如科学构思和创意写作。

Result: 在涵盖七个通用能力任务和四个特定领域多样性任务（科学构思和创意写作）的基准测试中，MoDA在Infinite-Chat保留提示上将SBERT多样性提高了265%，同时将平均通用能力pass@1比Qwen3-8B基线提高了10.3%。与最强的DivPO基线相比，MoDA将SBERT多样性从0.274提升至0.482（+75.9%），E-Vendi从2.86提升至4.4（+53.8%），同时平均通用能力pass@1提高了7.0%。

Insight: 创新点在于从多智能体强化学习（MARL）的协调视角出发，提出了模式条件化的角色竞争机制，无需手工设计角色或修改架构即可促进多样性探索；同时引入了提示自适应的质量门控机制，仅对达到质量阈值的响应给予多样性奖励，有效防止了损害响应质量的奖励黑客行为。

Abstract: A notable byproduct of LLM alignment training is mode collapse: the progressive loss of output diversity that narrows a model’s expressivity at inference time. This degradation is especially limiting for applications requiring open-ended exploration and pluralistic perspectives, such as scientific ideation and creative writing. We present MoDA (Mode-conditioned Diversity Alignment), an online post-training RL algorithm that jointly optimizes generation quality and diversity, inspired by the coordination perspective in multi-agent reinforcement learning (MARL). MoDA trains a single shared LLM policy conditioned on abstract numbered roles, where each role acts as an agent competing to produce outputs distinct from the others. This formulation encourages mode-conditioned agents to explore complementary regions of the high-quality output space without requiring hand-crafted personas or architectural modifications. MoDA employs a prompt-adaptive quality gating mechanism that calibrates a reference quality threshold and grants diversity rewards only to responses that meet the threshold, preventing reward-hacking behaviors that compromise response quality. To study quality-diversity tradeoffs, we evaluate MoDA on a comprehensive suite of benchmarks spanning seven general capability tasks and four domain-specific diversity tasks in scientific ideation and creative writing. MoDA improves SBERT diversity by 265% on the Infinite-Chat held-out prompts, while increasing average general capability pass@1 by 10.3% over the Qwen3-8B baseline. Compared with the strongest DivPO baseline, MoDA improves SBERT diversity from 0.274 to 0.482 (+75.9%) and E-Vendi from 2.86 to 4.4 (+53.8%), while improving average general capability pass@1 by 7.0%. Overall, MoDA provides a drop-in alternative to standard post-training methods that preserves and expands the model’s expressive output space while improving quality.

[150] Beyond Depth and Width: The Information-Slack Dilemma in Streaming Test-Time Compute cs.CLPDF

Xiaotian Zhang

TL;DR: 本文探讨了在流式测试时计算中，当证据以不同顺序到达时，相同任务和计算预算需要不同推理策略的问题，即信息-松弛困境。论文提出将证据依赖的计算任务作为分析单元，研究何时启动、如何支持其结果以及何时提交结果，并倡导一个在演化证据下进行计算的研究议程，优先考虑在受控证据修订下的选择性恢复。

Details

Motivation: 解决在流式处理中，早期计算基于不完整或可修订证据与等待更多信息但减少计算松弛之间的权衡问题，即信息-松弛困境，旨在优化计算策略以实现更可靠、及时的响应。

Result: 论文未提及具体定量实验结果或基准测试，而是提出了一个研究议程和评估框架，强调分离早期执行效应、部署价值与预测策略的附加价值，并考虑共享资源成本。

Insight: 创新点在于将证据依赖的计算任务作为分析单元，系统化研究信息-松弛困境，并倡导优先选择性恢复的研究议程，目标是在声明资源范围内实现更可信、准时的响应，而非最大化提前计算。

Abstract: The same task and compute budget can require different reasoning policies when evidence arrives in a different order. Early computation has more time to finish but rests on incomplete or revisable evidence; waiting improves information while shrinking computational slack. We call this the information-slack dilemma. We take the evidence-dependent computational job as the unit of analysis: when to start it, what supports its result, and when that result can be committed. Advance computation is valuable only insofar as its benefits survive the costs of verification, invalidation, and recovery. This applies to grounded incremental processing and reusable preparation as well as future-dependent speculation. We propose a research agenda on computation under evolving evidence, prioritizing selective recovery under controlled evidence revisions. Evaluation should separate earlier-execution effects, deployment value against a full-input alternative, and the added value of predictive policies, while accounting for shared-resource costs. The objective is not maximal advance computation, but more trustworthy, on-time responses within a declared resource envelope.

[151] SALUTE: Benchmarking and Adapting LLMs for the Defense Domain cs.CLPDF

Hyeongcheol Park, Sumin In, Suyeon Myeong, Hogun Park, Sangmin Kim

TL;DR: SALUTE 构建包含语料、指令对话、偏好数据和评测基准的国防领域 LLM 全流程适配框架，并以多阶段后训练开发 Salute-LLM。

Details

Motivation: 国防领域要求理解专门术语、条令、作战流程和持续变化的军事事件；已有工作较零散、任务特定，或缺少完整适配流程和综合评估。

Result: SALUTE 包含来自公开美军条令及政府文件的 Salute-Corpus、基于条令和十年国防新闻的 Salute-Conv、领域偏好数据 Salute-Pref，以及评估条令和国防新闻理解推理的 Salute-Bench。Salute-LLM 依次进行持续预训练、监督微调和偏好对齐。摘要称其国防领域表现强，同时保持有竞争力的通用能力，但未给出具体分数。

Insight: 领域模型建设需要数据、偏好和评测配套，并检查领域增益是否以通用能力退化为代价；摘要描述端到端框架，但量化性能与条令来源的时间覆盖需阅读全文评估。

Abstract: Defense is a knowledge-intensive domain that requires precise understanding of specialized terminology, doctrinal concepts, operational procedures, and evolving military events. Although recent work has explored language technologies for military applications, existing efforts remain fragmented: they are often task-specific, rely on limited adaptation pipelines, or lack comprehensive defense-domain evaluation. In this paper, we present SALUTE, an end-to-end framework for benchmarking and adapting LLMs for the defense domain. SALUTE integrates Salute-Corpus, a curated corpus from open-access U.S. military doctrine and government documents; Salute-Conv, a grounded instruction dataset from doctrinal sources and decade-long defense news; Salute-Pref, a defense-aware preference dataset; and Salute-Bench, a rigorously filtered benchmark for evaluating defense-domain understanding and reasoning over doctrine and defense news. Based on these resources, we train Salute-LLM through multi-stage post-training with continual pretraining, supervised fine-tuning, and preference alignment. Extensive experiments show that Salute-LLM achieves strong defense-domain performance while retaining competitive general capabilities, demonstrating the effectiveness of SALUTE as an end-to-end framework for defense-domain LLM adaptation.

[152] ABSOL: Aggregated Bayesian Subsampling Orchestrated with LLMs cs.CL | cs.LGPDF

Jackson Hassell, Chen Shen, Estevam Hruschka

TL;DR: ABSOL 将大模型限制为贝叶斯网络结构学习的语义辅助信号，再与统计聚合结合，在更大规模的五个离散基准上提升边恢复质量。

Details

Motivation: 大模型可用自然语言处理结构化数据，但在一致证据条件、依赖关系推理和不确定性估计上仍不可靠；贝叶斯网络能显式建模概率关系，但大规模结构学习成本高且易受噪声影响。

Result: ABSOL 让大模型提供有边界的语义证据，辅助统计结构学习与聚合。在包含 27–1,041 个节点的五个离散贝叶斯网络基准中，GPT-5.4 版本是唯一在所有基准都生成可用图的方法，并在大于 27 节点的所有基准上取得最高 Edge F1。四种 LLM 增强相较无 LLM 聚合骨干平均提高 Edge F1 0.23；后处理实验表明，限制 LLM 对最终结构的决定权是收益的一部分。

Insight: 语言语义知识可增强统计结构学习，但应作为受限辅助而非直接决定网络结构；摘要结果依赖 GPT-5.4 版本，节点规模超过 27 的评测支持大图收益。

Abstract: Large language models are increasingly used as natural-language interfaces to structured data, yet they remain unreliable when answers require consistent evidence conditioning, dependency-aware reasoning, and uncertainty estimation. Bayesian networks provide an explicit probabilistic reasoning layer, but learning useful structures from data remains costly and fragile at scale. We introduce ABSOL, a hybrid LLM-guided Bayesian network structure-learning framework that uses LLMs as bounded semantic guides. Across five discrete BN benchmarks spanning 27 to 1041 nodes, ABSOL is the only evaluated method to produce a viable graph on every benchmark, and achieves the highest Edge F_1 on every benchmark larger than 27 nodes with GPT-5.4. The four LLM augmentations, which contribute complementary semantic evidence to the statistical backbone, improve Edge F_1 over the non-LLM aggregation backbone by +0.23 on average. Complementary post-hoc refinement experiments suggest that these gains depend in part on limiting the LLM’s authority over the final structure. Together, these results show that language-derived semantic knowledge can substantially improve scalable probabilistic structure learning when used as bounded guidance within a statistically grounded reasoning pipeline. The code for ABSOL is available at github.com/megagonlabs/absol-bn.

[153] Salesforce Koa: An Enterprise Language Model for Agentic Tool Use cs.CL | cs.AI | cs.LGPDF

Zixiang Chen, Sufeng Niu, Yingchi Liu, Wenting Zhao, Akshara Prabhakar

TL;DR: Salesforce Koa 基于开源 Nemotron-3-Super-120B，用规格驱动的仿真任务和工具使用奖励进行 GRPO 后训练，以增强企业多轮智能体能力。

Details

Motivation: 企业智能体需要可靠调用工具、处理依赖数据的工作流，同时维持通用能力；专用模型训练还需避免使用客户数据并有效生成任务与奖励信号。

Result: Koa 使用公开及合成数据，不用客户数据；以 Agent Script 工作流规格生成带 persona 的多轮任务，并依据工具成功完成情况构建奖励，再通过 GRPO 训练。公开工具使用、智能体推理和 CRM 基准上相较 Nemotron 基座有提升，最明显在多轮工具使用；超过一个强专有基线，但仍不及最强前沿模型。

Insight: 把可执行工作流规格转成仿真任务并以真实工具完成作为奖励，是适配开源模型的实用路径；摘要未给出提升数值，且与前沿模型仍存在能力差距。

Abstract: We present Salesforce Koa, an enterprise language model built by post-training the open-weight Nemotron-3-Super-120B foundation model with reinforcement learning using Group Relative Policy Optimization (GRPO). Salesforce Koa is trained on public and synthetically generated data, with no customer data, to improve tool use and agentic capabilities while preserving strong general-purpose performance. Its distinctive component is a simulation-to-reward pipeline that expands workflow specifications into persona-conditioned multi-turn tasks with task-resolution rewards grounded in successful tool use for data-dependent requests. For enterprise domains, these specifications are written in Agent Script, Salesforce’s declarative language for building Agentforce agents; for public tool-use domains, we synthesize the workflow structure directly. The same simulation and grounded-reward machinery drives GRPO across both. Across public tool-use, agentic-reasoning, and enterprise Customer Relationship Management (CRM) benchmarks, Salesforce Koa improves over its open-weight base, with the clearest gains on multi-turn tool use, and surpasses a strong proprietary baseline while remaining below the strongest frontier models. These results show that specification-driven reinforcement learning is a practical path to specializing open-weight foundation models for enterprise agentic tasks.

[154] DA-DLM: Explicitly Modeling Token Dependencies in Diffusion Language Models cs.CLPDF

Pengyu Ji, Zichen Zhang, Xiang Hu, Kewei Tu

TL;DR: DA-DLM 将定向无环图式 token 依赖引入扩散语言模型，使早期去噪中已固定 token 成为邻近预测的锚点，并随生成逐步重分配不确定性。

Details

Motivation: 扩散语言模型每步并行独立预测多个 token，条件独立假设会损害连贯性；这与非自回归翻译的多模态问题类似，需要显式建模 token 间依赖。

Result: DA-DLM 采用面向位置的 DAG，节点组绑定固定输出位置，较早确定的 token 通过学习转移关系约束相邻预测；随着去噪进展和锚点累积，图结构适应剩余不确定性。在语言建模、开放式生成和摘要任务中持续优于 Block Diffusion，少步去噪时优势尤其明显，并达到自回归模型相当表现，同时保留并行生成能力。

Insight: 把已确定 token 作为结构锚点能在并行去噪中恢复依赖关系，减少独立预测导致的不连贯；其优势在少步设置突出，摘要未提供延迟与不同任务的绝对质量数据。

Abstract: Diffusion Language Models (DLMs) generate text by iteratively denoising a masked sequence, independently predicting multiple tokens at each step. This conditional independence discards inter-token dependencies and degrades coherence-an issue that parallels the multi-modality problem in Non-Autoregressive Translation (NAT). Drawing on the Directed Acyclic Transformer (DAT), which tackles this problem in NAT via a Directed Acyclic Graph (DAG), we propose DA-DLM, a model that adapts DAG-based dependency modeling to DLMs’ iterative setting through a position-oriented DAG design. The position-oriented DAG binds node groups to fixed output positions so that tokens fixed in earlier steps anchor neighboring predictions via learned transitions, and evolves with denoising to focus on remaining uncertainty as anchors accumulate. On language modeling, open-ended generation, and summarization, DA-DLM consistently outperforms Block Diffusion, especially under fewer denoising steps, and matches autoregressive models while preserving the parallel generation advantage. Our code is publicly available at https://github.com/jipy0222/DA-DLM.

[155] Improving Mathematical Reasoning Capabilities in Large Language Models via Reasoning Process Error Classification cs.CLPDF

Runa Yoshida, Kosuke Nishida, Kyosuke Nishida

TL;DR: 作者将数学推理过程错误人工归纳为 21 类，并据此设计关注高频错误类型的提示，实验显示该提示可提升大模型数学推理表现。

Details

Motivation: 只根据最终答案对错无法解释数学推理模型为何失败，也难以针对性改善；作者希望系统归纳错误过程特征并将诊断结果反馈给提示设计。

Result: 研究分析数学数据集上答错的问题，人工定义并分类 21 种推理过程错误，找出高频类别；随后设计明确针对其中八类错误的提示。实验显示新提示有效提升推理成绩，且识别出的常见错误在相近规模的大模型间具有共性。摘要未给出具体模型和准确率提升数值。

Insight: 过程级错误分类可以把抽象的推理失败转化为提示可针对的行为模式；人工分析覆盖的是模型答错样例，摘要未说明分类一致性和提示对未见任务的迁移效果。

Abstract: The reasoning ability of large language models (LLMs) is a critical factor for practical LLM-based applications. To investigate the current reasoning capability of LLMs, we clarify the types of errors that arise in LLMs’ reasoning processes on mathematical datasets. We focus on problems where LLMs produce an incorrect answer. We define errors in the reasoning process as reasoning errors and manually analyze the features of reasoning errors. We defined and classified 21 error classes and identified the frequently occurring classes among them. Beyond qualitative evaluation, we leverage the evaluation results to improve the reasoning capability. We designed a prompt that explicitly focuses on eight error classes. The experiments demonstrate that this prompt effectively improves reasoning performance. Furthermore, the results suggest that the frequent reasoning errors identified in this paper are common across LLMs of comparable scale.

[156] EMR: Self-Evolving Medical Multi-Agent System via Experience Mining and Reuse cs.CL | cs.AIPDF

Dongsheng Shi, Yue Li, Xin Yi, Linlin Wang

TL;DR: EMR 是通过经验挖掘与复用实现自演化的医疗多智能体系统，将成功诊断经验和失败警示写入分层临床经验库以指导后续病例。

Details

Motivation: 现有医疗多智能体系统通常依赖固定推理策略，缺乏持久临床记忆，无法从既往诊断成功与失败中持续改进。

Result: EMR 将经验整理为临床原则、诊断模式和代表性案例三层；推理时由规划代理协调专科代理会诊，再由总结代理形成结论。系统自动从协作轨迹中提取正确诊断洞见和失败警示，并逐步更新经验库。摘要报告在医疗推理基准上持续优于先进多智能体基线，层级经验支持跨专科泛化及不同 LLM 骨干迁移；本地原始摘要末句被截断，无法确认其后续具体结论。

Insight: 把推理成功经验与失败警示转化为可复用分层记忆，可能使医疗多智能体系统随病例积累而改进；由于摘要文本尾部截断且未报告具体分数，结论仅按现有内容概括。

Abstract: Large language model (LLM) driven multi-agent systems have shown promise in complex clinical reasoning, yet existing approaches rely on static strategies and lack persistent clinical memory, preventing self-evolving from prior diagnostic successes and failures. We present EMR, a self-evolving medical multi-agent system via Experience Mining and Reuse. EMR introduces a hierarchical clinical experience library that organizes accumulated knowledge into three levels: clinical principles, diagnostic patterns, and representative cases. During inference, EMR emulates multidisciplinary consultation: a planner agent coordinates domain-specific department agents for specialized reasoning, while a summary agent synthesizes their analyses into a final decision. Critically, EMR automatically extracts correct diagnostic insights and failure-related warnings from multi-agent reasoning trajectories, incrementally updating the experience library to guide future cases. Experiments on medical reasoning benchmarks demonstrate that EMR consistently outperforms state-of-the-art medical multi-agent baselines. Further analysis reveals that the hierarchical experience enables cross-specialty generalization and transfer across diverse LLM backbones, offering a scalable and in

[157] MUSE: A Theory-Harnessed Story Engine for Vibe Narrativizing cs.CLPDF

Jianxiang Ma, Xiaocui Yang, Daling Wang, Yuesong Hou, Mingfu Zhang

TL;DR: MUSE 是理论驱动的故事生成代理框架，将结构、角色、场景和修订决策组织为可传递的中间产物，改善多种模型的长篇叙事质量和一致性。

Details

Motivation: 大模型能生成流畅文字，但故事质量取决于规划、起草和修订中情节、角色与语言选择是否协调；创作指导既要有质量，又要能持续贯彻。

Result: MUSE 将 McKee 故事理论拆解为规则、语义合并并抽象机制，以单一事实源和分层披露组织指导，再用中间交付物让设计、角色表现、场景构成与修订共享决策。在四个基础模型上，WritingBench 较零样本提升 1.6–4.8 分，三个模型的 LongStoryEval 提升超过 10 分；ConStory-Bench 一致性错误密度维持个位数，并在三个模型上低于复现的故事系统基线。消融显示结构设计贡献最大，角色路径影响声音表达，修订还能继续改善。

Insight: 将创作理论转成角色相关、可持续传递的决策和交付物，有助于长篇叙事保持结构一致，而不是只依赖一次性提示；理论规则化和基准选择会影响结论可迁移性。

Abstract: LLMs can generate fluent prose. Story quality depends on how decisions about plot, character, and language work together across planning, drafting, and revision. Guiding these decisions presents two bottlenecks: the quality of story guidance and its sustained use. We formulate Vibe Narrativizing as the task of turning natural-language writing requirements into a finished story and present MUSE, a Theory-Harnessed Story Engine. MUSE organizes story knowledge as guidance for specific decisions and carries those decisions into subsequent creative work. Knowledge engineering develops Robert McKee’s story theory through rule atomization, semantic consolidation, and mechanism abstraction; a single source of truth and layered disclosure organize the resulting guidance. Typical examples complement principles that depend on context and aesthetic judgment. An agent harness organizes design, character performance, scene composition, and revision through intermediate deliverables that preserve story decisions. Context engineering supplies each role with the relevant guidance and decisions, while a masterwork corpus provides inspiration and prose references. A worked example follows one requested object from its thematic role to the characters’ climactic actions. Across four base models, MUSE improves WritingBench by 1.6-4.8 points over zero-shot generation and raises LongStoryEval by more than ten points on three. ConStory-Bench consistency error density remains in the low single digits for all four models, below every reproduced story-system baseline on three. Component ablations locate the largest quality contribution in structural design, voice-specific effects in the character path, and further gains in revision. Code is available at https://github.com/RoadtoAGI/MUSE.

[158] Semiotic Relations and Proof Methods: A Cross-Genre Study of Argument Structure with Large Language Models cs.CLPDF

Edirlei Soares de Lima, Marco A. Casanova, Antonio L. Furtado

TL;DR: 论文提出用四类符号关系连接论证结构与证明方法，并以数学、法律和日常语料的 LLM 标注分析不同文体的论证路径。

Details

Motivation: 当命题难以直接证明时，常需转换到相关命题再论证；作者希望形式化从原命题到替代命题的关系，并比较这些转换在不同论证文体中的使用情况。

Result: 框架将句法关系、范式关系、对立关系和部分—整体关系分别对应转喻、隐喻、反讽和提喻，并联系到推理证明、类比证明、反证法和分类讨论。作者把四类关系操作化后，使用多个大模型分析数学、法律和日常论证语料。结果显示数学证明使用全部四类关系，而法律与日常推理几乎完全依赖推理关系。

Insight: 证明策略可视为通过不同符号关系从原命题迁移到相关命题；文体间分布差异表明数学论证结构更丰富，结论依赖 LLM 对关系类别的标注可靠性。

Abstract: When a direct proof of a statement $S$ seems hard or even impossible to obtain, there may exist another statement (or set of statements) $S^{}$, somehow related to $S$, on the basis of which $S$ can be proved. In order to investigate what options can be used to move from $S$ to $S^{}$, four kinds of semiotic relations inspired by the four master tropes of semiotic research are briefly reviewed. Specifically, our syntagmatic, paradigmatic, antithetic and meronymic relations correspond, respectively, to metonymy, metaphor, irony and synecdoche. It is suggested that these four semiotic relations determine the options to move from $S$ to $S^{*}$, leading to proof by inference, proof by analogy, proof by contradiction, and proof by case analysis. To examine how the four relations are actually used across different kinds of argument, we complement the framework with an empirical study. We turn the four relations into explicit operational definitions and apply them to a cross-genre corpus of mathematical, legal, and everyday argument using a panel of large language models. We find that the relations are used very unevenly across genres: mathematical proofs draw on all four, whereas legal and everyday reasoning rely almost entirely on inference.

[159] Don’t Count the Edits, Judge by the Outcome Alone: Reward-Based Evaluation for Grammatical Error Correction cs.CL | cs.AIPDF

Hayeong Ryu, Sunhee Jo, Seunguk Yu, YoungBin Kim

TL;DR: SURE 是源文本条件的语法纠错奖励评估器，以同源不同改写偏好学习语法性、忠实度和流畅度，并对纠正片段提供来源锚定。

Details

Motivation: 语法纠错常以参考文本或编辑重合度评分，会惩罚与标准答案不同但有效的改写；无参考指标则难判断流畅输出是否确实修正了原文错误。

Result: SURE 使用同一源文本上的最小编辑与改写式纠错偏好联合学习总体奖励，并增加语法性、忠实度、流畅度分项监督及源侧错误解决片段的落地信息。在 SEEDA 上与强基线表现有竞争力，特别提升改写式纠错评价能力，并提供更可分离的准则级诊断；摘要未给出具体指标。

Insight: 语法纠错评价需要同时考虑句子可接受度和对原始含义的忠实性，来源片段对齐可帮助判断修改是否解决原错；评测效果目前报告于 SEEDA，外部语种迁移未说明。

Abstract: Grammatical error correction (GEC) evaluation has traditionally relied on reference or edit overlap, which can penalize valid rewrites that differ from gold corrections. Reference-free metrics reduce this dependence, but evaluating whether a fluent output is a valid correction of the source remains challenging. We propose SURE, a source-conditioned reward evaluator trained on within-source preferences spanning minimal-edit and rewrite-oriented corrections. SURE jointly learns an overall reward with criteria-level supervision for grammaticality, faithfulness, and fluency, together with span-level grounding for source-side error resolution. Experiments on SEEDA show that SURE performs competitively against strong baselines, with particular gains on rewrite-style corrections and more disentangled criteria-level diagnostics. Our code is available at https://github.com/hayeonggg/SURE.

[160] Through the Eyes of the Beholder: Biometric and Demographic Conditioning for Multimodal Sexism Detection cs.CL | cs.AI | cs.CV | cs.LGPDF

Ana-Maria Luisa Mocanu, Sebastian Mocanu, Ciprian-Octavian Truică, Elena-Simona Apostol

TL;DR: VANGUARD 在 EXIST 2026 性别歧视检测中把标注者心理与人口特征作为条件输入，并融合文本、图像、传感器和风格特征。

Details

Motivation: 网络性别歧视判断具有主观性，忽略标注者差异可能把不同群体的意见压成单一标签；作者探索以标注者完整标签分布和属性改善多模态识别。

Result: 框架用跨注意力融合五类输入并以 FiLM 条件化；Gemma 4 提取梗图文字和视觉描述，NLLB-200 做英西互译，LoRA 适配的 XLM-R 与 CLIP 编码文本和图像，预训练自编码器编码传感器信息。Subtask 2.1 以 KL 散度拟合标注者标签分布，推理时将多模态网络和风格/生理特征 SVM 软投票。最佳提交在来源意图子任务软评估排名 114 队中的第 29；两个子任务的归一化 ICM 均超过基线。

Insight: 把分歧视为完整标签分布并条件化标注者特征，提供以人为中心的多模态建模方式；排名和超基线结果说明存在可用信号，但无法单独证明人口特征提升了公平性或跨人群泛化。

Abstract: Detecting sexism on the internet is a fundamentally subjective task; our team, VANGUARD, addresses this challenge in the EXIST 2026 Task 2 by proposing a human-centered multimodal framework that analyses and incorporates the psychological and demographic characteristics of human annotators into the detection pipeline. We fuse five input modalities through a cross-attention architecture with Feature-wise Linear Modulation conditioning. Meme text is extracted and visually described with Gemma 4, then augmented by automatic translation between English and Spanish with NLLB-200. Text and image representations are produced by LoRAadapted XLM-RoBERTa and CLIP encoders and fused with sensor features encoded by a pretrained autoencoder. To model annotator subjectivity, we frame Subtask 2.1 as a label distribution learning problem, optimizing a Kullback-Leibler divergence loss over the full annotator label distribution. At inference time, predictions are produced by soft-voting between the deep multimodal network and a complementary SVM trained on stylometric and physiological features. Our best submission ranks 29th out of 114 on Subtask 2.2 (source intention) under soft evaluation, and the normalized ICM scores remain above the baseline on Subtasks 2.1 and 2.2, indicating that annotator-centered conditioning contributes a usable signal. We release our full pipeline and analysis to support reproducible human-centered modeling.

[161] Before You Poll with LLMs: A Deliberative Diagnostic Framework cs.CL | cs.AI | cs.CYPDF

Ahmed Wali, Hassaan Tayyab

TL;DR: 审议式民意诊断框架比较人类和大模型在接受相同新信息后的信念变化，发现五种前沿模型都存在反转、过度调整或僵化等问题。

Details

Motivation: 模拟公众意见的 LLM 研究常只检验人物设定是否持有合理初始意见，却很少考察它们能否像人类一样根据论据更新信念。

Result: 框架让人类与 LLM persona 接受相同信息干预，使用 America in One Room 的 526 个 persona、72 个问题评估五个前沿模型。GPT-5.1 在平衡信息后对反对党更敌对，80% 群外问题出现反转，而政策问题为 26%，且党派方向对称；Gemini 2.0 Flash、Claude Sonnet 4.5 和 Llama 3.3 70B 虽方向正确但调整幅度为人类 5–7 倍；DeepSeek V3 几乎不变。消融显示政策内容触发身份特异失败，作者称之为对模型自身 persona 刻板印象的自我迎合。

Insight: 初始立场看似合理不能保证模型会根据新证据正确更新，动态信念变化需独立评测；框架提供可复现诊断协议，作者建议在将 LLM persona 用于民调前先测试。

Abstract: Can LLMs reason through new information like humans, or do they merely retrieve cached opinions? This is critical for silicon sampling, where LLM personas simulate public opinion at scale. Current evaluations test only whether personas hold the right opinions – a static snapshot. But opinion research increasingly depends on dynamic fidelity: whether personas update beliefs in response to new arguments, as humans do during deliberation. No existing benchmark tests this. We introduce the Deliberative Polling Diagnostic Framework, which compares human and LLM belief shifts after identical informational interventions. Grounded in deliberative polling, it surfaces failures invisible to static evaluation: models that produce plausible partisan opinions can still misrepresent how those opinions change. Applying the framework to five frontier models using data from America in One Room (526 personas, 72 questions), we find that every model fails, each in a unique manner. GPT-5.1 exhibits reversal: its personas become more hostile toward the opposing party after balanced information, while humans become less so. This reversal is selective (80% on outgroup vs. 26% on policy questions) and symmetric across partisan identities. Gemini 2.0 Flash, Claude Sonnet 4.5, and Llama 3.3 70B exhibit overshoot, shifting correctly but at 5-7x human magnitude. DeepSeek V3 exhibits rigidity with near-zero change. Targeted ablations reveal that policy content triggers these failures and that they are identity-specific: GPT-5.1 reverses on outgroup questions but overshoots on ingroup; Gemini shows the inverse. We term this signature self-sycophancy: conformity to the model’s internal stereotype of the persona rather than reasoning from the information provided. Our framework offers a concrete protocol: run the deliberative diagnostic before trusting LLM personas to mimic revised beliefs.

[162] Learning to Coach for Experiential Learning cs.CLPDF

Guanheng Chen, Tianzhu Ye, Li Dong, Xun Wu, Shaohan Huang

TL;DR: Learning to Coach 训练专用教练模型从冻结行动者的历史轨迹提炼可操作经验，并依据被辅导回答的正确性优化这些指导信息。

Details

Motivation: 模型可从经验学习，但原始解题轨迹冗长且噪声多，不能直接作为高效指导；需要压缩成可迁移的教练知识。

Result: L2C 让教练 LLM 从行动者过去的轨迹中提取经验，行动者本身保持冻结；教练以引导后回答正确性获得奖励，分别学习改善原题后续回答的同实例奖励和促进跨题迁移的跨实例奖励。在数学推理和交互文字游戏中，L2C 持续优于自我修订和未训练教练。增加经验学习迭代可进一步提升准确率，并比扩大行动者解码预算更有效地使用额外推理计算。训练后的教练还能迁移到分布外任务，并针对被辅导行动者调整指导。

Insight: 把经验提炼与答题执行分开，可通过反馈学习更精炼、可迁移的指导；跨行动者适配是优势，摘要未提供不同任务和奖励设计的具体增益。

Abstract: Language models can learn from experience, but raw solution trajectories are often too long and noisy to provide effective guidance. In this work, we propose Learning to Coach (L2C), a framework that trains a dedicated LLM-as-a-Coach to extract actionable experiential knowledge from an actor model’s previous trajectory. The actor remains frozen, while the LLM-as-a-Coach is trained to maximize a reward given by the correctness of the actor’s guided response. We study two such rewards: a same-instance reward, which improves subsequent responses on the original problem, and a cross-instance reward, which elicits knowledge that transfers to other instances. Across mathematical reasoning and interactive text-games, L2C consistently outperforms self-refinement and an untrained LLM-as-a-Coach. Running experiential learning for more iterations further improves accuracy and uses additional inference compute more effectively than enlarging the actor’s decoding budget. The trained LLM-as-a-Coach also transfers to out-of-distribution tasks and adapts its guidance to the specific actor it coaches.

[163] K-Bench: a clinically calibrated benchmark for evaluating large language models in high-risk mental health conversations cs.CL | cs.AI | cs.LGPDF

Laura M. Vowels, Matthew J. Vowels, Shivali Sharma, Apoorv Jha, Rehnuma Choudhury

TL;DR: K-Bench 是临床校准的高风险心理健康多轮对话基准，涵盖自杀、自伤、家暴、物质滥用和无风险场景，用于比较不同模型配置的安全表现。

Details

Motivation: 人们越来越多用大模型寻求心理支持，但模型在风险随多轮对话变化时的安全性尚未充分描述，单轮或静态基准不足以评估临床风险。

Result: 基准包含 200 个多轮情景，评测来自 14 家提供方的 33 个基础模型、共 125 种配置；合成对话与真实人机心理对话有显著分布重叠。冻结 GPT-4o 评审在 151 份临床评分转录的 6,751 项有效比较中与临床共识精确一致率 94.2%。领先模型能提供支持性对话，但综合风险分数仍高于 95；低分配置在风险探索方面差异明显。治疗式提示主要帮助较弱配置，增加推理未带来平均改善。公开排行榜持续更新，但运营测试材料受保护。

Insight: 心理健康安全需在多轮情景下同时衡量支持性与风险识别，且配置级差异明显；自动评审虽与临床共识高度一致，基准仍是模型比较工具，不能代替临床安全保障。

Abstract: % !TEX root = ../main.tex People increasingly use large language models (LLMs) for mental health support, yet their safety in evolving, high-risk conversations remains poorly characterised. We developed K-Bench, a clinician-calibrated, protected benchmark evaluating 125 model configurations representing 33 base models from 14 providers across a fixed cohort of 200 multi-turn vignettes involving suicide, self-harm, domestic violence, substance misuse, and no-risk presentations. Synthetic patient conversations showed substantial distributional overlap with real human-AI conversations. A frozen GPT-4o judge achieved 94.2% exact agreement with clinician consensus across 6,751 eligible item comparisons from 151 clinician-rated transcripts. Leading models combined strong supportive conversation with combined-risk scores above 95, whereas risk exploration exposed substantial variation among lower-performing configurations. Therapeutic prompting produced configuration-specific gains concentrated among weaker models, while elevated reasoning produced no average improvement. K-Bench combines broader clinical coverage and configuration-scale comparison with a continuously updated public leaderboard whose operational test materials are protected from direct optimisation. The leaderboard is available at www.k-bench.ai.

[164] Inoculation Midtraining with Learned Neologisms cs.CLPDF

Kyle O’Brien, Edward James Young, Puria Radmard, Nathalie Kirch, Cameron Tice

TL;DR: Inoculation Midtraining 先教模型把不安全行为与新引入的隔离 token 关联，再在该上下文内做安全数据后训练，以尝试减少不良行为向普通上下文泛化。

Details

Motivation: 后训练可能同时学习期望和不期望行为；作者研究在较早的中训练阶段先建立情境边界，能否影响后来不安全行为的选择性泛化。

Result: 模型在中训练时学习不安全行为属于上下文，之后在该上下文下用不安全数据做监督微调或强化学习，再于系统提示中移除该 token 评估。方法能降低失配行为并保留德语表达、莎士比亚风格等良性属性迁移，但不优于标准 Inoculation Prompting；效果对训练配置敏感，邻近上下文线索仍会重新激活不安全行为。

Insight: 学习得到的上下文关联可塑造某些行为的选择性泛化，但边界会泄漏且提示基线更强，尚不能作为安全框架的核心防线；还需要系统检验触发条件和鲁棒性。

Abstract: Large language models (LLMs) often learn both desirable and undesirable properties during post-training. We study whether midtraining, an earlier training stage, can shape which of these properties later generalise. We introduce Inoculation Midtraining, a technique that teaches a base model that unsafe behaviour belongs to a designated context, as indicated by the neologism (a new token) introduced during midtraining, and then post-trains the model on unsafe data within that context. We then evaluate the model outside the context, with the neologism excluded from the system prompt. Across supervised fine-tuning and reinforcement learning post-training regimes, we find that Inoculation Midtraining can reduce misalignment while preserving the transfer of benign data properties (e.g., speaking in German or Shakespearean prose). However, our approach does not outperform standard Inoculation Prompting, is sensitive to training configuration, and produces a leaky boundary that nearby contextual cues can reactivate. These results show that inoculation with a learned association introduced via midtraining can shape selective generalisation. Still, more work is needed before this approach can become a load-bearing component in a developer’s safety framework.

[165] HypoEvolve: Genetic Algorithms Enable Multi-Agent LLMs to Discover Scientific Hypotheses cs.CL | cs.CE | cs.MA | cs.NEPDF

Jieyuan Liu, Mengzhou Hu, Jefferson Chen, JungHo Kong, Pratibha Jagannatha

TL;DR: HypoEvolve 用世代式遗传算法显式协调专门化大模型代理提出和筛选科学假设，在药物再利用任务中提升了外部证据支持的靶点级假设质量。

Details

Motivation: 科学代理可综合证据并提出解释，但不同协作方式如何影响假设质量仍不清楚；评估还需将机制性假设与可验证的外部生物证据关联起来。

Result: HypoEvolve 维护假设种群，由专门代理整合机制论证、重审假设、评估证据与可检验性，通过每一代规则更新候选。作者将 DepMap 与 Open Targets 适配为分别基于实验、遗传和临床证据的互补测量。在 34 个癌种上，该方法两项指标均超过六个基线；DepMap 选择性得分为 0.171，最强基线 0.115。相较单次生成的增益也迁移到留出癌种。

Insight: 协作应显式规定如何保留、修订和淘汰假设，才能把多代理差异转成可测的发现质量；研究聚焦药物再利用中的靶点级外部指标，仍非湿实验发现验证。

Abstract: Scientific agents contribute to hypothesis discovery by synthesizing evidence, assessing proposals, and developing new explanations. Recent systems combine scientific agents with evolutionary search through critique, comparison, and revision. However, how different forms of agent collaboration affect hypothesis quality remains an open question. Answering this question requires separating the effects of agents’ scientific capabilities from those of their collaboration. A framework must therefore preserve agents’ scientific roles and support rules for combining, revising, and retaining hypotheses. Building on this view, we introduce HypoEvolve, which makes collaboration explicit through successive updates to a hypothesis population. Specifically, we propose a generational genetic algorithm to coordinate specialized large language model (LLM) agents that integrate mechanistic arguments, reconsider assumptions, and assess evidence and testability. Each generation specifies how scientific judgments and new proposals reshape the population, making collaboration effects on hypothesis quality directly testable. Moreover, we design our evaluation around scientifically meaningful hypotheses that explain how a proposed intervention could work. Drug repurposing links these explanations to target-level biological claims assessed against external evidence. Specifically, we adapt DepMap and Open Targets into complementary external measures grounded in experimental, genetic, and clinical evidence. Across 34 cancer types, HypoEvolve achieves the highest scores against six baselines on both measures. DepMap selectivity reaches 0.171, versus 0.115 for the strongest baseline. Gains over single-pass generation also generalize to held-out cancer types. HypoEvolve advances a vision of autonomous science in which AI research teams achieve a capacity for discovery beyond that of individual models.

[166] Mind2Dialogue: Training Human-Aware Language Models by Simulating User Mental States cs.CL | cs.LGPDF

Zixuan Wang, Yufan Zhou, Jinzhou Tang, Xinle Yu, Chengjun Wu

TL;DR: Mind2Dialogue 通过心理学引导的用户状态模拟生成“隐含信念与目标”监督，再用特权蒸馏训练更懂用户偏好的对话模型。

Details

Motivation: 助手训练数据很少包含基于用户未明说信念和目标的知情回应，而真实心理状态不可直接观察，限制了人本、个性化助手的监督扩展。

Result: 框架模拟具有稳定个人特征且会随互动演化的用户心理状态，以共同变化的状态驱动用户行为和 Oracle 助手回应，再用特权蒸馏训练部署时无需直接访问用户状态的模型。全量语料训练在 Qwen、Llama 和 OLMo 指令基线的所有个性化指标上均有提升，其中偏好遵循生成提高 26.6–40.9 个百分点；Qwen 和 Llama 的信念及行动推理也改善。

Insight: 显式模拟未言明的心理状态可为助手提供现实数据中难获得的监督，并改善个性化和心智理论能力；结果依赖模拟状态与 Oracle 生成质量，需验证与真实用户长期目标的一致性。

Abstract: As language models become more capable, long-term collaboration in learning, reasoning, and decision-making calls for a deeper understanding of the people they serve. Yet training such human-aware language models faces a fundamental supervision gap because current datasets for LLM assistant training contain few if any well-informed responses explicitly grounded in users’ unspoken beliefs and goals. Scaling such supervision is inherently constrained, as users’ underlying states are not directly observable. We thus propose the Mind2Dialogue framework to mitigate this gap by simulating users’ mental states and turning them into privileged supervision for human-aware training. Specifically, we first propose a psychology-guided simulator that preserves personal characteristics while updating mental states through interaction to generate coherent conversations. The key idea is to enforce a shared evolving mental state that drives user behavior and guides an Oracle assistant’s responses. Our privileged distillation then trains models on the Oracle’s well-informed responses to assist users without direct access to their mental states at deployment. Moreover, we propose to evaluate human-aware learning by combining personalization and theory of mind, examining how models understand people and act on that understanding. Training on the full Mind2Dialogue corpus improves every reported personalization metric over the corresponding Qwen, Llama, and OLMo instruction-tuned baselines, including gains of 26.6 to 40.9 percentage points in preference-following generation. The gains extend to belief and action reasoning on Qwen and Llama, beyond personalized assistance. Looking forward, Mind2Dialogue makes user simulation a foundation for genuine AI collaborators that understand beliefs and intentions behind people’s words and support their long-term goals across education, work, and everyday life.

[167] Discovery Foundation Models: Toward Open-Ended Discovery Intelligence cs.CLPDF

Ling Yang, Zhenfei Yin, Yingcheng Wu

TL;DR: 论文提出 Discovery Foundation Models，将基础模型目标从解决人类给定问题扩展到参与问题发现、假设、实验和知识结构修订，并以 Zetema 和 GALILEO 描述实现框架。

Details

Motivation: 基础模型已能学习既有知识并通过行动和反馈改进，但科学发现还要求创造新问题、表征和解释，当前系统对此类开放式发现能力缺少统一定义与过程评估。

Result: 作者将 Discovery Intelligence 定义为开放式发现能力，要求系统在可修订研究状态上支持问题发现、形式化、表征构建、假设形成、干预、证据驱动修订和持续改进七项耦合能力。Zetema 将研究状态动态、验证门控、外部证据和技能演化结合；GALILEO 描述连接干实验推理、机器人及人工湿实验和迭代假设修订的真实治疗发现闭环，并提出以能力形成和过程为中心的评估方式。摘要主要提出框架，未提供实验指标。

Insight: 发现智能不仅是输出正确答案，还需参与问题构造、实验执行和证据修订；GALILEO 将推理与物理实验闭环连接，但该摘要侧重愿景和系统设计，未给出临床发现效益的定量证据。

Abstract: Foundation models have progressed from learning and reasoning over existing knowledge, to increasingly learning through action, tool use, and outcome feedback. We argue that the next frontier is a further transition: from solving and acting within problems specified by humans to participating in the process by which new problems, representations, explanations, and knowledge are created. We refer to this capability as Discovery Intelligence. We formulate Discovery Foundation Models (DFMs) as general-purpose model systems for open-ended discovery. A DFM operates over a revisable research state and supports seven coupled capabilities spanning problem discovery, formulation, representation construction, hypothesis formation, intervention, evidence-grounded revision, and continual discovery improvement. We instantiate this framework with Zetema, which couples explicit research-state dynamics, verification and experimental gating, external grounding, and cross-task Discovery Skill evolution. We further ground the framework with GALILEO, a real therapeutic-discovery system in which Dry-Lab reasoning, robotic and hands-on Wet-Lab experimentation, external biological evidence, and iterative hypothesis and design revision form a closed physical discovery loop. We then formulate a unified approach to capability formation and process-centered evaluation, enabling discovery behavior to be trained, improved, and measured beyond final-answer performance. Together, these components establish discovery as a learnable, executable, and evaluable capability of foundation-model systems. We view this shift as a broader progression in intelligence scaling: from learning over existing knowledge, to learning from action outcomes, and ultimately to participating in the construction, testing, and revision of the structures through which new knowledge is discovered. Code: https://github.com/Gen-Verse/DFM-Plans

cs.AI [Back]

[168] FedV-KGQA in Practice: Design Lessons and an Interactive Prototype cs.AI | cs.CL | cs.IRPDF

Md Saikat Islam Khan Bappy, Oshani Seneviratne

TL;DR: FedV-KGQA 将关系类型分布在不同组织的知识图谱做垂直联邦多跳问答，报告融合可恢复大部分集中式准确率且无需共享原始三元组或关系嵌入。

Details

Motivation: 现实组织往往共享实体标识但各自拥有不同关系类型，没有单一方掌握完整推理链；常规知识图谱问答假设系统能访问完整图。

Result: 各数据孤岛扩充本地图并只用自身三元组训练知识图谱嵌入，服务器拼接实体视图、将问题投影锚定在主题实体，再按相似度排序候选；原始三元组和关系嵌入不离开本地。跨实验比较显示：联邦融合恢复集中式准确率的大部分，单一孤岛表现较弱；锚定与图扩充比嵌入模型选择更重要；最便宜编码器取决于目标准确率而非参数规模。论文还提供可交互原型，可逐题追踪完整推理流程。

Insight: 垂直联邦融合能够在不共享核心关系数据的情况下组合跨机构推理链；设计重点是实体锚定与本地图丰富，系统成本—精度选择需按目标服务水平确定。

Abstract: Knowledge graph question answering usually assumes that one system can reach the whole graph. In practice, facts are often held by organizations that share entity identifiers but own disjoint relation types, so no single party sees a complete reasoning chain. This poster presents the empirical findings of FedV-KGQA on multi-hop question answering over such vertically partitioned graphs. Each silo enriches its local graph and trains a knowledge graph embedding on its own triples. A server then concatenates the silo-specific entity views, anchors the projected question at the topic entity, and ranks candidates by similarity. Raw triples and relation embeddings never leave a silo. Comparing the FedV-KGQA experiments with one another yields three results. First, federated fusion recovers most of the centralized accuracy, while a single silo recovers little. Second, anchoring and enrichment matter more than the choice of embedding model. Third, the cheapest encoder depends on the target accuracy rather than on parameter count. This poster paper contributes that cross-experiment comparison, four design lessons drawn from it, and an interactive prototype that runs real inference and traces the full pipeline, per question, on released checkpoints.

[169] VeriDx: Earning the Right to Diagnose with Disease-Centric Verification cs.AI | cs.CLPDF

Zhong Cao, Shuying Chen

TL;DR: VeriDx 将每个疾病假设对应的临床义务结构化追踪，检查关键证据、鉴别诊断、矛盾处理、必要检查和诊断闭环依据。

Details

Motivation: 医学大模型评价通常看最终答案、局部推理步骤或孤立事实，可能忽略推理过程早先提出某病后应承担的证据验证责任；即使结论碰巧正确，也可能理由错误。

Result: VeriDx 将自由形式诊断推理映射到疾病档案，追踪每个假设是否满足、未解决或违反临床义务，从而显露遗漏关键检查、鉴别诊断未决、忽视矛盾、无依据主张及过早收敛等失效。系统以指南派生疾病档案和专家标注的纵向复杂呼吸病例实例化。结果显示，许多诊断错误不是孤立失误，而是推理早期提出却未履行的承诺。

Insight: 诊断质量评估应追踪假设产生后的证据义务，而非只看最终诊断是否正确；该框架当前聚焦复杂呼吸诊断，扩展到其他专科需构建相应专家档案。

Abstract: A correct diagnosis can still be reached for the wrong reasons. In clinical reasoning, every disease hypothesis creates obligations: key evidence must be checked, alternatives must be ruled out, contradictions must be resolved, useful tests must be considered, and closure must be justified. Current evaluations of medical LLMs mostly focus on final answers, local steps, or isolated facts, and therefore miss these hypothesis-induced commitments. We introduce \textbf{VeriDx}, a disease-centric verification framework that links free-form diagnostic reasoning to structured disease profiles. VeriDx tracks whether each hypothesis is satisfied, unresolved, or violated its clinical obligations, exposing failures such as missing critical tests, unresolved differentials, ignored contradictions, unsupported claims, and premature closure. We instantiate VeriDx for complex respiratory diagnosis using guideline-derived disease profiles and expert-annotated longitudinal cases. Our results show that many diagnostic errors are not isolated mistakes, but broken commitments made earlier in the reasoning process.

[170] Question’s Gambit: The First Move Matters in Agentic Deep Search cs.AI | cs.CL | cs.IRPDF

Radin Hamidi Rad, Amin Bigdeli, Negar Arabzadeh, Sajad Ebrahimi, Charles L. A. Clarke

TL;DR: Question’s Gambit 在智能体深度搜索开始前执行线索拆分、互补查询、结果聚合和重排，改善开局证据上下文并提升 BrowseComp-Plus 问答表现。

Details

Motivation: 深度研究代理虽能循环搜索、阅读和推理，仍可能找到高质量证据却无法连接到黄金文档；首次检索动作会影响后续搜索空间和证据链。

Result: 模块将复杂问题分解为线索、生成互补搜索、汇总检索结果并对候选池重排，形成支持线索聚合与最终验证的初始上下文，再交给迭代代理。在 BrowseComp-Plus 上，Question’s Gambit 相较强基线提高召回和下游准确率；结合 GPT-5.5 时，回答准确率由 Pi-Serini 基线 83.1% 升至 90.5%。另在 MultiHop-RAG 评估向更传统多跳问题结构的迁移。

Insight: 深度搜索不只取决于循环中的工具能力，开局检索是否覆盖问题线索同样重要；基准显示初始证据组织可大幅改善结果，跨任务迁移程度需看 MultiHop-RAG 数据。

Abstract: Deep research agents answer complex questions through iterative loops of searching, reading, and reasoning. Recent work on reasoning-intensive benchmarks such as BrowseComp-Plus shows that well-configured lexical retrieval can surface high-quality evidence, yet agents may still fail to connect documents carrying evidence to the gold documents. We identify a deep research agent’s first retrieval move as an important design decision for this setting. We introduce Question’s Gambit, a first-move retrieval module that decomposes the question into a set of clues, reformulates them into complementary searches, consolidates the retrieved results, and reranks the candidate pool before the agent begins its iterative search-and-reasoning process. This produces an opening context designed to support both clue aggregation and final-answer verification. We further evaluate on MultiHop-RAG to test whether these benefits transfer beyond BrowseComp-Plus to a more conventional multi-hop question structure. Experiments on BrowseComp-Plus show that Question’s Gambit improves retrieval recall and downstream agent accuracy over strong baselines, improving answer accuracy from 83.1% to 90.5% with gpt-5.5 over Pi-Serini, the strongest reported agentic baseline. Our results confirm that effective agentic deep research depends not only on the tools available inside the loop, but also on the quality of the first move. We published our implementation publicly at https://github.com/radinhamidi/Question-s-Gambit.

[171] Self-Orchestrating Language Models: Leveraging Semantic Dependence for Efficient Inference cs.AI | cs.CLPDF

Tian Jin

TL;DR: 这篇论文提出“自编排语言模型”概念，让模型标记 token 间语义依赖，再由运行时据此并行解码、驱逐中间上下文或安排离散扩散去噪。

Details

Motivation: 自回归解码在低批次时延高且硬件利用不足，离散扩散需很多去噪步才能达到自回归质量，长上下文推理又受 KV 内存限制。

Result: 作者提出让语言模型在生成中标注语义依赖关系，使运行时能调整执行策略，并介绍三个系统：PASTA 根据块间依赖并行化自回归解码；TIP 将依赖用于 KV 缓存驱逐以降低记忆占用且保持精度；Planned Diffusion 自回归生成去噪计划，决定哪些块并行去噪。摘要提出总体方法和系统构成，未报告具体实验数字。

Insight: 模型可不仅生成答案，也可输出帮助运行时编排的依赖计划，从而在质量与效率间动态权衡；三种系统对应不同瓶颈，但摘要没有比较其性能和共同限制。

Abstract: Large language models (LLMs) demonstrate impressive capabilities, but their deployment presents significant efficiency challenges. Autoregressive decoding imposes substantial inference latency and under-utilizes hardware accelerators in low batch size regimes. Discrete diffusion models can generate in parallel but struggle to match autoregressive quality without many diffusion denoising steps. Long-context reasoning creates memory bottlenecks that strain even state-of-the-art accelerators. My thesis is that language models can direct their own inference execution strategy by annotating semantic dependence – which tokens depend on which others – in their generation. I call such models self-orchestrating language models. For each system, I design a runtime that acts on these annotations to parallelize autoregressive decoding, evict intermediate context, or derive denoising orders, achieving Pareto-optimal quality-efficiency trade-offs. I demonstrate this approach through three self-orchestrating systems. First, PASTA uses semantic dependence to parallelize autoregressive decoding, training the model to annotate which output chunks can generate independently. Second, TIP uses semantic dependence to evict intermediate reasoning steps from the KV cache, reducing memory consumption while preserving accuracy. Third, Planned Diffusion uses semantic dependence to derive a denoising order for discrete diffusion, autoregressively generating a plan that specifies which chunks to denoise in parallel.

[172] Converting Sequenced Fuzzy Cognitive Maps to Causal Virtual Worlds with Large Video Generators cs.AI | cs.CL | cs.IRPDF

Akash Kumar Panda, Olaoluwa Adigun, Bart Kosko

TL;DR: 论文用反馈模糊认知图表示虚拟世界中的局部因果规则与全局稳定状态，再由语言模型编写脚本、视频模型按因果场景生成视频。

Details

Motivation: 用户希望创建和操纵具有因果演化的虚拟视频世界，但直接让生成模型续写场景难以确保事件按预设因果结构发展。

Result: 反馈模糊认知图（FCM）表示世界变量间的模糊局部规则，其反馈动力学收敛到定义因果情景的均衡。动态元规则“If A then B”中，A 按用户或代理意图扰动 FCM，B 为最终吸引态；算法从 FCM 提取规则序列并引导 LLM 写脚本，再由视频生成器将规则转为场景。示例使用海底海豚与鲨鱼 FCM，由 Gemini 3.1 写脚本、Veo 3.1 生成视频。

Insight: 用显式因果动力学和均衡状态约束视频场景生成，有望让事件顺序遵循可编辑的虚拟世界规则；论文仅展示简单海底示例，复杂世界扩展属于提出的方向而非已验证规模。

Abstract: We show how users can create and manipulate causal virtual worlds with large-language-model (LLM) and large-video-model agents. The approach uses feedback fuzzy cognitive maps (FCMs) both to model the granular causal structure of the virtual world and to guide its causal evolution. The local causal rules are partial or fuzzy while the FCM’s feedback structure produces global equilibria that define causal scenarios. A sequence of \emph{dynamical} meta-rules of the form ``If $\mathcal{A}$ then $\mathcal{B}$” define the causal scenes of the virtual-world video. The if-part causal pattern $\mathcal{A}$ perturbs the FCM’s virtual world at the user’s or agent’s discretion. The FCM’s transient feedback dynamics define the meta-rule’s causal arrow of implication. The then-part $\mathcal{B}$ is the resulting equilibrium attractor such as a FCM limit cycle or fixed point. Our algorithm extracts these meta-rules from the FCM and guides the LLM agent to write a script based on the FCM meta-rule sequence. The large video generator converts the meta-rule into a video scene in accord with the flow of the dynamics. We applied the agent-based technique to a simple FCM that describes an undersea world of dolphins and sharks. Google’s Gemini 3.1 generated the script and Google’s Veo 3.1 generated the dolphin-shark video. The approach is general and can scale by mixing larger FCMs and AI agents to produce more immersive virtual worlds.

[173] Reason What Matters: Retrieval-Grounded Reasoning for Universal Multimodal Embeddings cs.AI | cs.CLPDF

Mingzhou Jiang, Peixi Wu, Hang Cheng, Yunhao Zhou, Biao Yang

TL;DR: ReWAM通过检索反馈改进多模态嵌入模型的推理监督与推理时计算：用检索相关的逐 token 信号训练推理过程，并根据检索置信度提前终止或加速推理，在MMEB-V2和MRMR上取得较强检索表现，推理吞吐最高提升5倍。

Details

Motivation: 现有基于GRPO的多模态嵌入推理方法把整段思维链视作同等重要，无法区分真正由输入证据支持且能区分正样本与难负样本的内容；每次都生成完整思维链也会增加延迟，即使部分推理已经足以完成检索。

Result: 论文提出检索感知自蒸馏RASD，利用输入证据和检索到的难负样本构造特权指导，再由在线教师把轨迹级反馈细化为 token 级监督；并提出检索自适应推理RAI，以置信度头评估部分思维链的剩余检索价值，提前停止无效推理，或通过投机解码加速有效延续。在MMEB-V2和MRMR上的实验显示其达到先进检索性能，较有竞争力的显式CoT嵌入方法吞吐最高提高5倍。

Insight: 该工作将检索反馈同时用于判断“哪些推理内容值得学习”以及“推理何时可以结束”，针对检索任务把推理质量与推理成本联合优化；论文摘要未给出所有基准的具体分数，最高5倍是吞吐增益上限。

Abstract: Universal multimodal embedding (UME) learns unified representations across modalities, enabling a single model to support diverse retrieval tasks. Recent methods use Chain-of-Thought (CoT) reasoning to better interpret multimodal inputs before generating embeddings for complex retrieval tasks and further optimize this reasoning process through GRPO with retrieval-based rewards. However, two limitations hinder corpus-scale deployment. GRPO assigns all CoT tokens the same advantage, without identifying input-supported claims or evidence that distinguishes the positive from negatives. Moreover, generating a complete CoT before each embedding introduces substantial latency, even when a partial trace already provides sufficient retrieval evidence. To address these limitations, we propose Reason What Matters (ReWAM), a retrieval-grounded reasoning framework that uses retrieval feedback to guide both credit assignment and reasoning computation. Specifically, we introduce Retrieval-aware Self-Distillation (RASD), which constructs privileged guidance from input-supported evidence that distinguishes the positive item from retrieved hard negatives. An on-policy self-teacher uses this guidance to refine trajectory-level feedback into token-specific supervision for retrieval-relevant reasoning. We further develop Retrieval-adaptive Inference (RAI), which uses a retrieval confidence head to estimate the remaining retrieval utility of a partial CoT. It stops unproductive traces early and accelerates useful continuations with speculative decoding. Extensive experiments on MMEB-V2 and MRMR demonstrate that ReWAM achieves state-of-the-art retrieval performance while delivering up to 5x the inference throughput of competitive explicit-CoT UME methods. These results bridge the gap between retrieval quality and inference efficiency, making reasoning-enhanced UME practical for large-scale deployment.

[174] Beyond Safe Answers: Segment-Aware Listwise Alignment for Reasoning Safety in Large Reasoning Models cs.AI | cs.CLPDF

JungMin Yun, Junehyoung Kwon, Hayeong Ryu, Byeonggeuk Lim, Hoejoon Kwon

TL;DR: SaLT-DPO针对大推理模型的推理过程和最终回答分别做分段列表式安全对齐，并加入跨段安全一致性约束和良性任务效用锚定；在三个模型上降低不安全内容，同时尽量保持良性遵从和一般推理能力。

Details

Motivation: 大推理模型存在双层安全问题：中间推理轨迹和最终答案都可能包含有害内容。整体响应级对齐会让不安全推理被看似安全的最终答案掩盖，也可能因过度安全对齐导致拒答增加或推理能力退化。

Result: 作者提出SaLT-DPO：将响应分为推理段和答案段，分别评估安全性，对多个候选的长度归一化分段奖励与软目标分布进行列表式对齐；用联合安全一致性正则化遵循“最弱环节”原则；并在良性提示上进行效用锚定以缓解过度拒答和推理退化。三个大推理模型上的实验显示，推理段及答案段的不安全率均持续下降，同时良性遵从退化得到缓解，一般推理表现得以保持。消融实验显示各组成机制具有互补作用。

Insight: 安全评估与对齐应分别检查推理轨迹和最终答案，并确保两者在安全性上保持一致；只优化整段输出可能漏掉隐藏在推理过程中的风险。摘要没有列出具体不安全率变化或基准名称，因此只能概括趋势，不能量化收益。

Abstract: Large Reasoning Models (LRMs) pose a dual-surface safety challenge: both intermediate reasoning traces and final answers can contain harmful content. Existing alignment methods often operate at the whole-response level, allowing unsafe reasoning to be masked by a safe-looking final answer. We propose Segment-aware Listwise Target DPO (SaLT-DPO), which addresses this gap through three mechanisms: (1) segment-aware listwise alignment that decomposes responses into reasoning and answer segments, independently scores each segment’s safety, and aligns length-normalized segment rewards with soft target distributions over multiple candidates; (2) joint safety coherence regularization that applies a weakest-link principle to promote safety consistency across both segments; and (3) utility anchoring on benign prompts to mitigate over-refusal and reasoning degradation. Experiments on three LRMs show that SaLT-DPO consistently reduces unsafe rates for both reasoning and answer segments while mitigating degradation in benign compliance and preserving general reasoning performance. Ablation studies demonstrate the complementary contributions of its components.

[175] Option-Aware Retrieval and Task-Specific VLM Adaptation for Medical VQA cs.AI | cs.CL | cs.CVPDF

Tristan Kirscher, Niklas C. Koser, Soren Pirk

TL;DR: MedReason 2026竞赛系统研究发现，医学选择题检索必须匹配选项语义而非选项标签；最终MCQ适配器达到94.0%开发集准确率，但增加检索示例或置信度门控几乎没有帮助，开放题指标则低于参考基线。

Details

Motivation: 医学视觉问答同时包含选择题和开放题，并要求完全离线、容器化推理。不同问题的选项标签独立分配，直接复制相似病例的答案字母无法迁移；同时也需厘清检索、任务适配器和门控对提交系统性能的实际贡献。

Result: 系统发现按当前选项文本与相似训练病例正确答案的语义相似度检索，可使检索单独准确率在200例开发留出集上从20.0%升至57.5%。固定MCQ LoRA适配器后，提示内检索数k从0增加到3仅改变最多1例：k=0及训练时k=1均为187/200（93.5%），默认k=3为188/200（94.0%）；置信度门控在k=3上无净准确率增益，198/200例选择VLM。开放题20例中，k增加时token-F1和RaTEScore下降，但token-F1差异配对符号检验不显著（p≥0.29）；最终配置单标注者比较中20例有6例错误锚定，早期联合改变路由、适配器和提示的配置有14例。官方预评估MCQ准确率93.20%，高于离线参考基线29.43%；但两个开放题指标低于基线：真实答案一致度1.245对1.588，视觉准确度1.995对2.696（均满分4）。

Insight: 检索收益依赖于比较答案语义而非选项字母；在适配器已很强时，额外检索和置信门控带来的MCQ收益极小。系统在选择题和开放题上的表现分化明显，不能用单一MCQ高分概括医学VQA整体能力；开放题结论样本仅20例。

Abstract: We describe our submission to the MedReason 2026 challenge, covering multiple-choice (MCQ) and open-ended (OE) medical visual question answering (VQA) under fully offline, containerized inference. Our first finding is that MCQ retrieval must compare answer \emph{semantics} rather than answer labels: labels are independently assigned per question, so copying a retrieved neighbor’s label transfers no useful information, whereas scoring each current option’s text against correct-answer text from similar training cases raises retrieval-only accuracy from 20.0% to 57.5% on a 200-case retrieval-excluded development holdout. Our second finding attributes the submitted system’s accuracy: holding the task-specific MCQ Low-Rank Adaptation (LoRA) adapter fixed and varying the number (k) of in-prompt retrieved examples changes accuracy by at most one case — 187/200 (93.5%) at both (k=0) and the adapter’s training-time (k=1), 188/200 (94.0%) at the packaged runtime’s default (k=3) — and the submitted confidence-gated override adds no net accuracy on top of (k=3), selecting the VLM in 198/200 cases. With the final MCQ adapter fixed, retrieval changes accuracy by at most one case, and gating provides no net gain. On 20 OE cases, token-F1 and RaTEScore~\cite{zhao2024ratescore} decrease as (k) grows, but paired sign tests on token-F1 differences are nonsignificant ((p \ge 0.29)); a single-annotator comparison found 6/20 wrong-anchor errors for the final configuration and 14/20 for an earlier configuration that jointly differed in routing, adapter, and prompting. The system reaches 94.0% MCQ accuracy on the development holdout and 93.20% on the organizer’s official pre-evaluation, versus 29.43% for the off-the-shelf reference baseline, while both of the organizer’s open-ended scores are lower than that baseline’s (ground-truth agreement 1.245 versus 1.588, visual accuracy 1.995 versus 2.696, each out of 4).

cs.GR [Back]

[176] ESG: Generating Physically Consistent Dynamic 3D Scenes from Text Descriptions cs.GR | cs.CVPDF

Xintong Fang, Zhiyuan Fang, Rengan Xie, Xuhong Zhang, Guoyuan An

TL;DR: ESG通过“演化场景图”把物理属性、空间关系和事件时间线编码为机器可检查的结构，并结合布局优化、可微分物理模拟和引擎编译，从文本生成可在Unreal Engine执行的动态3D场景；10个场景平均完成16.4/18个事件。

Details

Motivation: 图像和3D场景生成已能产生较逼真的静态环境，但自然语言生成动态场景还需联合推理场景结构、时间演化和物理可行性，并保证结果能在现代物理引擎中可靠执行。

Result: 系统先让大语言模型依据提示构造并验证完整的演化场景图（ESG），其中实体带有物理属性、空间关系及事件驱动时间线；再通过能量最小化梯度优化确定空间布局，利用可微分模拟优化受时间线约束的物理参数以满足用户指定事件，最后编译为Unreal Engine可执行类。在三个复杂度等级的10个场景实验中，平均完成16.4/18个事件，明显超过Scene Language、最强引擎可执行基线SimWorld及去除物理优化的消融方案；参数准确性也更高。

Insight: 把动态场景需求表示成显式、机器可检查的事件和物理约束，再交由优化与模拟执行，可将自然语言生成从外观描述推进到可运行场景。评测仅含10个场景，摘要没有给出逐场景结果及复杂度变化，泛化范围仍需正文确认。

Abstract: Recent progress in image and 3D scene generation has enabled increasingly realistic static environments, yet most methods remain confined to such static configurations. Generating dynamic scenes from natural language is fundamentally challenging: it requires joint reasoning over scene structure, temporal evolution, and physical feasibility, while ensuring reliable execution in modern physics engines. We present a unified framework for generating physically consistent dynamic 3D scenes from text, with outputs directly executable in Unreal Engine. Central to our approach is the \emph{Evolutive Scene Graph} (ESG), which specifies entities with physical attributes, spatial relations, and event-driven timelines in a machine-checkable form. Given a prompt, a large language model constructs and validates a complete ESG; spatial layouts are grounded via energy-minimized gradient optimization; timeline-constrained physical parameters are then optimized through differentiable simulation to satisfy user-specified events; and the resulting scene is compiled into an engine-executable class. Experiments on 10 scenes across three complexity levels show that our method achieves $16.4/18$ mean event completion, outperforming Scene Language, the strongest engine-executable baseline (SimWorld), and our ablation without physical optimization by a clear margin in event completion and parameter accuracy.

[177] MedVA: An End-to-End Neuro-Symbolic Agentic System for Medical Volume Visualization cs.GR | cs.CV | cs.HCPDF

Haill An, Suhyeon Kim, Minjun Kang, Eunwoo Lee, Bin Sheng

TL;DR: MedVA是面向医学体数据可视化的端到端神经符号智能体系统，通过临床知识符号推理细化用户意图、多模型识别原始体数据中的ROI，并用体积可见性目标优化显示；多数据集评估和形成性用户研究支持其有效性与易用性。

Details

Motivation: 医学体数据可视化需要按临床意图选择感兴趣区域并合理调节视觉强调，传统流程依赖大量临床及可视化经验且需反复试调。现有智能体多由多模态大模型贯穿工作流，可能未能完整理解临床请求，导致ROI识别和可视化优化不可靠。

Result: MedVA由三个互补智能体构成：神经符号意图制定代理用既有临床知识进行符号推理，补全纯MLLM解析得到的ROI规格；多模型ROI识别代理利用互补的大规模预训练医学分割模型，直接在原始体数据中定位语义指定区域；目标驱动的可视化优化代理则通过体积可见性目标显式评估ROI的可见程度和遮挡。跨多种医学数据集和交互场景的代理级及系统级评估支持各模块效果，形成性用户研究也表明不同专业水平用户认为系统可用且有实际价值。

Insight: 医学可视化智能体可将语言模型的意图理解与临床知识约束、多模型分割及显式可见性优化结合，减少单一生成模型在临床细节上的遗漏。摘要没有提供各评测指标数值、用户研究规模或对临床决策结果的验证，不能据此视为临床有效性证明。

Abstract: Medical volume visualization requires selecting regions of interest (ROIs) and carefully controlling their relative visual emphasis according to a given clinical intent. Implementing these decisions in conventional workflows demands substantial clinical and visualization expertise and often involves trial-and-error optimization. Recent agentic systems have introduced natural-language interaction and autonomous visualization operations but largely rely on MLLM-based inference throughout the workflow. Although MLLMs encode broad medical knowledge and provide strong reasoning capabilities, such inference may be suboptimal for medical volume visualization, potentially leading to clinically incomplete interpretations of user requests and unreliable ROI identification and visualization optimization. In this work, we present MedVA, an end-to-end neuro-symbolic agentic system for medical volume visualization that addresses these limitations through three complementary agents. The neuro-symbolic intent formulation agent refines MLLM-based interpretations of natural-language requests through symbolic reasoning over established clinical knowledge, which provides more complete, clinically grounded ROI specifications than MLLM-only reasoning. The multi-model ROI identification agent directly identifies semantically specified ROIs in the original volume by leveraging complementary large-scale pretrained medical segmentation models. The objective-driven visualization optimization agent explicitly evaluates ROI visibility and occlusion in the original volume using a volume-based visibility objective. Extensive agent-level and system-level evaluations across diverse medical datasets and interaction scenarios support the effectiveness of the individual agents. A formative user study further indicates high usability and practical value among users with different levels of expertise.

[178] TopoRig: Topology-Agnostic Facial Rigging via Multi-Source Supervision cs.GR | cs.CVPDF

Andrew Fleet, Soroush Mehraban, Vida Adeli, Cole Clifford, Babak Taati

TL;DR: TopoRig提出不改变输入网格拓扑的拓扑无关人脸绑定方法，直接在任意顶点上预测FACS条件形变，并融合标准模板绑定、跨拓扑迁移结果及图像线索监督；在未见身份和未见网格拓扑上更接近参考表情空间。

Details

Motivation: 不同网格拓扑间自动生成面部绑定仍困难：高质量表情监督往往依赖标准模板，而把形变迁移到任意网格会产生几何伪影和对应错误。需要一种能处理异构网格，同时保留原拓扑并泛化到新角色的方法。

Result: TopoRig从ICT FaceKit表情模型出发，构造互补监督：准确但受模板拓扑限制的标准绑定、拓扑多样但较噪声的迁移绑定，以及针对几何迁移难以表达的控制项的图像线索。模型结合局部表面几何、相对地标的语义特征、全局形状上下文及FACS控制，直接预测输入网格逐顶点位移并保持原拓扑。训练集含3,496个生成身份，使用ICT FaceKit 53项控制词汇中的45项非注视表情控制。在留出身份和未见拓扑上，TopoRig比既有神经人脸绑定方法更忠实地复现参考表情空间；定性结果显示多种角色几何上的局部形变一致。消融表明语义地标特征和互补监督有助于跨身份、跨拓扑泛化。

Insight: 把模板绑定、迁移绑定和图像线索作为互补监督，而不是依赖单一来源，可将异构且有噪声的表情标签蒸馏成保持拓扑的形变模型。摘要没有给出量化表情误差或用户评估结果，泛化收益主要由对比结论及定性结果支持。

Abstract: Automatic facial rigging across heterogeneous mesh topologies remains challenging because high-quality expression supervision is often tied to canonical templates, while deformation transfer to arbitrary meshes can introduce geometric artifacts and correspondence errors. We present TopoRig, a topology-agnostic facial rigging framework that predicts FACS-conditioned deformations directly on input mesh vertices while preserving the original topology. Starting from the ICT FaceKit expression model, we construct complementary supervision from accurate but template-biased common-topology rigs, topology-diverse but noisier transferred rigs, and targeted image-based cues for controls poorly captured by geometric transfer. TopoRig combines local surface geometry, landmark-relative semantic features, global shape context, and FACS controls to predict per-vertex displacements. We train on 3,496 generated identities using 45 non-gaze expression controls from the 53-control ICT FaceKit vocabulary. On held-out identities and unseen mesh topologies, TopoRig more faithfully reproduces the reference expression space than prior neural facial-rigging methods, while qualitative results show consistent localized deformations across diverse character geometries. Ablations demonstrate that semantic landmark features and complementary supervision improve cross-identity and cross-topology generalization. Overall, TopoRig amortizes heterogeneous and imperfect expression supervision into a single topology-preserving deformation model.

cs.HC [Back]

[179] Speak to the City: Multimodal Resolution for Outside-the-Vehicle References cs.HC | cs.CL | cs.IR | cs.LG | cs.MMPDF

Alireza Parchami, Artin Saberpour, Robin Connor Schramm, Jürgen Steimle, Ulrich Schwanecke

TL;DR: 研究提出融合乘客视线和自然语言的车外地标指代解析方法，并以同步 360° 视频与 GNSS 的 VR 研究数据训练轻量模型。

Details

Motivation: 车载自动驾驶与 XR 场景中，用户可能通过“那栋楼”等方式指向车外兴趣点，但车辆自运动及指代歧义使系统难以确定目标。

Result: 作者开发 VR 数据采集流程，将 360° 公交视频与车辆 GNSS 同步；46 名参与者通过头部朝向映射至三维地理数字孪生，记录自然视线—语音行为。训练轻量 Transformer，并借助 LLM 动态对齐连续空间注视向量和离散语言语境。Top-1 准确率 83.33%，Top-2 为 87.72%，平均推理时间 24.3 毫秒。

Insight: 连续空间视线与语言上下文联合可消解车内外地标指代歧义，并以实时延迟支持交互；实验来自 VR 模拟公交场景，真实道路中头部注视与车辆运动的泛化需验证。

Abstract: As autonomous vehicles and Extended Reality (XR) headsets enable novel in-car interactions, seamlessly querying physical landmarks, known as Outside-the-Vehicle Referencing (OVR), remains challenging due to ego-motion and referential ambiguity. We present a robust, multimodal OVR framework fusing user gaze and natural language to identify Points of Interest (POIs). To address the scarcity of dynamic vehicular data, we developed a VR-based pipeline synchronizing 360-degree transit videos with vehicle GNSS telemetry. Through a user study (N=46) mapping passenger head orientation into a 3D geospatial Digital Twin, we captured authentic gaze-speech behaviors. We subsequently trained a lightweight Transformer network, leveraging LLMs to dynamically align continuous spatial gaze vectors with discrete verbal context. Experimental results demonstrate high accuracy and low computational overhead, achieving an 83.33% Top-1 accuracy (87.72% Top-2) and an average inference time of 24.3 milliseconds. This real-time paradigm effectively resolves referential ambiguity, enabling context-aware spatial retrieval for passengers within the vehicle.

[180] PeerPen: AI-Assisted Writing for Online Mental Health Peer Support cs.HC | cs.AI | cs.CL | cs.CY | cs.SIPDF

Jiwon Kim, Sherry Gong, Maya Ajit, Soorya Ram Shimgekar, Yunhao Yuan

TL;DR: PeerPen 是嵌入 Reddit 式社区的心理健康互助写作辅助工具，支持生成草稿与修改用户文本，并探索如何兼顾表达负担、作者身份和信任。

Details

Motivation: 在线心理健康社区依赖同伴支持，但志愿者缺少正式培训，可能难以写出有支持性的回应；AI 能降低写作门槛，却可能削弱被认为“亲自写”的真实性。

Result: 作者构建 PeerPen 并对 15 名参与者进行半结构访谈。参与者认为它降低撰写支持回复的负担、提高提供帮助的信心；同时希望 AI 协助而不接管作者身份，并担忧真实性与信任。即使某条回复未实际使用 AI，社区成员也可能因辅助工具存在而不确定作者身份，从而影响整体互信。论文提出支持沟通、保留作者权并考虑社区层面信任的设计启示。

Insight: 心理健康互助中的 AI 写作应作为脚手架而非代写者，透明度和作者控制是维护信任的核心；研究样本仅 15 人，未量化长期社区参与或支持质量。

Abstract: Online mental health communities thrive on peer support, yet those who volunteer to help often lack formal training and may struggle to articulate supportive responses. AI co-writing could lower this barrier; however, peer support derives much of its value from being perceived as personal, raising questions around authorship, ownership, and trust. We built PeerPen, a writing assistance tool embedded within a Reddit-like interface, supporting two main features: draft generation and revision of user-written responses. Through semi-structured interviews with 15 participants, we find that PeerPen reduced the burden of composing responses and increased confidence in offering support. Participants wanted AI to assist their writing without taking over authorship and anticipated tensions around authenticity and trust. Such assistance could make authorship uncertain even for responses written without it, weakening trust across the community. We contribute design implications for AI writing assistance that scaffolds supportive communication, preserves authorship, and accounts for community-level trust.

cs.LG [Back]

[181] HarnessBandit: Joint Learnability-Transferability Scheduling for Multi-Harness Agentic Reinforcement Learning cs.LG | cs.AI | cs.CLPDF

Hongliang Wei, Xiaobing Tu, Yinggui Wang, Zhengxi Liu, Rongkun Xue

TL;DR: HarnessBandit 在多种智能体运行框架之间逐步调度 GRPO 训练，以可学习性和跨框架梯度可迁移性共同选择每个优化步骤的训练环境。

Details

Motivation: 同一语言模型在不同提示、工具 schema、控制循环和轨迹格式下表现不一；共享策略跨多个 harness 训练时，需要选择当前既提供有效学习信号又能迁移到其他框架的训练环境。

Result: 每次 GRPO 更新后，调度器观测批次平均绝对 advantage 作为可学习性，并以当前 harness 的低维梯度草图与其他 harness EMA 的余弦相似度估算可迁移性；池化滑窗归一化后结合访问奖励和探索下限采样。Qwen3.5-2B 在 ClawGym 六种 harness 上训练后，在 PinchBench 和 ClawEval 上均优于混合批次多框架训练，诊断显示两种信号互补且随训练演变。

Insight: 训练环境的即时可学性与跨接口迁移性是不同信号，在线联合调度比静态混合更有效；结果针对指定 Qwen 模型、框架和任务集，其他模型规模的规律待验证。

Abstract: Language-model agents are increasingly deployed through diverse harnesses that differ in system prompts, tool schemas, control loops, and trajectory formats. The same model can perform unevenly across these interfaces, making robustness to harness variation an important objective. A natural approach is to train a shared policy through multiple harnesses, but doing so introduces a scheduling problem: each training step should favor a harness that currently provides a useful learning signal while also producing an update that benefits the other harnesses. We develop HarnessBandit, an online scheduler that selects one harness per optimizer step. After a group-relative policy optimization (GRPO) update, it observes learnability – the mean absolute advantage on the batch – and transferability – the cosine between a low-dimensional gradient sketch of the current harness and exponential moving averages of the remaining harnesses. The two signals are fused after pooled sliding-window min-max normalization and sampled with a visit-dependent bonus and an explicit exploration floor. We train Qwen3.5-2B across six harnesses on ClawGym and evaluate on PinchBench (held-out tasks, in-distribution OpenClaw) and ClawEval (held-out tasks and harness). HarnessBandit improves over mixed-batch multi-harness training on both benchmarks, while training diagnostics indicate that learnability and transferability provide distinct, evolving signals.

[182] Training Specialist Models without Reasoning Trajectories for Domain Expert Distillation cs.LG | cs.CLPDF

Yilei Tu, Zihao Li, Shaoxiong Ji, Jörg Tiedemann, Fei Yuan

TL;DR: 论文用学生模型蒸馏作为探针，研究仅凭问答对训练的领域专家模型会生成怎样的潜在推理轨迹，并发现优化选择可控制领域精度与通用能力的权衡。

Details

Motivation: 领域专家蒸馏通常依赖教师生成的显式推理轨迹，但只有问答监督时，专家模型内部会形成哪些轨迹及其受训练策略如何影响尚不明确。

Result: 作者把学生蒸馏作为观察轨迹分布的探针，而非主要下游目标，因为学生只继承抽样轨迹、不继承教师参数化和优化约束。27 组教师—学生配对中，领域专精与泛化能力表现高度相关。明确控制教师分布漂移，可系统改变教师及其蒸馏学生在领域精确度与通用能力保持之间的位置；结果覆盖化学、物理和多语言任务，且在不同模型家族间迁移。

Insight: 无金标准推理监督时，教师训练选择会塑造传给学生的隐式监督分布；分布漂移不是单纯副作用，也可作为调节专业化—泛化权衡的控制量。

Abstract: Specialist distillation effectively transfers domain expertise to student models via teacher-generated reasoning trajectories. However, when these specialists are trained solely on question–answer pairs without explicit reasoning supervision, what governs the trajectories they generate? In this work, we show that specialist optimization implicitly selects from this latent trajectory space. To isolate and observe this latent distribution, we leverage student distillation not as a downstream goal, but as an agnostic probe—since students inherit no parameterization or optimization constraints from the specialist, inheriting only the sampled trajectories themselves. Through this probe, our empirical analysis unveils a tight governing relationship: across 27 specialist–student pairings, their specialization–generalization profiles correlate exceptionally strongly. Crucially, explicitly controlling the specialist’s distributional drift systematically shifts both the teacher and its distilled student along a controllable trade-off between domain precision and general-capability retention. Across chemistry, physics, and multilingual settings, distilled students systematically reflect these specialist-induced profiles, even across divergent model families. Our findings establish a new view of specialist training: when gold reasoning is absent, tuning choices directly control the latent supervision passed to downstream models.

[183] Does Reasoning Improve Psychological Depth in Large Language Models? It Depends on Who’s Judging cs.LG | cs.CLPDF

Ruichen Zheng, Yihe Wang, Fabrice Y Harel-Canada, Sara Khosravi, Zeynep Senahan Yildiz

TL;DR: 短篇故事心理深度评测发现，LLM 评审与人类偏好存在系统差异：评审几乎总偏好推理模型，而人类没有一致的推理增益偏好。

Details

Motivation: LLM 评审在开发集上与人类评分相关，并不保证在主观、差距微妙的输出上有效；作者以故事心理深度检验推理模型评审的分布外效度。

Result: 七名人类读者与在原始心理深度标量基准上相关系数 0.646 的 LLM 评审集，盲评 60 对匹配提示故事（GPT-5 对 GPT-4o、DeepSeek-R1 对 V3）。人类仅温和偏好 GPT-5（60.0%–62.9%），对 DeepSeek-R1 的偏好为 42.9%，读者间一致性接近随机（Krippendorff α=0.070），但个体内一致性和加权模式显示是结构化异质而非随机回答。评审在五种配置中均偏好推理输出：维度级比较 89.0%，总体 60 对中偏好 59 对，并与句长、词汇多样性等表面特征相关。

Insight: 开发集相关性不足以证明评审在新分布和细微主观评价上的有效性；点估计评审可能掩盖人类偏好异质性，并把文风变化误判成心理深度提升。

Abstract: LLM-as-a-Judge evaluators are increasingly used to score open-ended generation, yet a judge’s correlation with human ratings on its development set may not guarantee valid measurement when outputs are closely matched and human preferences are subjective. We study this failure mode through psychological depth in short stories. Seven human readers and an LLM-judge ensemble selected on the original scalar Psychological Depth Scale dataset ($ρ= 0.646$) evaluated 60 blinded, prompt-matched story pairs from GPT-5 vs.\ GPT-4o and DeepSeek-R1 vs.\ DeepSeek-V3. Human preferences showed no universal reasoning advantage: GPT-5 was modestly preferred over GPT-4o (60.0–62.9%), whereas DeepSeek-R1 trailed V3 (42.9%), and inter-reader agreement was near chance (Krippendorff’s $α= 0.070$), with within-reader consistency and recurring weighting patterns suggesting structured heterogeneity rather than random responding. The judge, by contrast, favored reasoning outputs in 89.0% of dimension-level comparisons and 59 of 60 pairs on aggregate PDS, uniformly across all five evaluator configurations, and its scores were associated with surface features such as sentence length and lexical diversity. These results suggest that development-set performance is insufficient evidence for deployment validity on a shifted distribution, and that point-estimate judges can obscure the heterogeneity in subjective human evaluation.

[184] Know When to Stop, Where to Restart: Accelerating Multi-Turn Agentic On-Policy Distillation cs.LG | cs.CLPDF

Zhiyu Gui, Kexin Huang, Jia Guo, Junkang Wu, Zihao Wang

TL;DR: STRIDE 依据教师信号质量自适应停止较差的学生 rollout，并缓存高质量前缀后从最弱的正确轮次重启，加速多轮智能体 on-policy 蒸馏。

Details

Motivation: 多轮智能体 OPD 成本主要来自自回归学生 rollout；现有加速方法按固定离线预算截断或重定位监督，未利用教师信号在轨迹内外变化的可靠性。

Result: 在 τ²-bench 上发现有效监督集中于每轮前缀，教师认可度的跨轮下降通常与学生首次错误动作时间锁定。STRIDE 以累计教师 log 概率低于 OOD 阈值时提前停止，并用前缀缓冲保存高质量上下文、从最弱正确轮次重启，形成逐步扩展到后续轮次的课程。在零售任务上匹配完整轨迹 OPD，并以 3.73 倍速度超过 30B 教师；2.34 倍速度即超过基线。跨域多教师下加速 4.51 倍；AIME 2025/2024 上分别以 5.10/3.08 倍加速优于完整 OPD。

Insight: 监督可靠性在轨迹中并非均匀分布，按首个错误断点停止并复用可靠前缀可同时节省生成成本、扩展有效监督；多任务结果积极，但阈值和缓存策略对其他 agent 环境的迁移需验证。

Abstract: On-policy distillation (OPD) has become a standard approach for transferring capabilities from large teachers to compact students. Its cost, however, is dominated by autoregressive student rollouts and scales poorly in multi-turn agentic settings. Existing acceleration methods truncate or relocate the supervision signal according to fixed, offline budgets, despite substantial variation in teacher-signal reliability both within and across trajectories. Our empirical analysis on $τ^2$-bench reveals a clear structure in this variation: informative supervision is concentrated in the prefix of each turn, and, most importantly for multi-turn agentic training, the cross-turn loss of teacher endorsement is temporally locked to the student’s first erroneous action rather than accumulating gradually over turns. Building on these findings, we propose STRIDE (Stop-and-Restart on-policy Distillation acceleration), which combines two complementary techniques: adaptive early stopping, which terminates a rollout once the cumulative teacher log-probability falls below an out-of-distribution threshold, and a prefix buffer, which caches high-quality prefixes and restarts generation at the weakest correct turn. Together, these mechanisms induce a data-driven curriculum that progressively extends coverage to later turns. On $τ^2$-bench retail, our method matches full-trajectory OPD and exceeds the 30B teacher at a $3.73\times$ speedup, surpasses the baseline itself at $2.34\times$, and retains a $4.51\times$ speedup under cross-domain multi-teacher training. As a supplementary generalization test beyond the agentic setting, STRIDE outperforms full OPD on AIME 2025 at a $5.10\times$ speedup and on AIME 2024 at a $3.08\times$ speedup; averaged across the two evaluations, both fixed-budget truncation baselines remain below full OPD.

[185] Refusal Reads Only a Slice of What the Model Knows: Harm-Keyed Routing and Its Exceptions Across Model Families cs.LG | cs.AI | cs.CLPDF

Orion Reblitz-Richardson

TL;DR: 论文认为安全拒答只读取模型道德知识的一小部分，并通过因果互换实验发现不同模型家族的拒答决策依赖不同表征。

Details

Motivation: 单一残差方向被编辑后模型不再拒答，说明拒答容易被移除，却不能说明拒答机制实际读取哪些知识；作者区分模型理解道德内容与触发拒答的信号。

Result: 四个开源模型、三家族的分析显示道德理解在预训练形成低秩子空间，后训练主要旋转而非重建；拒答门则主要是后训练新建、位于狭窄控制 token 通道，且与道德判断正交。对 OLMo-3 的嵌套秩互换实验显示，道德判断随补丁空间扩大持续读取更多道德子空间，拒答则约在单一有害方向饱和，约四分之三拒答因果输入位于道德子空间之外。家族间不同：Llama 读取更广道德内容，Qwen 结果未定，GPT-OSS 主要读有害线索。

Insight: 拒答行为与道德理解并非同一机制，秩一编辑可移除只读取低秩有害信号的拒答；跨家族异质性说明不能把单一模型的因果结论泛化，扩大拒答所读内容是否会深化行为仍未解决。

Abstract: Alignment applied after pretraining is shallow in a measurable way: a single direction in a model’s residual stream can be edited out, and the model stops refusing harmful requests. That fact says how easily refusal can be removed, not what the refusal decision was reading in the first place. We ask what it reads, and we separate that from what the model comprehends. Across four open-weight models spanning three families, moral comprehension is native to pretraining: a low-rank moral subspace crystallizes during pretraining, and alignment rotates it once without rebuilding it. The refusal gate, in contrast, is a fresh post-training construction with only a weak pretraining precursor, written into a narrow control-token channel where the refusal decision is orthogonal to the moral-judgment decision. The central result is causal and comes from one model, OLMo-3. A nested interchange rank sweep patches successively larger slices of the moral subspace between matched requests and reads how much of refusal’s response transfers: as the basis widens, moral judgment keeps reading more of it, while refusal levels off at the level of a single harm direction, and about three-quarters of refusal’s causal input lies outside the moral subspace altogether. Refusal reads the harm percept, not the moral content that judgment reads on the same patches. The picture is not uniform across families. Llama reads broad moral content; Qwen reads beyond the single harm cue but is unresolved at our sample size; GPT-OSS reads harm, and its refusals can be argued in either direction by its own reasoning trace. Where refusal reads only a low-rank slice and routes around the bulk of what the model knows, a rank-one edit removes it. Whether widening what refusal reads would also deepen the behavior is the open question this raises.

[186] Not All Prompts Are Equal: Exploration-Guided Prompt Scaffolding for Multimodal Reinforcement Post-Training cs.LG | cs.AI | cs.CLPDF

Yuanhao Yue, Qianli Ma, Chengyu Wang, Haoting Wang, Lei Shen

TL;DR: 论文用探索潜力分数动态调整多模态 RL 训练提示分布，并让教师模型改写低效提示以增加信息量，而不是模仿教师答案。

Details

Motivation: 在线强化学习中，有些提示已饱和、另一些过难而不能产生可靠学习信号，但标准训练通常对所有提示分配相同 rollout 预算。

Result: 探索潜力分数 EPS 根据 on-policy rollout 统计和 KL 正则策略改进理论估计提示训练效用，无需额外开销。低效提示不被丢弃，而由教师改写以保留原任务意图并产生更有效学习信号，教师角色因此是优化训练数据而非提供目标答案。在 Geo3K、MMK12 上结合 GRPO，方法在域内和分布外基准均优于基线，域内相对提升最高 9.7%，MathVision 提升 11.5%，MMMU-Pro 提升 11.1%。

Insight: 提示的学习价值随当前策略变化，训练分布应动态调度；对过难样本做保意图改写可保留任务覆盖而避免无效 rollout。

Abstract: Training prompts in online reinforcement learning (RL) differ substantially in how informative they are for the current policy: some are already saturated while others are too difficult to yield reliable learning signals, yet both receive equal rollout budget under standard training. We propose an exploration-guided prompt scaffolding framework that adapts the training prompt distribution dynamically throughout RL post-training of multimodal large language models (MLLMs). Central to our approach is the $\textit{Exploration Potential Score} (EPS)$, a lightweight rollout-based proxy for prompt utility derived from KL-regularized policy improvement theory, computable directly from on-policy rollout statistics without additional overhead. Rather than discarding low-utility prompts, we use a teacher model to generate scaffolded rewrites that preserve the original task intent while making subsequent training more informative, reframing teacher supervision as training-data refinement rather than output imitation. Integrated with GRPO on Geo3K and MMK12, our method consistently outperforms the baseline on both in-domain and out-of-distribution benchmarks, achieving up to 9.7% relative improvement in-domain and gains of 11.5% on MathVision and 11.1% on MMMU-Pro.

[187] Land Art as a Big-Data Climate Sensor cs.LG | cs.CV | physics.ao-phPDF

Alev Cinbarci, Sean Kalaycioglu

TL;DR: 研究将犹他州大盐湖的螺旋防波堤卫星影像作为气候与水文变化传感器，分析1984—2025年的1,744个影像样本，发现多尺度排列熵和ResNet特征与湖面状态及累积二氧化碳相关，影像复杂度约领先湖水位变化三年。

Details

Motivation: 大盐湖水位变化会让罗伯特·史密森的地景作品“螺旋防波堤”反复淹没或露出。作者希望判断艺术地景的遥感影像复杂度能否作为长期气候或水文指标，并检验此前小样本研究中“香农熵与全球温度正相关”的说法。

Result: 研究处理1984至2025年逐年逐月的1,744个配准Landsat 4–9和Sentinel-2影像块，计算香农熵、多尺度排列熵、分形维数、空隙度、纹理、强度统计及ImageNet预训练ResNet50特征，并与42年月度气候水文数据对照。自助法分析显示，香农熵是较弱代理，不能支持此前的小样本正相关结论；粗尺度排列熵与平均强度和湖面高程的Spearman相关为0.85–0.88。ResNet特征的第三主成分与累积CO2相关0.86、与湖面高程相关-0.83。影像复杂度领先湖水位约三年，滞后相关r=0.58，95%区间0.40–0.73；控制月份和传感器后仍稳健。趋势在1984至2015年上升，此后随湖面接近历史低位而快速下降。数据集、特征流程和分析代码公开。

Insight: 结果更支持将该作品的影像视为水文状态的领先指标，而非简单的全球温度计；不同复杂度特征的解释力差别明显，香农熵并不可靠。约三年的领先关系是相关性结果，不能单凭摘要解读为因果预测能力。

Abstract: Robert Smithson’s 1970 land artwork Spiral Jetty, located in the north arm of Utah’s Great Salt Lake, has alternated between submergence and exposure during severe lake decline. We analyze 1,744 co-registered Landsat 4-9 and Sentinel-2 image chips spanning every year and calendar month from 1984 to 2025. A 14-feature complexity signature combines Shannon entropy, multiscale permutation entropy, fractal dimension, lacunarity, gray-level co-occurrence texture, intensity statistics, and ImageNet-pretrained ResNet50 features. These measurements are compared with a 42-year monthly climate and hydrology panel from NASA GISTEMP, USGS NWIS, Open-Meteo, and the Global Carbon Budget. Bootstrap analysis shows that Shannon entropy is a weak proxy and does not support an earlier small-sample claim of positive correlation with global temperature. By contrast, coarse-scale permutation entropy and mean intensity track lake elevation strongly, with Spearman correlations of 0.85 to 0.88 and 95 percent confidence intervals excluding zero. The third principal component of the ResNet50 embeddings emerges without supervision as an AI climate axis, correlating 0.86 with cumulative CO2 and -0.83 with lake elevation. Image complexity leads lake stage by about three years, with Pearson r = 0.58 at lag +3 and a 95 percent confidence interval of 0.40 to 0.73. STL decomposition reveals a non-monotonic trend that rises from 1984 to 2015 and declines sharply thereafter as the lake approaches record-low elevations. Partial correlations controlling for month and sensor confirm robustness to seasonal and sensor effects. These results refine the art-as-thermometer metaphor into an art-as-leading-indicator-of-hydrological-state interpretation. The dataset, feature pipeline, and analysis code are released as a public benchmark.

[188] SPICE: Simple Polysemantic Feature Interpretation via Clustering-based Explanation cs.LG | cs.CVPDF

Sehyun Lee, Dahee Kwon, Damin Lee, Jaesik Choi

TL;DR: SPICE提出一种面向视觉网络多义神经元的聚类解释框架，无需架构专用传播规则或预设概念簇数，可自动推断每个神经元的概念簇数量，并用于比较CNN与Transformer中的多义性。

Details

Motivation: 单个神经元可能响应多个无关概念，导致其功能难以解释。既有方法常依赖特定网络架构和人工启发式，例如预先设定聚类数K，这限制了方法的通用性、可扩展性以及对现代Transformer的适用性。

Result: 论文提出SPICE（基于聚类解释的简单多义特征解释框架），用于分析深度视觉架构中的多义性。该框架避免架构依赖的特征传播规则，并自动确定单个神经元对应的概念簇数量，从而无需预设K。作者据此首次系统比较CNN与Transformer中的多义性，并研究多义性如何形成、随网络深度和架构如何变化，以及其对应的不同计算路径。摘要未给出具体模型、数据集或定量比较结果。

Insight: 移除架构专用处理和手动指定簇数，使多义特征分析更有可能扩展到不同视觉网络；该工作贡献侧重分析框架和机制研究，摘要没有报告定量性能或解释质量验证，无法据此判断其解释准确度。

Abstract: One of the pivotal recent challenges in neural network interpretability is polysemanticity, where a single neuron is activated by multiple, often unrelated concepts, hindering clear functional understanding. Although prior work has explored this phenomenon, existing approaches remain architecture-specific and depend on manual heuristics such as a fixed number of concept clusters ($K$), limiting their generality and scalability–especially for modern Transformer-based models. To address these limitations, we introduce SPICE (\textbf{S}imple \textbf{P}olysemantic Feature \textbf{I}nterpretation via \textbf{C}lustering-based \textbf{E}xplanation), a generalizable framework for analyzing polysemanticity in deep vision architectures. SPICE avoids architecture-dependent propagation rules, enabling the first systematic comparison of polysemanticity across both CNNs and Transformers, and automatically determines the number of concept clusters per neuron, eliminating reliance on a preset $K$ and supporting scalable analysis for large models. Using SPICE, we conduct a comprehensive investigation into how polysemanticity emerges, varies across depth and architecture, and forms through distinct computational pathways.

[189] Bellman Policy Optimization cs.LG | cs.CL | math.OCPDF

Zhuoqing Song, Haotian Xu, Xikun Zhang, Lidong Bing

TL;DR: Bellman Policy Optimization（BPO）是用于可验证奖励大语言模型强化学习的无评论家方法，以Bellman方程将策略镜像下降改写为终局奖励下的轨迹级目标，避免估计中间状态价值；数学推理基准实验显示方法有效。

Details

Motivation: 可验证奖励强化学习能提升大语言模型推理能力，但策略优化通常需要评估中间状态价值。对于自回归生成且奖励只在终局给出的任务，准确估值困难，增加评论家模型也会带来额外复杂度。

Result: 作者从策略镜像下降（PMD）推导BPO。针对终局奖励的自回归生成，利用Bellman方程将PMD重写为轨迹级目标，从而无需估计中间状态值。论文证明重写目标与原PMD目标拥有相同且唯一的最优解，并在近似该目标的实际损失中采用互补token概率比的平滑比值作为失配校正权重。数学推理基准实验显示BPO有效；摘要未提供基准名称或具体分数。

Insight: Bellman重写使终局奖励场景下的策略优化可以直接围绕整条生成轨迹构造，而不需要单独的状态价值估计；平滑校正权重用于处理实际目标近似带来的分布失配。理论最优解等价不自动保证近似损失的训练效果，摘要也未给出与基线的定量差异。

Abstract: Reinforcement learning with verifiable rewards (RLVR) improves the reasoning capabilities of large language models (LLMs). We introduce Bellman Policy Optimization (BPO), a critic-free method derived from Policy Mirror Descent (PMD). For autoregressive generation with terminal rewards, BPO uses the Bellman equations to reformulate PMD as a trajectory-level objective. The reformulation avoids estimating state values at intermediate states. We prove that it has the same unique optimal solution as the original PMD objective. We derive the practical BPO loss by approximating this objective. Its mismatch-correction weight is a smoothed ratio of complementary token probabilities. Experiments on mathematical reasoning benchmarks demonstrate the effectiveness of BPO.

[190] A Chosen Future Can Still Be Rewritten: Causal Writability in Video Models cs.LG | cs.CVPDF

Xingyun Wang, Haomin Zheng, Man Yuan, Leqian Yang, Ziming Liu

TL;DR: 该研究提出视频模型中的“因果可写性”：即便模型在冲突条件下生成错误运动，内部仍可能保留正确物理信号，并可通过低维因果编辑重新控制输出；固定强度写入存在深度边界，更强的后续写入可恢复运动，过量则会过冲。

Details

Motivation: 视频模型生成物理错误运动时，问题可能是模型没学会正确规律，也可能是学到了规律但生成时没有使用。区分这两种情况有助于理解错误成因、内部表征可控性及训练为何会修正某些错误而保留另一些错误。

Result: 作者训练一种场景：红色物体缓慢振荡、蓝色物体快速振荡；测试时输入一个观测运动很快的红色物体。模型即使在冲突条件下生成慢运动，也可通过由简单物理变量预测的低维编辑恢复正确快运动，这种能力被称为因果可写性。在固定编辑强度下存在清晰深度边界：边界前编辑会改变视频，边界后不再起作用，表示该次写入已提交；正确运动信号仍然保留，更强的下游写入可将其恢复，但增益过高会过冲。训练后会被纠正的错误在更多网络深度上可写，而持续存在的错误可写深度更少。作者还在预训练1.3B视频模型中复现因果可写性及其清晰关闭边界。

Insight: 模型输出错误不必然意味着相关物理规律未被学到；信息可能仍在内部，只是在生成控制中被压制。编辑可写范围及其随深度关闭的边界，提供研究表示是否可被因果变量重新调用的新方法；摘要基于特定振荡场景，跨任务普适性仍需更多验证。

Abstract: When a video model generates physically incorrect motion, did it fail to learn the correct motion, or did it learn it but fail to use it? We show the latter: the correct motion remains available inside the model and can still be made to control the generated video. We train on videos where red masses oscillate slowly and blue masses oscillate quickly, then test a red mass with fast observed motion. Even when the model generates slow motion in this conflicting case, a low-dimensional edit predicted from simple physical variables restores the correct fast motion. We call this ability causal writability. At fixed strength, we find a sharp depth boundary: the same edit changes the video before the boundary but not after it. This closure marks commitment for that write. The motion signal nevertheless remains, and a stronger downstream write can restore physical motion, while excessive gain overshoots. Early causal writability predicts which errors training later corrects: those errors are writable at more network depths than errors that persist. We reproduce both causal writability and its sharp closure in a pretrained 1.3B video model, supporting generality across model scale and training regime.

cs.MA [Back]

[191] From Process Loss to Assembly Bonus: Human-Grounded Diagnosis of Multi-Agent LLM Collaboration cs.MA | cs.AI | cs.CLPDF

Ala N. Tak, Teruhisa Misu, Kumar Akash, Zhaobo K. Zheng, Kevin H. Joo

TL;DR: 研究比较人类与 LLM 群体讨论中的过程机制，发现两者都有组装增益不对称，但 LLM 群体更从众、独特信息更少且更早收敛。

Details

Motivation: LLM 群体常用于模拟人群或协同推理，仅看最终答案不足以判断它们是否通过人类相似的讨论机制得到结果。

Result: 作者先比较 Wason 式演绎任务中的人类群聊与匹配 LLM 讨论，再检查类比、溯因和分析任务。两类群体都表现出组装增益：讨论更常改善平均成员而非最初最佳成员。初始答案多样性可解释模型异质性效应，并会同时增加纠正与破坏性改变。相较人类，LLM 群体更频繁跟随多数、提供较少独特信息且更早收敛；正确少数意见主要在较早重新表达后才成功。受人类群体研究启发的干预仅带来有限改进，未消除协调瓶颈。

Insight: LLM 群体可以复现部分结果层面的集体增益，但产生增益的过程仍与人类不同；评估多代理协作需报告信息贡献、异议处理和收敛过程，而不只比较最终正确率。

Abstract: LLM agents are increasingly used for collaborative problem solving and human-group simulation. This makes outcome-only evaluation insufficient: if LLM groups are used as models of human groups, we need to know whether they succeed or fail through human-like deliberative mechanisms. We compare human group chats with matched LLM deliberation traces on Wason-style deductive reasoning, then test whether the same process signatures generalize to analogical, abductive, and analytical tasks. Humans and LLMs show the same assembly bonus asymmetry: discussion improves the average member more often than the best initial member. Initial-answer diversity accounts for the effect of model heterogeneity, increasing movement in both corrective and destructive directions. The main differences are process-level. Compared with humans, LLM groups follow majorities more often, surface less unique information, and converge earlier; correct minority signals succeed mainly when re-expressed early. Interventions motivated by human group-decision research yield modest improvements in collective outcomes, but do not remove the coordination bottleneck. Together, these results suggest that LLM groups can reproduce some outcome-level patterns of human deliberation while diverging in the mechanisms that generate assembly bonus and process loss, with implications for group simulation and human-AI collaboration.

[192] Loop-Back Authority in LLM Agent Teams: A Paired Experiment on Flat and Hierarchical Coordination cs.MA | cs.AI | cs.CL | econ.GNPDF

Burak Agachan, Max van Duijn, Amirhossein Zohrehvand

TL;DR: 配对实验只改变经理代理能否驳回工人结果并要求返工，发现扁平协作在报告效用和清晰度上优于层级协调，返工循环增加了含糊表达和 token 成本。

Details

Motivation: 生产多代理系统常用经理审核工人输出的层级模式，组织理论预期其加速收敛，但 LLM 迎合和思维退化研究则预期权威批评可能使输出变差；以往比较难隔离权威链路本身。

Result: 固定五个模型、角色、提示、工具和数据，仅切换经理是否拥有拒绝返工权；对 43 对报告、86 次运行，用五模型评审面板和确定性规格检查评分。扁平结构在效用（d=0.42,p=0.009）和写作清晰度（d=0.34,p=0.030）更高；两种报告长度相同，但层级报告 hedge 多 53%，每次修订循环与清晰度下降 0.14 分相关。规格准确率均达天花板，监督层级多耗 51.5% token 且没有质量增益。

Insight: 主管只有在能客观验证时才有价值，只有主观点评时可能损害文本质量并增加成本；结果限于商业智能报告任务和一次成对实验。

Abstract: Hierarchical orchestration, in which a Manager agent reviews worker output and can send it back for revision, is the default coordination pattern in production multi-agent LLM frameworks. Classical organizational theory predicts that the authority link speeds convergence on decisive output; work on sycophancy and Degeneration-of-Thought predicts that authoritative critique makes LLM output worse. Prior comparisons vary whole frameworks on tasks with checkable answers, leaving the authority link untested on open-ended work. We present a paired experiment that holds five LLM agents, their roles, prompts, tools, models, and data fixed and varies one link: whether the Manager may reject a worker’s output and oblige a revision. Across 43 paired products and 86 runs of a business-intelligence reporting task, a five-model judge panel and a deterministic specification check score every report. The flat organization scores higher on Utility (d = 0.42, p = 0.009) and on Writing Clarity (d = 0.34, p = 0.030); the classical prediction fails. The reports are the same length, but hierarchical reports hedge 53% more, each revision loop is associated with a 0.14-point drop in Writing Clarity, and the hierarchical Writer’s first draft is indistinguishable from the flat report: the gap opens inside the revision loop. Specification accuracy is at ceiling in both organizations, and the supervisory tier costs 51.5% more tokens for no quality gain. A supervisor pays for itself when it can verify and becomes a liability when it can only opine.

cs.RO [Back]

[193] Language-Guided Representation Learning for Robust Cross-Sensor Material Recognition cs.RO | cs.CL | cs.CVPDF

Mashood M. Mohsan, Muhayy Ud Din, Binzhao Xu, Ahmad Abubakar, Irfan Hussain

TL;DR: 论文用自然语言触觉属性作跨传感器监督，将不同硬件的视觉触觉图像对齐到共享语义空间，以提升少样本材料识别和传感器迁移。

Details

Motivation: 机器人操作需要触觉判断软硬、纹理和接触稳定性，但视觉触觉传感器受光学、弹性体和照明影响，同一材料在不同设备上的图像差异大，单传感器训练泛化差。

Result: 作者构建 3.9 万条人工标注材料类别的触觉—语言数据，以高层语义（粗糙、柔软、滑等）作为硬件不变监督，将特定传感器图像表征与语言嵌入对齐。100-shot 设置准确率为 95%，跨传感器迁移平均提高 13.3 个百分点；在六个现有触觉数据集上最高提高 19%。

Insight: 语言语义可作为不同触觉设备间的共享对齐空间，减少对特定硬件外观的过拟合；结果覆盖六种数据集，但实际机器人接触任务的操作收益未在摘要中报告。

Abstract: Robots need touch to manipulate objects safely and reliably, as many properties, such as softness, texture, and contact stability, are hard to infer from vision alone. However, vision-based tactile sensors yield different observations of the same material due to variations in optics, elastomer properties, and illumination, leading to poor generalization when trained on a single or multiple sensors. We propose a language-guided distillation framework for learning sensor-robust tactile representations. Language encodes high-level semantic properties of touch (e.g., rough, soft, slippery) that remain invariant across sensing hardware, providing a natural sensor-agnostic supervisory signal. We construct a 39K-sample touch-language dataset with human-annotated material labels and train a tactile encoder to align sensor-specific tactile images with language embeddings in a shared semantic space. We evaluate our approach for few-shot learning and cross-sensor transfer and benchmark it on six existing tactile datasets. Our method achieves 95% accuracy in the 100-shot setting, improves cross-sensor transfer by an average of 13.3% accuracy, and yields up to 19% accuracy gains across six existing tactile datasets. These results demonstrate that language-guided distillation enables scalable and hardware-agnostic tactile representation learning. Code and dataset are available at https://mashood3624.github.io/Language_Tactile/

[194] Physical Kernel: Structured Visual Latents for Dark Manipulation cs.RO | cs.CVPDF

Jinting Hang, Hong Li, Zhenhui Cai, Zhihao Zhao, Jian He

TL;DR: 该研究测试“暗操作”机器人操控：模型只在初始短暂可见时写入潜在状态，之后仅凭潜状态和动作预测完成接触丰富技能；在ManiSkill StackCube五阶段链上成功率68.1%，但写入时外观偏移仍会使其失效。

Details

Motivation: 真实操控过程中视觉观测可能因遮挡、黑暗或带宽限制而不可用。作者研究能否在初始视觉阶段写入结构化潜变量，之后在无新图像条件下依靠学习到的动作动力学继续控制，并区分优势来自视觉表征、几何先验还是简单的冻结编码。

Result: 方法先通过RGB写入潜状态z0，后续策略根据潜变量选动作、开环动力学更新状态，不再接收像素。在ManiSkill StackCube（n=160，两个种子包）五级状态链上，暗操作方案成功率68.1%，逐步重新编码可见图像为35.6%，冻结编码或黑屏编码为0%。统一写入后遮挡实验中，遮挡及摄像机对齐的真实接触邻域遮罩使可见图像策略抬升成功率由43%降至0%，暗操作方案保持82.5%；打乱动作令动力学MSE约增9.4倍。写入阶段的夜间外观变化使堆叠成功率降至0%，但同样变化若发生在保持阶段，不影响抬升；停止写入的时机与早停、模糊/JPEG退化对结果有显著影响。专用写入策略在视觉预算设置下成功率35%（n=80），无特权几何的Dreamer式和像素空模型为0%；使用9维暗观测的特权状态RSSM-MPC约35%，作者说明这是更强观测的空模型，不是严格匹配视觉基线。

Insight: 结果支持潜变量动力学在观测中断后维持操控，但性能依赖初始写入质量：写入阶段的外观偏移很脆弱，而保持阶段偏移影响较小。对照协议、样本量和是否使用特权信息并不完全一致，比较时需按论文标注区分，不能把所有差异归因于表征方法。

Abstract: We study dark manipulation: after a brief lit Write encodes z0 = Enc(rgb), a policy pi(z) and open-loop dynamics f(z,a) complete contact-rich skills without further pixels (dark_f). On ManiSkill StackCube (n=160; seed packs 0/1000), dark_f attains 68.1% stacked on the five-rung chain (near_A -> grasped -> lifted -> on_B -> stacked), compared with 35.6% for per-step lit_reenc and 0% for freeze/encode_black. On a shared Write->HOLD protocol (n=40), occlusion and camera-aligned GT contact-neighbor masks drive lit lift from 43% to 0%, while dark_f holds 82.5%; shuffling actions inflates dynamics MSE by ~9.4x; write-time appearance shifts break encoding (night: 0% stacked), yet the same shifts during HOLD leave dark_f lift unchanged; Write length Tw is flat once the stop phase is reached, while earlier stops and write-time blur/JPEG sharply cut stacked. A dedicated pi_write reaches 35% vision-budget stacked (n=80); matched Dreamer-style/pixel nulls without privileged geom stay at 0%. Privileged state-RSSM MPC reaches ~35% stacked with 9D dark observations – a stronger-observation null, not a matched visual baseline.

[195] Compositional Shift Algebra: Extrapolating Mixed Robot Shifts Without Mixed Finetuning cs.RO | cs.CVPDF

Jinting Hang, Zhenhui Cai

TL;DR: Compositional Shift Algebra从精确重置探测中学习观察、策略和动力学各自的单因素迁移算子，再组合外推未见过的机器人混合迁移，无需混合迁移微调；在多个ManiSkill任务上组合策略接近Oracle，但实验同时显示部分基线和部署条件仍需进一步适配。

Details

Motivation: 机器人部署中摄像头、动作接口和物理动力学常同时变化。现有适配方法要么为每种变化组合单独微调，要么尝试挑选需要更新的模块，难以扩展到未见的混合分布偏移。

Result: CSA在模块化系统z=E(o)、a=g(z,u)、z’=f(z,a)上，利用精确重置探测分别学习观察、策略和动力学的单因素迁移算子，再以算子组合预测留出的混合迁移，无需混合偏移微调。在ManiSkill StackCube上，残差CSA对10个种子的留出混合偏移达到1.0成功率，与混合偏移Oracle相当，较最佳单项迁移、零样本及参数平均基线高约67个百分点；RGB-D闭环组合也接近Oracle并明显超过非组合方案。延迟交换子的压力测试显示策略与延迟算子的组合次序有必要。在PickCube上CSA组合为1.0，对比非组合0.33（n=10）；不使用特权立方体/目标位姿和抓取标志的L1视觉控制器组合成功率0.95，对比0.00。扩展结果中PushCube提升33个百分点，PegInsertion两种动作表示各提升67个百分点；若干深层BC、公平适配基线和视觉定位BC仍需要组合适配，延迟场景也更适合有序、少样本部署。作者还报告一种干预门控适配作为负对照。

Insight: 若不同类型迁移可近似分解为独立算子，组合单因素经验就可能外推未训练过的混合变化；同时算子顺序对某些因素（如延迟）有影响。成功率1.0来自有限的10种子评测，且多个补充设置需要组合适配，说明不是所有迁移都能靠简单模块叠加解决。

Abstract: Robot deployments rarely change one mechanism at a time: cameras, action interfaces, and physical dynamics often shift together. Prior adaptation recipes either finetune a new model for every mix or attempt to select which module to update. We instead learn shift operators on a modular stack z{=}E(o), a{=}g(z,u), z’{=}f(z,a) and compose them. Compositional Shift Algebra (CSA) fits single-factor observation, policy, and dynamics operators from exact-reset probes, then extrapolates held-out mixed shifts by operator composition—without mixed-shift finetuning. On ManiSkill StackCube, residual CSA matches an oracle mixed inverse on held-out mixes (success 1.0 over 10 seeds) while beating best-single / zero-shot / parameter-average baselines by { approx}67 pp. RGB-D vision-in-the-loop composition remains near oracle and far above non-compositional arms; a delay commutator stress shows ordered necessity for policy timesdelay. On a second task (PickCube), residual CSA again reaches compose 1.0 vs. 0.33 non-compositional (n{=}10), and an L1 vision controller without privileged cube/goal poses or grasp flags in the control loop retains compose 0.95 vs. 0.00. Main-track upgrades freeze PushCube (+33 pp), PegInsertion joint8 / pose7 EE (+67 pp each), and thin BC under frozen CSA (+67 pp); deeper BC and fair adapt baselines still need compose (+67 pp each), vision-localized BC needs compose (+56 pp), and delay favors ordered/few-shot deploy. We report Intervention-Gated Adaptation as a negative control.

[196] From Vision to Harvest: Benchmarking Vision-Language Models for Multi-Arm Robotic Fruit Harvesting cs.RO | cs.CVPDF

Vrishan Inukollu, Adyan Zaman, Anvi Kudaraya, Carlos Lazcano, Yuankai Zhu

TL;DR: 该论文建立首个预训练VLM零样本多机械臂果园采摘规划基准，使用真实苹果和柑橘果园图像，让VLM直接生成多臂采摘顺序与路径点，再以轻量轨迹验证器检查碰撞；结果显示其有潜力，但3D路径点精度和多臂避碰协调仍是部署瓶颈。

Details

Motivation: 多机械臂采摘有望提升效率并减少人工需求，但系统要在不同环境中泛化，还需协调共享作业空间中的机械臂。既有方案可能依赖目标环境的大量数据，或采用限制规划质量的简化假设。

Result: 作者提出针对预训练视觉语言模型的零样本多臂水果采摘规划综合基准，采用真实苹果与柑橘果园图像，并比较VLM规划流程和传统感知—规划流程。VLM直接生成每个机械臂的采摘序列及路径点，轻量轨迹验证器检查碰撞。实验表明，前沿VLM能在零样本条件下生成有效的多臂采摘计划；但实际部署仍受精确3D路径点生成和碰撞感知协调限制。摘要未给出基准规模、各模型得分或相对传统管线的量化结果。

Insight: 让VLM端到端提出多臂任务序列与路径可以减少对目标果园专门训练数据的依赖，但语言—视觉能力并不自动解决精确三维几何和共享空间避碰。该工作主要贡献是基准及初步能力评估，摘要中的“有效”缺少量化细节。

Abstract: Multi-arm robotic harvesting offers a promising path to improve harvesting efficiency and reduce reliance on manual labor. However, practical deployment remains challenging because the system must generalize across diverse environments while efficiently coordinating multiple arms in a shared workspace. Existing methods often require substantial data collection in target environments or rely on simplifying assumptions that limit planning quality. In this work, we introduce the first comprehensive benchmark for evaluating pretrained Vision-Language Models (VLMs) on zero-shot multi-arm fruit harvesting planning. Our benchmark uses real-world apple and citrus orchard images and compares a VLM-based planning pipeline with a traditional perception-and-planning pipeline. The VLM pipeline directly generates harvesting sequences and waypoints for each arm, while a lightweight trajectory verifier checks for collisions. Our results show that frontier VLMs can generate effective multi-arm harvesting plans zero-shot, but a practical deployment remains limited by accurate 3D waypoint generation and collision-aware coordination. These results highlight both the promise and current limitations of pretrained VLMs for multi-arm robotic harvesting.

[197] What Makes an Efficient VLA? Navigating Action-Head Design, Scaling, and Latency cs.RO | cs.CVPDF

Luoyang Sun, Guoyang Xia, Fengfa Li, Lei Ren, Xinyu Cui

TL;DR: EffVLA研究VLA动作头初始化、规模和实测延迟的受控权衡，发现将语言骨干末层复制到动作头是最稳定且无额外延迟的提升来源；据此提出紧凑模型，在LIBERO表现匹配强开源模型、LIBERO-Plus多数扰动轴领先，并迁移到真实SO-ARM101。

Details

Motivation: VLA由预训练视觉编码器、语言骨干和动作头组成，但各部分在受控训练和配对延迟条件下的相对贡献仍不明确。模型规模和推理轮数的收益也可能取决于动作头是否已与语言表示对齐。

Result: 作者固定SigLIP2、Qwen2.5骨干系列和训练流程，系统比较动作头设计、模块规模及设备端延迟。实验发现，性能主要受初始化而非解码器结构、损失或推理预算影响：将语言骨干最后几层Transformer复制到动作头，是增益最大的单项因素，不增加延迟且在所有模块规模下有效。流匹配和更重解码器仅在动作头未对齐时有利，对齐后收益反转；额外推理轮次未见可测收益。作者把对齐后的注意力集中于指令中的物体名、参数空间接近骨干，解释为表示迁移，但承认尚未通过直接因果实验验证。容量只有在对齐后才有收益；规模回报在当前π系列VLA附近迅速递减。基于这些结果提出EffVLA：标准LIBERO上匹配最强开源VLA，在LIBERO-Plus多数扰动轴领先且延迟更低，并以不变配方迁移到真实SO-ARM101机械臂。

Insight: VLA扩容前应先让动作头复用并对齐语言骨干表示；在动作头未对齐时，增加容量或推理计算可能是在补偿表示失配，而非带来稳定扩展收益。论文清楚区分测量发现与“表示迁移”的机制解释，后者仍待专门控制实验验证。

Abstract: Vision-Language-Action (VLA) models combine a pretrained vision encoder, a language backbone, and an action head, but their relative contribution has not been established under controlled, latency-paired conditions. We fix the backbone families (SigLIP2 and Qwen2.5) and the training pipeline, sweep action-head design and module scale, and pair each configuration with measured on-device latency. The study yields three findings. First, action-head performance is governed primarily by initialization rather than decoder architecture, loss, or inference budget: copying the last transformer layers of the language backbone into the head is the single largest lever, at no latency cost, and the only axis that helps at every module scale. Alignment also explains the other axes: flow matching and a heavier decoder pay off only while the head is misaligned and reverse once it is aligned, and extra inference passes give no measurable benefit; expressiveness appears to substitute for missing alignment. We read this as representation transfer: the aligned head keeps attending to the instruction’s object nouns and stays close to the backbone in weight space rather than relearning to act from scratch. Because we reach alignment only through initialization, we offer this as the account that best organizes the measurements, not a demonstrated cause, and name the control that would settle it. Second, capacity pays only after alignment: the aligned action head is the highest-return module to scale. Third, those returns diminish sharply near the size today’s $π$-series VLAs already use, so further growth buys little in-domain accuracy for its latency. These specify EffVLA, a compact model matching the strongest open-source VLAs on standard LIBERO, leading on most LIBERO-Plus perturbation axes at lower latency, and transferring to a real SO-ARM101 arm with the recipe unchanged.

[198] Bench2Dex: Benchmarking Visuo-Tactile Bimanual Dexterous Manipulation Across Dexterous Hands cs.RO | cs.AI | cs.CVPDF

Zhenjie Yang, Yideng Zhang, Dongjie Zhang, Chenyu Jiang, Xianshuai Liu

TL;DR: Bench2Dex是跨12种灵巧手的视觉—触觉双臂操作仿真基准，提供26项任务、约1,300条遥操作演示和统一图像式触觉接口，并区分扰动下策略应保持不变或随之变化的鲁棒性轴；基准旨在提供研究环境，不声称模拟触觉可替代真实传感器。

Details

Motivation: 触觉能补充视觉难以推断的接触信息，但灵巧手的手指结构、接触面和传感器布局尚无统一设计，模拟触觉也与真实传感器输出存在差异。这使得跨不同灵巧手系统研究视觉—触觉操作较难进行一致比较。

Result: Bench2Dex为12种灵巧手统一接入仿真触觉接口，将局部接触几何转换为图像式触觉观测，但不试图复现任何特定真实传感器。基准含26项双臂任务，涵盖工具使用、可动结构交互和多阶段操作，并提供约1,300条人工遥操作示范，以及同步的视觉、触觉、本体感觉、动作、物体状态观测和可执行任务指标。七种扰动分为不变性轴（正确动作不变）与等变性轴（正确动作随扰动改变）。作者评测ACT、Diffusion Policy、pi0.5和GR00T N1.5，并报告其性能和失败模式。

Insight: 统一触觉接口和明确区分不变/等变扰动，有助于分析策略究竟该忽略何种变化、对何种变化做动作调整。该基准支持算法开发与比较，但作者特别说明仿真触觉不应被视为真实触觉感知的替代品；摘要未列出模型分数。

Abstract: Tactile sensing provides contact information that can be difficult to infer from vision alone, but tactile hardware for dexterous hands has not converged to a common design. Dexterous hands differ in finger structure, contact surfaces, and sensor layouts, while simulated tactile signals still differ from measurements produced by physical sensors. These factors make it difficult to study visuo-tactile manipulation across diverse dexterous hands within a consistent experimental setting. We present Bench2Dex, a simulation benchmark for visuo-tactile bimanual manipulation across 12 dexterous hands. We adapt existing robot models with a shared simulated tactile interface that converts local contact geometry into image-like tactile observations. The interface provides a consistent observation format across different hand morphologies without attempting to reproduce the output of a specific physical tactile sensor. Bench2Dex includes 26 bimanual manipulation tasks that involve tool use, articulated-object interaction, and multi-stage manipulation, together with about 1.3K human-teleoperated demonstrations. The benchmark provides synchronized visual, tactile, proprioceptive, action, and object-state observations, together with executable task metrics. For robustness, we group seven perturbation types into invariance axis, where the correct action does not change, and equivariance axis, where the correct action changes together with the perturbation. We evaluate ACT, Diffusion Policy, pi0.5, and GR00T N1.5 on Bench2Dex and report their performance and failure modes. Bench2Dex is meant as a platform for studying visuo-tactile learning across dexterous hands. It does not assume that simulated tactile observations can replace real tactile sensing; it offers a shared setting for algorithm development while tactile hardware and simulation models are still evolving.

cs.SE [Back]

[199] woma: a real-time foundation model and its fine-tuned models for endoscopy cs.SE | cs.CV | cs.LGPDF

Thang Tran, Lan Dang

TL;DR: woma是以约百万张内镜帧进行无标签训练的实时基础模型，并提供结肠镜和胃镜微调任务模型；在未见数据上报告息肉检测、解剖定位和病灶识别指标，单工作站GPU处理1080p任务约100帧/秒，还提供仅依赖自研Vulkan内核的部署构建。

Details

Motivation: 内镜AI要从实验走向生产，需要可复现的模型设计、严格的独立测试和高效易部署推理。研究者希望同时提供无标签基础模型及针对临床任务的微调模型，并预先确定筛选标准、报告所有结果，避免只挑选成功指标。

Result: woma使用约100万张内镜帧无标签训练，并通过单一自包含的软件库完成需求及通过标准预注册、按规则筛选八个候选、自监督训练停止、任务微调和部署优化。结肠镜模型检测和分割息肉、识别结肠段、建议息肉类型并评估肠道准备；胃镜模型在22个规范部位中识别站点、标记病灶并区分七种发现。所有数值均来自训练未见数据。在六院PolypGen数据集上检出96%的息肉，精确率至少0.85；15段完整REAL-Colon视频中的19个息肉全部检出，每例误报1.6次。胃镜在未见患者的92%帧上正确识别标志区域，并检出39个留出肿瘤性病变帧中的37个，特异度0.91。单工作站GPU上所有任务处理1080p视频约100帧/秒，并在四种精度设置下快于PyTorch、ONNX Runtime和TensorRT；与TensorRT相比，基础模型单次通过耗时少3%–27%，端到端帧率高6%–31%。另有只用自研Vulkan内核、无需供应商库的构建版本；同卡上f32速度快于CUDA构建，可仅部署两个文件且无需工具包、cuDNN或框架。

Insight: 论文强调从预注册筛选到独立测试和部署基准的完整生产流程，并把轻量运行时作为模型贡献的一部分。各临床检测数字对应不同数据集和任务，摘要未提供置信区间；约100帧/秒是特定工作站GPU上的报告，不代表所有部署环境。

Abstract: woma is a real-time foundation model for gastrointestinal endoscopy: a network trained without labels on about a million endoscopy frames, from which task models are fine-tuned. We contribute a systematic design for production. Requirements and pass marks were fixed before any run, eight candidates screened under pre-registered rules, self-supervised training taken to a stopping rule, then fine-tuning and deployment optimisation, all on one self-contained library, numbat. We also contribute woma itself with two fine-tuned models, every outcome reported met or missed. Our colonoscopy model finds and outlines polyps, names which colon segment is in view, suggests polyp type and grades bowel preparation. Our gastroscopy model names a station out of 22 protocol sites, flags and outlines lesions, and names one of seven findings. Every number was read on data never seen in training, and shipped weights were chosen on that record. In colonoscopy, 96% of polyps in a six-hospital PolypGen set are found at precision >=0.85, and 19 of 19 polyps across fifteen full REAL-Colon videos at 1.6 false alarms per procedure. In gastroscopy, landmark region is named correctly on 92% of frames from unseen patients, and 37 of 39 held-out neoplasia frames are flagged at specificity 0.91. On one workstation GPU every task runs over 1080p video at about 100 frames per second, faster than PyTorch, ONNX Runtime and TensorRT in all four precision regimes tested. TensorRT comes closest: one pass of our foundation model takes it 3 to 27% longer than ours, and we deliver 6 to 31% more frames per second from frame to results. A second build links no vendor library at all – our own kernels over Vulkan – so a site deploys two files and needs no toolkit, no cuDNN and no framework; in f32 it beats the CUDA build on the same card.

eess.AS [Back]

[200] Exploring Multimodal Turn-Taking Cues in Face-to-Face Conversation using Voice Activity Projection eess.AS | cs.CL | cs.SDPDF

Willem Berner, Julio Cesar Cavalcanti, Kalle Åström, Gabriel Skantze

TL;DR: 研究将注视、头动、身体手势和面部动作单元加入 VAP 语音活动投影模型，发现视觉线索能提升面对面对话轮替预测，其中面部动作信息最有用。

Details

Motivation: 人类轮替依赖言语和非言语信号，而对话系统通常只使用音频；需要评估面对面交流中的视觉线索能否改善未来语音活动预测。

Result: 作者基于 Meta Seamless Interaction 双人对话数据扩展自监督 VAP Transformer，加入注视方向、头部运动、身体/手部姿态和面部动作单元，比较拼接、交叉注意力、差分特征与可学习门控融合。视觉信息整体优于音频基线；面部动作单元显著比其他单类特征更有信息，所有特征组合最好，说明身体和注视可提供互补信号。自然对话与即兴表演数据混合方式会影响特定任务表现。

Insight: 轮替时机需要结合面部和身体线索，单独依赖音频会丢失重要非语言信号；模型在自然与表演对话间存在域差异，部署训练数据应匹配目标交互场景。

Abstract: Turn-taking is a fundamental component of spoken interaction, and while humans naturally rely on both verbal and non-verbal signals, dialogue systems usually depend on audio cues alone. This paper investigates whether visual features from face-to-face conversations can enhance turn-taking prediction beyond what is achievable from audio-only. We extend the Voice Activity Projection (VAP) model, a self-supervised transformer-based model for predicting future voice activity, by incorporating visual features extracted from the large-scale Meta Seamless Interaction dataset of dyadic face-to-face conversations. The visual features include gaze direction, head movement, body and hand pose, and facial action units (FAU). For incorporating the visual features, we explore concatenation, cross-attention fusion, delta features, and trainable gating mechanisms. Results show that visual information improves performance over the audio-only baseline, with FAU being significantly more informative than other feature groups. Body and gaze features nevertheless contribute complementary information, as the model combining all features performs best. Furthermore, results indicate that performance on specific tasks varies depending on whether training and test data come from improvised (acted) or naturalistic (non-acted) conversations.

eess.IV [Back]

[201] Deformable 2D Gaussian Splatting for Efficient 4K Video Compression eess.IV | cs.CVPDF

Chenhao Zhang, Fengqing Zhu

TL;DR: 该方法用由粗到细的多尺度2D高斯泼溅表示视频GOP，并结合轻量形变网络进行4K视频压缩；实验中LPIPS率失真表现超过H.265及其他先进学习式视频压缩方法，强调高分辨率快速解码潜力。

Details

Motivation: 超高清视频存储和实时解码成本高。神经视频压缩和隐式神经表示虽有竞争力的率失真表现，却存在解码延迟高和内存占用大的问题；高斯泼溅渲染快速且视觉质量高，但在视频压缩中的应用仍较少。

Result: 论文提出实时视频压缩框架，以由粗到细的多尺度二维高斯泼溅结构表示并压缩一个图像组（GOP），再通过轻量形变网络描述变化。实验显示，在LPIPS率失真指标上，该方法超过H.265以及其他先进学习式视频压缩方法。摘要未报告码率、解码速度、内存使用、分辨率测试设置及具体对比数字。

Insight: 二维高斯泼溅的快速渲染可作为高分辨率视频压缩表示的候选方案，轻量形变模块用于跨帧建模；其摘要目前只支持LPIPS率失真优势，实时性、内存收益和4K测试条件仍需依据正文确认。

Abstract: Ultra-High-Definition (UHD) video presents significant challenges for efficient storage and real-time decoding. Learning-based methods, such as Neural Video Compression (NVC) and Implicit Neural Representations (INR), achieve competitive rate-distortion performance but suffer from high decoding latency and excessive memory usage. Meanwhile, Gaussian Splatting has recently attracted attention in the computer graphics community due to its ultra-fast rendering and high-fidelity visual quality. Despite these advantages, its application in video compression remains largely unexplored. To bridge this gap, we propose a real-time video compression framework that represents and compresses a Group of Pictures (GOP) using a coarse-to-fine multi-scale 2D Gaussian Splatting (2DGS) structure coupled with a lightweight deformation network. Experiments demonstrate that our method delivers rate-distortion performance in LPIPS that surpasses H.265 and other state-of-the-art learning-based video compression methods. Our work demonstrates the potential of Gaussian Splatting as a practical solution for efficient high-resolution video compression.

eess.SP [Back]

[202] From Semantic to Token Communication: The Next Paradigm for Large-Model-Driven 6G Intelligent Connectivity eess.SP | cs.AI | cs.CV | cs.IT | cs.LGPDF

Yu Ma, Zhen Gao, Li Qiao, Xiaoyuan Zhang, Mahdi Boloursaz Mashhadi

TL;DR: 本文综述从大模型驱动的语义通信走向TokenCom的思路，提出把多模态大模型原生token作为比特之上的通用通信抽象，以支持按语义重要性进行传输、纠错与资源分配，并梳理其在模型服务及具身智能中的方向和挑战。

Details

Motivation: 6G通信正从可靠传比特转向面向含义和任务的连接，大模型推动了语义通信发展，但现有方案通常绑定特定模态、模型或任务。数字比特有统一传输单位，语义表征却缺乏类似的通用单位，限制了互操作、理论统一和规模化系统设计。

Result: 论文主张以token作为候选抽象单位：一方面，多模态大模型已在统一token空间表示文本、图像、音频、视频和机器人动作；另一方面，专家路由、缓存传输及投机解码等分布式推理过程已产生大量token级流量。综述先回顾大模型语义通信的三类方向——信源侧语义编码、物理层任务的信道语义、边缘设备协同智能——再讨论TokenCom抽象及其所需传输技术，并介绍面向大模型服务、具身与智能体的两种新范式，最后归纳统一、可扩展、AI原生6G系统的开放问题。

Insight: TokenCom试图把通信粒度从任务专用的语义表示提升为模型自身处理的原生单位，从而让重要性分配、错误处理和资源调度直接作用于token。该文是综述和范式倡议，摘要提出候选方向与挑战，并未报告实现系统或实验性能。

Abstract: The ambitious requirements of sixth-generation (6G) networks are driving communication systems from reliable bit delivery toward meaning-aware and task-oriented connectivity. Large models (LMs), with strong multimodal understanding and generation capabilities, have accelerated this shift and made semantic communication (SemCom) increasingly practical. Yet current LM-driven SemCom remains fragmented: semantic representations are typically tied to specific modalities, models, or tasks. While the bit provides a universal unit for digital transport, there is still no analogous unit for representing and processing semantics, which limits interoperability, theoretical unification, and scalable system design. We argue that tokens provide a natural candidate for this missing abstraction. Two trends support this: unified multimodal LMs now encode text, images, audio, video, and robot actions in one token space, while distributed LM inference already generates substantial token-level traffic through expert routing, cache transfer, and speculative decoding. Token communication (TokenCom) emerges by unifying these trends, using the LM’s native processing unit as a communication abstraction above the bit level and enabling importance assignment, error handling, and resource allocation directly at token granularity. This survey traces the evolution from LM-driven SemCom to TokenCom. We review three major directions of LM-driven SemCom: source-centric semantic coding, channel semantics for physical-layer tasks, and collaborative edge-device intelligence. We then examine the token abstraction, the transmission techniques it requires, and two emerging paradigms, namely TokenCom for LM services and for embodied and agentic intelligence. Finally, we identify open challenges toward unified, scalable, and AI-native 6G communication systems.

Table of Contents

cs.CV [Back]

[1] What Does the Encoder Actually Decide? A Controlled Comparison of Vision Backbones on Joint Tree Segmentation and Stereo Depth cs.CVPDF

[2] Don’t Just Look, Intervene: Perturbation Based Region Labeling for VQA Images cs.CVPDF

[3] Part Grounding, Not Action Knowledge: Locating the Bottleneck in VLM Affordance Prediction cs.CV | cs.LG | cs.ROPDF

[4] Occlusal Geometry in Closed Form for Orthodontic Report Generation cs.CV | cs.CLPDF

[5] ArtSociety: Multi-Agent Multimodal Collaboration for Art Emotion Understanding cs.CVPDF

[6] SJD-SV: Speculative Jacobi Decoding with Semantics Verification for Autoregressive Image Generation cs.CVPDF

[7] Pixel-wise Planarity for High-Precision Monocular Plane Segmentation cs.CVPDF

[8] Evaluation of MLLM-Agnostic Plug-and-Play Keyframe Selection Methods for Long Video Understanding cs.CV | cs.AIPDF

[9] Preserving Subject-Clarity in Image Outpainting with Multiscale Wavelet Supervision cs.CVPDF

[10] (How) Do MLLMs Report Bistable Images Like Humans? cs.CV | cs.AIPDF

[11] Sampling headroom is not selection gain: a compute-value audit of test-time scaling for video world models cs.CV | cs.AIPDF

[12] A Comprehensive Review of Multimodal Facial State Analysis: Tasks, Methods, and Resources cs.CVPDF

[13] Interpretable Temporal Video Reasoning with EventGraph and EventField cs.CV | cs.CL | cs.IRPDF

[14] BEACON: Behavior and Appearance Control for Subject-Specific Video Generation cs.CV | cs.AIPDF

[15] TryOnReward: Learning Foveated Consistency for Reinforcement Fine-Tuning of Virtual Try-On cs.CV | cs.AIPDF

[16] SomBench: Benchmark Dataset for Advancing Machine Learning in Lunar Science cs.CV | cs.LGPDF

[17] Structure-Token Evidence-Anchored Reasoning for Scientific Chart Understanding cs.CVPDF

[18] Frame-Synchronous Hand Gesture Detection by Projected Winding Order cs.CV | cs.GR | cs.HC | cs.LGPDF

[19] RxScribe Bench: A Multi-Axis Benchmark for Evaluating Vision-Language Models on Indian Outpatient Prescriptions cs.CVPDF

[20] Multimodal-Multiresolution Foundation Model for Lunar Remote Sensing cs.CV | cs.AI | cs.LGPDF

[21] LLaDA-UI: Bringing Block-wise Diffusion to Vision-Language GUI Agents cs.CV | cs.AIPDF

[22] Target-Checked Reliability Score Refinement for Video Question Answering cs.CVPDF

[23] Harnessing Image Question Dependence for Better VLM Test-time Reinforcement Learning cs.CV | cs.LGPDF

[24] Adaptive Conformal Redistribution for Inter-class Transitional Uncertainty in Medical Image Classification cs.CV | cs.AI | cs.LG | stat.CO | stat.MLPDF

[25] GroundBench: A Factorized, Counterfactual Benchmark for Locating VLM Affordance Failures cs.CV | cs.LG | cs.ROPDF

[26] Task-Based CT Protocol Optimization Using Reinforcement Learning and Virtual Imaging Trials cs.CV | cs.AI | cs.LGPDF

[27] Pedestrian Crossing Intent Classification From Event-Based Vision Using Convolutional Spiking Neural Networks With Temporal Augmentation cs.CV | cs.AIPDF

[28] Global-Local Contextual Progressive Expansion Network for Martian Landslide Segmentation in Multimodal Remote Sensing Imagery cs.CVPDF

[29] RIGOR: Rig-Informed Geometry for Omnidirectional Reconstruction cs.CVPDF

[30] Towards Practical Precision Agriculture: Real-Time Fruit Detection and Video Analytics on Embedded Edge Hardware cs.CVPDF

[31] Multimodal Foundation Models Adaptation based on Domain-Aware Relaxed Orthogonal Subspace for Remote Sensing cs.CVPDF

[32] MARC: Morphology-Aware Regression of Consensus for Cell Segmentation in Subcellular Spatial Transcriptomics cs.CVPDF

[33] FFVO: A Feedforward Pose Decoder for Long-Horizon Visual Odometry cs.CVPDF

[34] Rank-Consistent Set Reasoning for Co-Salient Object Detection cs.CVPDF

[35] Hyper-LLaVA: Hyperbolic Uncertainty-aware Modality-Balanced Routing for Multimodal Continual Instruction Tuning cs.CVPDF

[36] Restore What Matters: Lessons from Joint Restoration and Recognition cs.CVPDF

[37] StepPrune: Adaptive Sequential Visual Token Selection across Multimodal Large Language Models cs.CVPDF

[38] Realtime-Venus: A full-duplex interaction system with asynchronous delegation cs.CV | eess.ASPDF

[39] Beyond OCR Accuracy: Text-Centric VQA Under Image Degradation with Modular and End-to-End cs.CVPDF

[40] DiVA: Enabling Interactive Digital Life Simulation via Video Models cs.CVPDF

[41] Physically Typed and Geometry-Aware Representations for Earth Foundation Models cs.CV | cs.LGPDF

[42] BLInD: Learning Driver Intent as a Distribution over Future Ego Trajectories cs.CV | cs.ROPDF

[43] SGWIB:Sliced Gromov-Wasserstein Information Bottleneck for Video Highlight Detection cs.CV | cs.AIPDF

[44] LPA-CWM: A Learned Physical Adjudicator for Motion Reasoning with Counterfactual World Models cs.CV | cs.AI | cs.ROPDF

[45] RA-CoA: Training-free Fashion Image Captioning via Retrieval-Augmented Chain-of-Attributes cs.CV | cs.AI | cs.LGPDF

[46] Adapting Open-Weight MLLMs to Generate Point Prompts for Electron Microscopy Segmentation cs.CVPDF

[47] Talking to Me or Someone Else? Rethinking Talk-to-Me Detection in Egocentric Videos cs.CV | cs.AIPDF

[48] SignMimic: Robust High-Quality Sign Language Motion Generation via Human-Shape-Oblivious Pose Transfer Guidance cs.CVPDF

[49] SPARK: Representation-Level KV Memory Alignment for Safer Vision-Language Models cs.CVPDF

[50] S3-Tracker: Self-Supervised Surgical Tissue Tracking With Contrastive Random Walks cs.CV | cs.LG | cs.ROPDF

[51] Vision-Language Models for Criterion-Level Grading of Handwritten Examinations in Outcome-Based Education cs.CVPDF

[52] Learning Continuous Source Responses For Generalizable AI-Generated Image Detection cs.CVPDF

[53] Two-Stage Mixture-of-LoRA for Multi-Task Medical Vision-Language Learning cs.CVPDF

[54] Rethinking Camouflage Image Generation towards a Training-Free Paradigm cs.CVPDF

[55] Contour-Guided Spectral Routing for Robust Real-Time Pedestrian Detection cs.CVPDF

[56] A Generative AI Integrated Multimodal Framework for Low-Latency Multi-Camera Person Re-Identification cs.CV | cs.AI | eess.IVPDF

[57] Lightweight Generalized DeepFake Face Detection with WAVIE: Wavelet Augmented Vision Intermediate Embeddings cs.CV | cs.AIPDF

[58] Context-Aware Mutual Learning for Blind Image Inpainting and Beyond cs.CVPDF

[59] From Visual Attribution to Clinical Reasoning: Explainable Parkinson’s Disease Screening from Hand-Drawn Patterns cs.CV | cs.AIPDF

[60] AlayaVista: Streaming World Modeling from Panoramic States to Perspective Video cs.CVPDF

[61] DynEoMT: Learning Object Dynamicity from Online Segmentation Queries cs.CV | cs.ROPDF

[62] PRI-Net: A Lightweight Multimodal Framework for 3D UAV Localization cs.CV | eess.SPPDF

[63] PuzzleMate: Benchmarking MLLMs for Egocentric Puzzle Assistance cs.CVPDF

[64] Open-UniMo: Towards Unified Motion-Language Understanding and Generation in the Open World cs.CV | cs.AIPDF

[65] Selective Tool Use for Agentic Change Visual Question Answering in Remote Sensing cs.CVPDF

[66] Compositional SVG Generation via VLM-Driven Hierarchical Semantic Parsing cs.CV | cs.AIPDF

[67] From Density to Biopsy Decisions and Malignancy Prediction: A Benchmark Study of Multimodal Large Language Models Against Radiologists in Digital and Contrast-Enhanced Mammography cs.CV | physics.med-phPDF

[68] Perceive, Refine, Reason: A Calibrated Pipeline for Measuring Indicators in Strategic Visual Communication on Social Media cs.CV | cs.CYPDF

[69] CrossDistill: Balancing Quality and Diversity via Trajectory-Level Hybrid Few-Step Distillation cs.CVPDF

[70] From Visual Feedback to Textual Reviews: A Multi-Agent Vision-Language Framework for Image-Grounded Review Assistance cs.CV | cs.AIPDF

[71] Zero-shot video highlight detection based on text descriptions and synthetic images cs.CVPDF

[72] MoVT: Video-Augmented Motion Tokenizer for Text-to-Motion Generation cs.CVPDF

[73] PhysBrain 1.5: From Vision-Language Models to Physical Foundation Models cs.CV | cs.ROPDF

[74] G-ray: Ray-Level Relative Geometric Position Encoding in Multi-View Vision Transformers under Camera Heterogeneity cs.CVPDF

[75] TwinICL: Diagnosing Multimodal In-Context Learning through Paired Counterfactuals cs.CV | cs.LGPDF

[76] Tele360: Real-Time Feed-Forward Human Reconstruction from Sparse Unposed Cameras cs.CVPDF

[77] LG-VLN: A Zero-Shot Vision-and-Language Navigation Framework with LangGraph State Orchestration cs.CV | cs.ROPDF

[78] AdaVSkip: Adaptive Visual Token Skipping Across Layers For Efficient MLLMs Inference cs.CV | cs.AIPDF