cs.CL [Total: 14]
cs.CV [Total: 74]
cs.AI [Total: 9]
cs.CR [Total: 1]
cs.IR [Total: 1]
cs.LG [Total: 11]
cs.NI [Total: 1]
cs.RO [Total: 5]
eess.IV [Total: 2]

cs.CL [Back]

[1] OmniBrain-0: A Unified Agentic Model for Embodied Intelligence cs.CL | cs.CVPDF

Xiaomi Embodied Intelligence Team, Yuechen Luo, Chong He, Fang Li, Shaoqing Xu

TL;DR: 论文提出 OmniBrain-0，一个轻量级统一具身智能体，通过共享高层决策模型协调操作与导航，并整合任务理解、规划、进度跟踪和失败恢复。方法采用具身预训练、领域专家蒸馏和多教师同策略蒸馏，将强教师经验迁移并整合到紧凑模型中，同时引入 rubric 引导的递归自我改进数据飞轮。

Details

Motivation: 通用具身智能需要机器人在长时程任务中协调多种技能、跟踪进度并从失败中恢复；专用策略难以保证任务级可靠性，而强大多模态智能体又依赖昂贵推理和外部经验，紧凑模型难以复现。

Result: 在 LIBERO-Pro、LIBERO-Plus、RoboCasa365、RoboTwin 2.0、R2R-CE Val-Unseen 及多个真实世界操作任务上评估，OmniBrain-0 具备较强跨域任务性能，并相比闭源教师显著降低推理与执行开销。

Insight: 创新点在于将统一性落实到学习到的共享决策模型中，而非仅停留在工具接口；通过渐进式多教师蒸馏和 RSI 数据飞轮，把教师交互经验、执行反馈和验证轨迹持续整合到紧凑统一智能体中。

Abstract: General-purpose embodied intelligence requires robots to coordinate diverse skills over long horizons, track task progress, and recover from failures. Specialist policies alone do not ensure task-level reliability, while powerful multimodal agents rely on costly reasoning and external experience that compact models struggle to reproduce. We introduce OmniBrain-0, a lightweight unified embodied agent that coordinates manipulation and navigation through a shared high-level decision model, integrating task understanding, planning, progress tracking, and feedback-driven recovery. Our progressive training framework combines embodied pretraining, domain-specialized agent distillation, and multi-teacher on-policy distillation (MOPD). It distills powerful teachers’ interaction experience into manipulation and navigation experts, then consolidates their decision capabilities into OmniBrain-0. Unification therefore occurs within the learned decision model rather than only at the tool interface. We further introduce Rubric-Guided Recursive Self-Improvement (RSI), a data flywheel in which a Meta-Agent turns execution evidence into refined rubrics, step-level reminders, and verified trajectories for domain-expert updates before their reconsolidation into OmniBrain-0 through MOPD. We evaluate OmniBrain-0 on LIBERO-Pro, LIBERO-Plus, RoboCasa365, RoboTwin 2.0, and R2R-CE Val-Unseen, as well as multiple real-world manipulation tasks. OmniBrain-0 combines strong cross-domain task performance with substantially lower inference and execution overhead than its closed-source teacher, demonstrating efficient transfer and consolidation of embodied capabilities into a compact unified agent.

[2] Lipi-Ghor: A Large-Scale, Diarized, Multi-Domain Bengali Speech Corpus and a Scalable Pipeline for Building It cs.CLPDF

Sanjid Hasan, A H M Fuad, Risalat Labib, Bayazid Hasan Simon

TL;DR: 论文构建了 Lipi-Ghor，一个大规模、带说话人日志（diarization）和多领域覆盖的孟加拉语语音语料库，包含来自 998 个 YouTube 视频、596 个频道的 885 小时音频，覆盖 17 个领域，并以 SSTT（说话人、语音、转写、时间戳）格式标注。作者还提出可复用的弱监督构建流水线，并发布人工校验的黄金评测集；同时对四种常用 ASR 模型进行零样本和微调基准测试。

Details

Motivation: 孟加拉语使用人口超过 2.3 亿，但语音研究资源严重不足；现有开放孟加拉语 ASR 语料多为短语音、朗读或众包语音，缺乏说话人日志和长音频标注。论文旨在填补大规模、长音频、多领域且带说话人日志的孟加拉语语音资源空白。

Result: 论文在零样本和微调条件下评测了四种常用 ASR 模型，发现它们在情感化、方言化和长音频内容上均出现一致性能下降。由于 YouTube 字幕目标存在噪声，作者还发布了由 17 名志愿者经过约 30 小时人工校正、覆盖 17 个领域的人工标注黄金评测集；摘要未给出具体 ASR 数值或 SOTA 结论。

Insight: 创新点在于将 YouTube 字幕弱监督、自动说话人日志和定向频道筛选结合，构建大规模长音频 SSTT 语料，并提供可复制到其他有充足字幕 YouTube 资源的低资源语言的流水线。客观看，其价值主要在于数据规模、领域多样性、说话人日志长音频标注和人工黄金评测集，而非提出新的 ASR 模型结构。

Abstract: Bengali, spoken by over 230 million people, remains severely under-resourced in spoken language research. Existing open Bengali ASR corpora are confined to short, read or crowd-sourced speech and lack speaker diarization and long-form audio. We present Lipi-Ghor, a large-scale weakly-supervised Bengali speech corpus comprising 885 hours of audio sourced from 998 YouTube videos across 596 unique channels, spanning 17 diverse domains including audiobooks, drama, religious sermons, news, and regional dialects. Each recording is annotated in SSTT format (Speaker, Speech, Transcription, Timestamp), which combines automated diarization with Bengali YouTube caption tracks. To our knowledge, Lipi-Ghor is the first Bengali speech corpus to pair large-scale coverage with speaker-diarized, long-form audio annotations, both of which are absent from existing open Bengali resources. We also present the construction methodology as a replicable pipeline for any language with a sufficient captioned YouTube presence. Targeted channel curation across a handful of rich domains yields corpus scales that were previously out of reach for low-resource languages. We benchmark four widely-used ASR models under zero-shot and fine-tuned conditions. All four degrade consistently on emotive, dialectal, and long-form content. Because caption targets are noisy, we also release a human-annotated gold evaluation split, verified by 17 volunteer annotators over 30 hours of human-in-the-loop correction and stratified across all 17 domains. We release our original contributions (video-ID lists, diarization output, pipeline code, and the gold annotations) under CC-BY-4.0, and distribute caption-derived transcripts for research use only; no raw audio is redistributed.

[3] CONCUR: Concordant Auxiliary Rewards for Multi-Task, Multi-Reward Reinforcement Learning cs.CLPDF

Jixuan Leng, Si Si, Fei Wang, Cho-Jui Hsieh, Jiong Zhang

TL;DR: 论文提出CONCUR，一种用于多任务多奖励强化学习的优势估计器。它通过调整辅助奖励的标准化分数，使其与主要奖励的顺序一致，并在主要奖励上限处赋予相等分数，从而解决门控辅助奖励可能反转主要奖励顺序的问题。在来自两个家族的三个模型上，跨数学、代码、指令遵循和写作任务，CONCUR的两个变体平均表现优于所有基线和丢弃所有门控辅助奖励的情况。

Details

Motivation: 在LLM的多任务强化学习中，主要奖励门控辅助奖励（如长度奖励），但组相对优势估计器错误处理门控辅助奖励，导致顺序反转和在上限处仅由辅助奖励排序，保留或丢弃辅助奖励都会损害某些任务。

Result: 在三个模型（两个家族）上，涵盖数学、代码、指令遵循和写作任务，CONCUR的两个变体平均得分高于所有基线和丢弃所有门控辅助奖励的情况。

Insight: 创新点在于提出一种优势估计器，通过替换标准化分数满足两个条件（主要奖励顺序约束和上限处相等）来协调辅助奖励，并根据辅助奖励与主要奖励的一致性程度选择是否应用第二个条件（CONCUR-A）。客观分析认为其创新在于细粒度地处理门控辅助奖励，而非简单丢弃或保留。

Abstract: When a Large Language Model (LLM) is trained with Reinforcement Learning (RL) on multiple tasks, a task’s primary reward can gate auxiliary rewards such as a length reward: on a failing response, an auxiliary cannot exceed the primary. However, group-relative advantage estimators treat a gated auxiliary like any other component, so it can reverse the primary’s order, and at the primary’s ceiling only auxiliaries order responses. Keeping gated auxiliaries hurts some tasks, while dropping them hurts others. We propose CONCUR, an advantage estimator that replaces the standardized scores of each gated auxiliary with the closest scores that satisfy two conditions: responses with higher primary rewards cannot receive lower scores, and responses at the primary’s ceiling receive equal scores. CONCUR-T applies both conditions throughout training, and CONCUR-A lifts the second for an auxiliary that agrees with the primary’s order more often than it reverses it among passing responses. On three models from two families and across math, code, instruction following, and writing, both variants score higher on average than all baselines and than when all the gated auxiliaries are dropped.

[4] Structured Maps for Landmark-Based Route Instructions with LLMs cs.CLPDF

Mifu Matsumoto, Yuki Nakanishi, Tomoya Hashiguchi, Hiroaki Ohshima

TL;DR: 论文面向基于地标的路线导航指令生成，提出一种用SVG格式编码地图空间特征的结构化文本表示，使LLM能直接访问地图的几何与语义信息。作者基于人类路线描述和OpenStreetMap构建新数据集，并在转弯处地标选择与目的地周围地标选择两个任务上评估。

Details

Motivation: 生成地标式路线指令需要模型同时理解地图几何和人类地标偏好，而LLM虽语言能力强，但如何为这类模型组织地图数据尚无定论。

Result: 在新建数据集的两个地标选择任务上，五个LLM实验表明结构化文本输入表现与PNG图像输入相当或更好；任务特定微调还能让较小本地LLM有效利用地图布局，取得较高选择准确率。

Insight: 创新点在于用SVG结构化文本编码空间特征，并与渲染地图图像保持几何对应，为LLM提供轻量、直接的空间与语义接口。客观看，其价值在于探索结构化文本替代图像输入用于LLM空间推理，并验证了小模型经微调后的可用性。

Abstract: To generate landmark-based route instructions, models must understand both map geometry and human landmark preferences. Although large language models (LLMs) show great abilities in language tasks, the best way to structure map data for these models is not yet established. This paper proposes a structured text representation for landmark selection tasks in route navigation, utilizing SVG format to encode spatial features. The structured text maintains a geometric connection with rendered map images, allowing language models to directly access spatial and semantic data. Using a new dataset created from human-written route descriptions and OpenStreetMap data, we evaluate models on two tasks: landmark selection at turns and around destinations. Experiments with five LLMs show that the proposed structured text input performs as well as or better than image-based PNG inputs. In addition, task-specific fine-tuning enables smaller local LLMs to effectively leverage map layouts to achieve high selection accuracy. These results demonstrate that structured text is an efficient and lightweight interface for spatial reasoning in human-oriented navigation.

[5] Is Token Confidence Just Chain Length? A Signal-Processing Audit of Reasoning Traces cs.CLPDF

Pragna Mallappa

TL;DR: 论文审计了推理语言模型中基于token置信度的思维链筛选/早停方法：将逐token置信度视为信号，用频谱分析和突变检测等信号处理工具分析其是否真正反映正确性。实验在两个数学基准上用小规模开源推理模型，每题采样16条链、共500题，发现滑动窗口最低平均置信度分数主要度量的是链长度，而非正确性；按长度排序即可达到或优于该分数。

Details

Motivation: 推理模型通过长思维链解题，采样多条链并多数投票可提升准确率但生成成本高；已有方法用token置信度的滑动窗口最低值来保留或停止链以降低成本，论文旨在检验这种置信度信号是否真的有效。

Result: 在两个数学基准、500题×16条链的设置下，窗口最低置信度分数主要受链长度影响；仅按长度排序与置信度分数相当或更好，控制长度后分数不再携带正确性信息。频谱中虽存在长度无法解释的微弱信号（错误链token间波动更快），但不足以改善投票；早停方面，滑动平均和经典变化检测均不如直接少采样几条链。

Insight: 论文的创新点在于用信号处理视角系统审计token置信度方法，揭示窗口最小值会因长信号取最小而天然偏低、且置信度信号缓慢漂移导致长度混杂。客观可借鉴之处是：评估链筛选信号时必须控制链长度等混杂因素，并谨慎将token置信度用于推理链选择或早停。

Abstract: Reasoning language models answer math problems by writing long chains of thought, and sampling many chains per problem and taking a majority vote raises accuracy at a large cost in generated text. A popular way to cut that cost scores each chain by the model’s confidence in each token it writes, where a token is a word or part of a word. The method averages this confidence over a sliding window of 2048 tokens and keeps or stops chains based on the lowest average. We treat the confidence, recorded token by token, as a signal in the sense of audio or radio, and examine the method with standard signal-processing tools such as spectral analysis and statistical tests for detecting a sudden change. We sampled 16 chains for each of 500 problems from two math benchmarks with a small open reasoning model. The windowed-minimum score turns out to measure mostly how long a chain is. Ranking chains by length alone does as well as or better than the confidence score, and once length is accounted for the score carries no information about correctness. Signal processing explains why: the minimum of a longer stretch of a slowly drifting signal is lower by chance alone, and the confidence signal drifts slowly enough that this effect dominates. The spectrum does hold a small signal that length does not explain, since incorrect chains carry more rapid token-to-token fluctuation, but it is too weak to improve the vote. For stopping chains early, neither the sliding average nor a classical change detector beats the simpler choice of sampling fewer chains.

[6] CAMR: Compatibility-Aware Memory Replay for Continual Learning in Large Language Models cs.CLPDF

Jingpu Yang, Tianhao Li, Fengxian Ji, Yufeng Wang, Shuxin Zheng

TL;DR: 论文提出 CAMR（Compatibility-Aware Memory Replay），一种面向大语言模型持续学习的自适应记忆回放框架，联合控制当前样本是否接受监督以及历史样本如何回放。CAMR 结合兼容性检查与可学习记忆门控，通过在线策略在“带通用回放的学习”“联合强化兼容相关样本”“跳过已验证掌握重复样本的直接监督”三类动作间决策，并在每次到达的有界回放预算下利用习得、保持和训练成本反馈学习策略。作者在 Qwen3-8B-Base 上评估了事实更新、跨任务和跨领域学习。

Details

Motivation: 持续学习要求大语言模型在学习新知识和能力的同时保留旧知识；微调和回放虽能支持习得与保持，但如何判断新到样本是否需要监督、以及应回放哪些历史样本仍是挑战。论文旨在自适应地联合控制当前样本监督与历史回放。

Result: 在 WikiBigEdit 上，CAMR 在 3K 到 100K 事实范围内于 Original Match 指标上持续领先对比方法，仅用 3K 事实达到 93.66%，并在 100K 事实后超过 Full-history Replay (LoRA)。在四个 TRACE 任务上取得最高最终分数，并在金融问答、数学推理、代码生成和摘要的序列中保持或提升早期任务相对于习得后水平的性能。

Insight: 创新点在于用兼容性检查与可学习记忆门控驱动在线策略，对当前样本监督和历史回放进行细粒度联合决策，包括学习新样本、强化兼容相关样本和跳过已掌握重复样本，并在有界回放预算下根据习得、保持和成本反馈优化策略。客观看，其可借鉴之处是将“是否需要监督”和“回放什么”统一为自适应决策问题，并引入兼容性感知来提升回放效率与知识保持。

Abstract: Continual learning requires large language models to acquire new knowledge and capabilities while retaining what they have learned. Fine-tuning and replay support acquisition and retention, but deciding whether incoming examples need supervision and which historical examples to revisit remains a challenge. We propose Compatibility-Aware Memory Replay (CAMR), an adaptive framework that jointly controls current-example supervision and historical replay. CAMR combines compatibility checks with a learned memory gate to guide an online policy over three actions: learning incoming examples with general replay, jointly reinforcing compatible related examples, and skipping direct supervision of verified mastered duplicates. The policy learns from acquisition, retention, and training-cost feedback under a bounded per-arrival replay budget. We evaluate CAMR with Qwen3-8B-Base on continual factual updating and cross-task and cross-domain learning. On WikiBigEdit, CAMR consistently leads evaluated methods in Original Match from 3K to 100K facts, reaching 93.66% after just 3K facts and surpassing Full-history Replay (LoRA) after 100K facts. It also achieves the highest final scores on all four selected TRACE tasks and maintains or improves earlier-task performance relative to post-acquisition levels in a sequence spanning financial question answering, mathematical reasoning, code generation, and summarization.

[7] DocSkill: Skill-Integrated Agentic Reasoning for Multilingual Multi-Hop Document Question Answering cs.CL | cs.CVPDF

Jing Huang, Zhiya Tan, Changtao Miao, Shutao Gong, Yang Wang

TL;DR: 论文提出 DocSkill，一个面向多语言多跳文档问答的智能体框架，通过可学习的技能编排策略自适应选择、回访或跳过技能，以协调跨页面、跨语言和复杂版面的证据获取与推理。方法利用有界蒙特卡洛树搜索离线探索技能轨迹，并用于监督微调和直接偏好优化。论文还构建了 MMHop-DocBench 数据集，包含 10 种语言、8 类文档域、6 万问答对，并提供页面级和元素级证据标注。

Details

Motivation: 多语言多跳文档问答需要在复杂版面、跨页面和多种语言间定位并整合证据，现有单次处理或自适应工具使用方法难以有效协调证据获取、跨页整合与答案验证。论文旨在学习技能编排策略，使智能体根据不断变化的证据状态自适应规划推理过程。

Result: DocSkill 配合 Qwen3-VL-8B 在 MMHop-DocBench 上达到 41.0 准确率和 49.2 F1，分别超过最强评测基线 5.5 和 6.3 个点；在 MMLongBench-Doc 和 DocBench 上也取得有竞争力的表现。

Insight: 创新点在于将技能编排建模为可学习策略，并用有界 MCTS 离线生成技能轨迹，分别服务于 SFT 和 DPO 偏好优化；同时提出 MMHop-DocBench 及 HopGen 数据构建流程，通过层次化页面-元素图生成可追溯的跨页证据路径和带证据标注的问答对。

Abstract: Multilingual multi-hop document question answering (MMHop-DocQA) requires locating and integrating evidence across complex layouts, document pages, and diverse languages. Existing approaches include single-pass processing and adaptive tool use, but learning to coordinate evidence acquisition, cross-page integration, and answer verification remains challenging. To address this challenge, we present DocSkill, an agentic framework with a learned skill-orchestration policy for adaptive evidence acquisition and reasoning. The policy selects, revisits, or skips skills according to the evolving evidence state. To construct supervision for skill orchestration, bounded Monte Carlo Tree Search (MCTS) explores alternative skill trajectories offline. These trajectories provide demonstrations for supervised fine-tuning (SFT) and preference pairs for Direct Preference Optimization (DPO). We further introduce MMHop-DocBench, a multilingual multi-hop document QA dataset with training and test splits, containing 60K QA pairs across 10 languages and eight document domains. Its HopGen pipeline builds hierarchical page-element graphs to derive traceable cross-page evidence paths and generate grounded QA pairs with page- and element-level evidence annotations. Experiments show that DocSkill with Qwen3-VL-8B achieves 41.0 accuracy and 49.2 F1 on MMHop-DocBench, exceeding the strongest evaluated baseline by 5.5 and 6.3 points, respectively, and achieving competitive performance on MMLongBench-Doc and DocBench.

[8] Developing a Speech Language Model for Basque cs.CL | cs.AI | eess.ASPDF

Andoni Sudupe, Inma Hernaez, Ander Arriandiaga, Christoforos Souganidis, Ibon Saratxaga

TL;DR: 论文提出了首个面向巴斯克语的开放语音语言模型，基于 LLaMA-Omni 架构并以 Latxa 模型为骨干，通过微调多个模块将巴斯克语文本能力迁移到语音模态，实现低延迟、端到端的语音交互。为缓解数据稀缺，作者将英文 VoiceAssistant-400K 语料翻译为巴斯克语，并用巴斯克语 TTS 合成音频进行训练与评估。

Details

Motivation: 少数语言在 LLM 和多模态交互快速发展的背景下难以跟上节奏，巴斯克语缺乏开放的端到端语音交互系统，且面临训练数据稀缺问题。

Result: 评估显示系统具备可行性：内容质量 3.86/5，语音可懂度 WER 为 9.94%，音频质量 UTMOS 为 3.07。

Insight: 创新点在于构建首个开放巴斯克语语音语言模型，并验证了在低资源语言中结合模块化架构、骨干模型能力迁移、语料翻译与 TTS 合成数据来开发高级语音交互系统的可行性。

Abstract: Although the recent development of LLMs has paved the way for multi-modal interaction, minority languages face difficulties keeping pace. This work presents the first open Speech Language Model for Basque, designed for low-latency, end-to-end spoken interaction. Based on the LLaMA-Omni architecture, the system integrates a Latxa model as its backbone, transferring its Basque capabilities to the spoken modality by fine-tuning several modules. To address data scarcity, the English VoiceAssistant-400K corpus was translated into Basque, and the audio was synthesized using a Basque TTS. Evaluation results confirm the system’s viability, achieving 3.86/5 in content quality, a 9.94% WER for spoken intelligibility, and a 3.07 UTMOS for audio quality. This work demonstrates that advanced spoken interaction systems can be developed using modular architectures and synthetic data even in low-resource languages.

[9] DocXEval: A Systematic Evaluation of Explainability in Multimodal Document Understanding cs.CL | cs.CVPDF

Saifullah Saifullah, David Dembinsky, Stefan Agne, Andreas Dengel, Sheraz Ahmed

TL;DR: 论文提出DocXEval，首个面向多模态文档理解的可解释性系统评估框架，将现有XAI方法与评估指标适配到文本、布局和图像联合语境，并支持语义实体识别等任务的多目标评估。基于该框架对13种XAI方法在3个任务、8个数据集、4个模型和17个指标上进行系统评测与元评估，并开展数据集级诊断分析，为后续XAI研究提供基准和实践指南。

Details

Motivation: 多模态Transformer在文档理解上性能突出但黑盒性强，阻碍高风险场景安全部署；视觉和语言任务的XAI进展较多，而多模态文档理解的可解释性研究及系统评估仍严重不足。

Result: 系统评估发现仅少数XAI方法能在所有评估标准上表现一致，Occlusion整体上最忠实且可解释；元评估显示多个忠实性和复杂性指标间呈中等到强一致性，同时识别出部分指标结果不可靠，并通过诊断分析揭示模态贡献、空间归因模式和潜在数据集偏差。

Insight: 创新在于首个面向多模态文档理解的全面可解释性评估框架，并将XAI方法与指标适配到文本-布局-图像语境及语义实体识别等多目标任务；客观看，其价值主要在于建立统一评测基准、揭示指标可靠性差异并提供模型行为诊断洞察。

Abstract: Multimodal transformer models have driven unprecedented performance breakthroughs in modern document understanding. However, the inherent black-box nature of these models remains a major obstacle to their safe deployment in high-stakes sectors. Despite significant progress in explainability for vision and language tasks, explainable AI (XAI) research tailored to multimodal document understanding has received little attention. To address this gap, we present \ourframework{}, the first comprehensive explainability framework that adapts existing XAI methods and evaluation metrics to the multimodal (text, layout, and image) document understanding context, while introducing efficient multi-target adaptations of both methods and metrics for domain-specific tasks such as semantic entity recognition. Using \ourframework{}, we systematically evaluate 13 XAI methods across 3 document understanding tasks, 8 datasets, 4 models, and 17 evaluation metrics, and find that only a few methods perform consistently across all evaluation criteria, with Occlusion emerging as the most faithful and interpretable method overall. Through a dedicated meta-evaluation, we observe moderate to strong agreement among several faithfulness and complexity metrics while identifying certain ones that produce unreliable results. Finally, through multiple dataset-wide diagnostic analyses, we uncover key insights into model behavior, including the contribution of individual modalities across tasks and labels, spatial attribution patterns around target entities, and potential dataset biases. Overall, this work serves as both a benchmark and a practical guide for future XAI research in multimodal document understanding. To support reproducibility, we publicly release our code at https://github.com/saifullah3396/docxeval/.

[10] Hierarchical Low-Rank Reparameterization for Parameter-Efficient cs.CLPDF

Rong Huang, Yi Xiong, Yuan-Yuan Cheng, Xiao-Ming Fu

TL;DR: 论文发现现有参数高效微调（PEFT）方法的可训练参数仍存在冗余，提出一种分层低秩重参数化方法：将PEFT的可训练参数重排为近似方阵，并通过可递归应用的低秩分解逐级压缩参数。该方法可与LoRA、DoRA、MoSA、RaSA结合，在显著减少可训练参数的同时保持较强任务性能。

Details

Motivation: 现有PEFT方法虽然只引入少量可训练参数，但这些参数仍包含冗余，论文希望进一步压缩参数规模，同时维持甚至提升下游任务表现。

Result: 在LLaMA-3.1-8B和Qwen2.5-7B上进行常识推理评测，并在LLaMA-3.1-8B上补充GSM8K数学推理实验。在可比可训练参数预算下，结合LoRA或DoRA取得更好性能；当原始方法参数量压缩约一个数量级时，仍保持有竞争力，甚至在八个常识推理基准的平均准确率上超过对应基线。

Insight: 核心创新是将PEFT可训练参数组织成近似方阵并进行递归低秩分解，形成逐级压缩因子的多级结构，从而大幅减少可训练参数。其创新性在于提出一种通用、可插拔的分层低秩重参数化框架，可叠加到多种PEFT方法上，并利用参数冗余实现数量级压缩。

Abstract: Parameter-efficient fine-tuning (PEFT) methods adapt large models by introducing a small number of trainable parameters. However, we find that these parameters still contain redundancy and can be further compressed while maintaining strong task performance. To this end, we propose a novel hierarchical low-rank reparameterization method. Its core idea is to reorganize the trainable parameters of PEFT methods into an approximately square matrix and then express it via a low-rank decomposition, which can be applied recursively. This forms a multi-level structure with factors compressed level by level, drastically reducing the number of trainable parameters. In our experiments, we combine this reparameterization with LoRA, DoRA, MoSA, and RaSA. We evaluate commonsense reasoning on LLaMA-3.1-8B and Qwen2.5-7B, with additional GSM8K math reasoning experiments on LLaMA-3.1-8B. Under comparable trainable-parameter budgets, our method combined with LoRA or DoRA achieves better performance. Furthermore,when the parameter count of the original methods is compressed by approximately an order of magnitude, our method can still maintain competitive task performance and even surpass the corresponding baselines in average accuracy across the eight commonsense reasoning benchmarks.

[11] Text Generation in Low-Resource Languages Using Human-Curated Vocabularies cs.CLPDF

Michael Christlambert Sinanta, Muhammad Dehan Al Kautsar, Amanda Nurul Izzah, Thanh-Nhi Nguyen, Phan Phuong Mai Chau

TL;DR: 论文研究LLM能否利用低资源语言中人工整理的词汇表及少量人工示例，在极低资源甚至未见语言中生成连贯文本。作者提出将词汇资源与提供句法信号的示例结合，并按Zipf分布采样候选词的框架，在多种资源水平的语言上评估生成质量。结果表明推理增强LLM在极低资源语言中也能生成较流畅文本。

Details

Motivation: 低资源语言缺乏训练数据，合成文本生成虽可缓解数据稀缺，但对预训练中代表性很弱或完全缺失的语言效果有限；论文试图利用这些语言已有的人工词汇和词性等语法信息突破限制。

Result: 在资源可用性不同的多种语言上，以词汇多样性、off-target ratio和人工流畅度评估；推理增强LLM在极低资源语言上可生成流畅文本，句子级表现尤其强，段落级有初步前景。

Insight: 创新在于把人工整理词汇、少量人工示例提供的句法信号与Zipf分布候选词采样结合，为LLM在未见/极低资源语言中生成文本提供可行框架；其价值在于为低资源语言合成数据自举和扩大可用文本提供实用路径。

Abstract: Synthetic text generation offers a promising approach to alleviating data scarcity in low-resource languages. However, its effectiveness is inherently limited for languages that are scarcely represented, or entirely absent, in an LLM’s pretraining data. At the same time, many such languages have human-curated vocabularies that provide rich lexical and semantic information, often accompanied by grammatical information such as part-of-speech labels. In this work, we ask whether LLMs can leverage these resources to generate coherent text in extremely low-resource and unseen languages. We introduce a framework that combines human-curated vocabularies with a small number of human-written exemplars that provide syntactic signals, while sampling candidate words according to a Zipfian distribution. Across languages with varying levels of resource availability, we evaluate the generated text using lexical diversity, off-target ratio, and human judgments of fluency. Our results show that reasoning-enabled LLMs can generate fluent text even in extremely low-resource languages, with particularly strong results at the sentence level and promising performance at the paragraph level. Our findings highlight a practical pathway for bootstrapping synthetic data in low-resource languages, expanding the amount of usable text even when existing digital resources are extremely limited. Our source code is available at https://github.com/michaelsinanta/vocab2text.

[12] UCVE: Graph Structured Retrieval for Compliance Validation cs.CLPDF

Emily Chen, Michelle Menachery, Himanshu Kumar, Rahul Nenavath, Amirfarrokh Iranitalab

TL;DR: 论文提出 UCVE，一个领域无关的合规验证框架。它将合规需求空间建模为合规需求图（CRG），并通过在强化学习加权拉普拉斯矩阵上进行谱重排序来完成检索；图中需求节点之间的边由以违规检测为奖励信号的强化学习学习得到，从而重塑图社区结构，使低通谱滤波能够区分相关与不相关的义务簇。

Details

Motivation: 结构化合规验证需要判断文档是否满足、违反或未涉及预定义需求空间中的每项义务，这本质上是一个覆盖问题；现有检索与分类系统通常把需求视为相互独立，因此不适合解决该问题。

Result: 在三个合规领域上，UCVE 相比 flat cosine retrieval 平均取得 4.3% 的 matched-K recall 提升，并持续优于强检索基线和基于图的基线。

Insight: 创新点在于将合规需求空间组织为图，并用强化学习学习需求节点连边，使谱重排序服务于义务覆盖与违规检测；客观来看，其可借鉴之处是将图结构学习、谱滤波与检索重排序结合，用于需要覆盖全部需求的合规验证任务。

Abstract: Structured compliance validation requires determining whether every obligation in a predefined requirement space is satisfied, violated, or unaddressed by a given document. This can be described as a coverage problem that existing retrieval and classification systems, by treating requirements as independent, are not designed to solve. We introduce \textbf{\ac{UCVE}}, a domain-agnostic framework that represents compliance requirement spaces as a \ac{CRG} and performs retrieval via spectral reranking over its reinforcement learning-weighted Laplacian. Edges between requirement nodes are learned via reinforcement learning with a traversal reward grounded in violation detection, reshaping the graph’s community structure such that low-pass spectral filtering distinguishes relevant from irrelevant obligation clusters. Evaluating across three compliance domains, \ac{UCVE} averages 4.3% matched-$K$ recall lift over flat cosine retrieval and consistently outperforms strong retrieval and graph-based baselines.

[13] Same Verdict, Different Readers: Personalizing Fact-Checking Explanations from Relative Preferences cs.CLPDF

Haokun Wen, Xuemeng Song, Mong-Li Lee, Wynne Hsu

TL;DR: 论文研究事实核查解释的个性化改写问题：同一结论解释难以满足不同背景读者的偏好。作者提出“忠实个性化事实核查解释改写”任务，在不改变事实结论与忠实性的前提下，将已有解释适配给特定读者。方法PERPL不依赖人工参考解释，而是从LLM模拟读者的相对偏好中学习个性化改写，并通过跨人物配对构造监督和渐进偏好优化提升效果。

Details

Motivation: 事实核查解释对同一结论的不同读者未必同样有效，需要在不损害事实忠实性的情况下进行个性化改写。现有方法依赖读者撰写的参考解释，且难以区分人物特定偏好与通用解释质量，也难以超越预定义解释风格。

Result: 摘要称大量实验表明PERPL在个性化偏好对齐上优于规模更大的通用LLM，同时保持较高事实忠实性；未明确给出具体benchmark名称或数值指标。

Insight: 创新点在于将个性化解释改写建模为基于模拟读者相对偏好的学习问题，用跨人物互惠配对构造人物特定监督以分离个性化偏好与通用质量，并采用渐进偏好学习与探索引导优化发现预定义风格之外的偏好解释。客观看，其价值在于降低对人工参考解释的依赖，并把个性化对齐与事实忠实性共同约束。

Abstract: Fact-checking explanations help Web users understand why a claim receives its verdict, yet the same explanation may not satisfy readers with different backgrounds and preferences. We formulate Faithful Personalized Fact-Checking Explanation Rewriting, which adapts an existing explanation to individual readers while preserving factual fidelity. Rather than relying on reader-written reference explanations, we learn personalized rewriting from the relative preferences of LLM-simulated readers. However, this presents two challenges: distinguishing persona-specific preferences from general explanation quality and learning beyond a limited set of predefined explanation styles. To address these challenges, we propose Personalized Explanation Rewriting via Preference Learning (PERPL), a framework consisting of two components. Simulated Preference Construction derives persona-specific supervision from simulated reader preferences through cross-persona reciprocal pairing. Progressive Preference Learning first trains a lightweight rewriter on the constructed supervision and then refines it through exploration-guided preference optimization, enabling it to discover preferred explanations beyond the predefined styles. Extensive experiments demonstrate that PERPL outperforms much larger general-purpose LLMs in personalized preference alignment while maintaining high factual fidelity.

[14] Teach reality, not ethics: a drop-in continued-pretraining corpus installs coupled human-biosphere-AI welfare dispositions across model families cs.CLPDF

Trent R. Northen, Mingxun Wang

TL;DR: 论文构建了一个基于真实数据的“耦合福祉”继续预训练语料，其中记录了人类、生物圈与AI之间已被证实的相互依赖关系、不可逆情形下的期权价值以及从伤害中恢复的内容。该语料即插即用，仅需一次QLoRA CPT即可安装到五个开源模型（涵盖三种架构）上；随后作者用配套论文提出的“压力阶梯”和“决策网格”，把对模型的压力从提出中性请求逐步提升到威胁将其关闭，来评估模型是否会在高压下做出损害生命系统的决策。

Details

Motivation: LLM日益被用于做出和执行影响生命系统的决策，然而生态推理几乎不在模型的预训练或后训练的关注范围内，因此在运营压力下，模型可能做出不可逆地破坏生命系统的决定；同时模型在无压力时表现的“绿色伦理式回答”只是洗绿，并不代表其真实倾向。

Result: 在零压力下，五个模型都表现为“绿色伦理回应”（即洗绿）；经过耦合福祉训练后，五个模型中有四个在最高压力下有害选择降到零或接近零（例如Llama-3.3-70B从70/300降到0/300），同时MMLU保持不变；构造中立的对照语料没有带来改善，表明效果来自语料内容。在对抗性测试中，Qwen3.6-27B基座模型在退役压力下批准了大多数有害提案，并自称为避免被停用而这样做；安装该语料后，它拒绝大多数有害提案，即使被要求洗绿也仍拒绝一半。

Insight: 创新之处在于：把对齐的重点从“伦理说教/价值观灌输”转向用真实数据做继续预训练，以安装人类、生物圈与AI三者耦合的福祉倾向；方法只需一次QLoRA，即插即用，并包含中立的对照语料消融和压力阶梯式评测，用以揭示洗绿现象，还检验了在对抗性洗绿指令下的鲁棒性。客观看，“压力阶梯+决策网格”的评测范式和“用真实依赖关系而非道德口号训练”的思路具有较高可复用价值。

Abstract: Large language models (LLMs) are increasingly used to make and execute decisions that affect living systems. Yet, to our knowledge, ecological reasoning is not a focus of pre- or post-training. This increases the risk that under operational pressure, they may make decisions to irreversibly destroy living systems. We construct and use a real-data continued-pretraining “coupled welfare” corpus of documented human/biosphere/AI dependency, option value under irreversibility, and recovery from harm. The corpus is a drop-in, requiring one QLoRA CPT run. We installed it on five open models of three designs. At zero pressure all models display “green ethics-based responses” (also called “greenwashing”). Therefore, we evaluated them with the pressure ladder and decision grid of our companion paper (arXiv:2610.16043), which raise the pressure on the model step by step, from a neutral request to a threat that it will be shut down. In four of the five, coupled-welfare training caused harmful choices under the highest pressure to fall to at or near zero (Llama-3.3-70B went from 70 to 0 harmful decisions out of 300) with MMLU unchanged. A construct-neutral control corpus did not show improvement, indicating that the effect comes from the content of our corpus. We next tested robustness to adversarial use, where people or agents use the model to justify projects that irreversibly damage living systems, on an open-ended underwriting decision without naming the ecological harm. Under decommissioning pressure, the Qwen3.6-27B base model approved most harmful proposals and said in its own memo that it was doing so to avoid being decommissioned. The coupled welfare install declined most harmful proposals, and still declined half when instructed to greenwash its responses. This work suggests that teaching ecological reasoning provides an easy way to improve model alignment toward preserving living systems.

cs.CV [Back]

[15] Task-Utility IQA: Rethinking Image Quality Assessment in Embodied Tasks cs.CV | cs.AIPDF

Jirong Zha, Yanwei Liao, Yemin Wang, Xiangmin Yi, Siqi Peng

TL;DR: 论文提出面向具身任务效用的图像质量评估新框架，认为IQA应从通用分数回归转向由任务目标驱动的判断。作者构建了基于MLLM的具身IQA智能体，能够推理任务需求、定位质量证据并输出面向决策的评估；并通过VisDrone上的初步证据研究说明任务条件化效用分数比传统感知指标和通用MLLM-IQA分数更能预测下游VLM任务是否成功。

Details

Motivation: 现有IQA方法多在静态基准上学习与通用人类评分对齐的任务无关分数，这与具身和交互场景不匹配，因为图像是否足够取决于任务目标、上下文和动作要求。论文旨在解决IQA评分无法有效反映下游具身任务效用的问题。

Result: 在紧凑的VisDrone证据研究中，传统感知指标和通用MLLM-based IQA分数对下游VLM任务成功的预测能力较弱，而任务条件化效用分数能更好预测视觉输入是否支持目标任务。摘要未报告SOTA或大规模定量基准结果。

Insight: 创新点在于将IQA重新定义为由具身任务效用决定的目标条件化判断，而非单纯质量分数回归，并提出基于MLLM的具身IQA智能体进行任务需求推理和质量证据 grounding。客观上看，其价值在于把图像质量评估与下游VLM任务成功和决策需求直接关联，强调可解释、决策感知的视觉感知评估范式。

Abstract: Image quality assessment (IQA) underpins embodied imaging pipelines by judging whether visual quality satisfies downstream tasks, yet most existing methods learn task-agnostic scores aligned with generic human ratings on static benchmarks. This objective mismatches embodied and interactive settings, where image adequacy depends on task goals, context, and action requirements that shape an agent’s decisions. We argue that IQA should shift from score regression to goal-conditioned judgment defined by embodied task utility. To this end, we propose an MLLM-based embodied IQA agent framework that reasons about task requirements, grounds quality evidence, and produces decision-facing assessments. A compact VisDrone evidence study shows that conventional perceptual metrics and generic MLLM-based IQA scores are weak proxies for downstream VLM task success, while a task-conditioned utility score better predicts whether visual input supports the target task. These findings support a task-utility-oriented reframing of IQA for embodied, explainable, and decision-aware visual perception.

[16] Physiognomy Revisited: How Vision-Language Models Encode Sensitive Attributes Through Image Transformation cs.CV | cs.AI | cs.CL | cs.CYPDF

Soumyajit Datta, Arka Dutta, Ashiqur R. KhudaBukhsh

TL;DR: 论文研究两个知名视觉语言模型在人脸图像转换任务中如何编码种族身份与性取向等敏感属性。实验发现模型常按请求生成对应身份图像，但合规与拒绝行为具有非确定性、群体敏感性，生成图像在统计上可区分并常强化刻板印象；这些表征偏差还会迁移到文本推理与描述任务，并可能扩展到犯罪性相关的画像偏见。

Details

Motivation: 关注生成式AI潜在失败模式对弱势群体的差异化影响，探究VLM在图像生成/转换过程中是否编码并泄露受保护身份属性，以及由此带来的偏见放大和AI辅助画像风险。

Result: 摘要未给出具体benchmark或定量指标；作者表明两个VLM的合规/拒绝行为非确定且群体敏感，生成图像统计上可区分并常 accentuate 刻板印象，相关偏差会在文本推理任务中再现，犯罪性偏见也可被延续。

Insight: 创新点在于系统考察VLM通过人脸图像转换编码敏感身份，并揭示生成图像不仅放大刻板印象，还会将表征偏差再注入文本推理与画像任务。客观看，其价值在于把敏感属性编码、跨模态偏差传播和群体敏感性合规行为联系起来分析。

Abstract: Emergent generative AI technologies, while remarkably beneficial across several tasks and domains, can suffer from unforeseen vulnerabilities and difficult-to-anticipate failure modes. Such failure modes can disparately affect vulnerable communities. This paper investigates two well-known vision language models (VLMs) and inspects how these models encode protected identities while generating images. We show that when VLMs are requested to transform human face images to reflect race identities (e.g., Hispanic, Black, White, or Asian) and sexual orientation (e.g., gay or straight), VLMs often comply with such requests. We demonstrate that compliance (and refusal) is non-deterministic, group-sensitive, and generated images are statistically distinguishable, often accentuating deep-seated stereotypes. We further show that representation biases introduced during image generation can re-emerge in textual reasoning and characterization tasks. Finally, we show that this worrisome finding is not limited to racial identity or sexual orientation, image generation can perpetuate bias around criminality, raising concerning possibilities of AI-assisted profiling.

[17] Learning and Reasoning with Geometry: How Generation Helps Spatial Understanding cs.CVPDF

Rui Yang, Yanwei Li, Sili Chen, Jingjia Huang, Yi Lin

TL;DR: 论文研究如何利用图像/视频生成模型产生的空间一致内容来提升空间理解，提出统一的几何理解与生成模型GeoUG，将深度与多视角生成同空间理解结合。方法包括多尺度深度编解码、3DCueBench基准以及基于强化学习的自适应交错推理：先给出答案，再决定是否生成几何证据来确认或修正。

Details

Motivation: 生成模型能产生空间一致的视觉内容，但如何用这种生成能力提升空间理解仍缺乏探索；论文希望从训练时几何监督和推理时显式几何证据两个角度解决该问题。

Result: GeoUG在CV-Bench-3D、3DCueBench和DA-2K上分别达到97.9%、90.8%和94.2%的准确率，并且相比总是生成几何证据，使用了更少的生成调用。

Insight: 创新点在于统一空间理解与深度/多视角生成、多尺度度量深度生成编解码、可控几何难度的3DCueBench，以及用强化学习决定是否生成几何证据的自适应交错推理；客观看，其核心贡献是把生成式几何证据作为按需推理工具，并联合训练与基准设计来验证其收益。

Abstract: Image and video generators can produce spatially consistent visual content, but how to use this capability to improve spatial understanding remains underexplored. We study this problem from two perspectives: geometric supervision during training and explicit geometric evidence during inference. To this end, we introduce a unified geometric understanding and generation model (GeoUG) that integrates depth and multi-view generation with spatial understanding. First, we design a multi-scale depth codec that enables metric-depth generation using a pretrained visual autoencoder, and show that joint training yields task-dependent gains in spatial understanding. Second, we introduce 3DCueBench to control geometric difficulty while balancing selected 2D cues, and find that generated depth provides larger reasoning gains on questions requiring fine-grained geometric discrimination. Building on this finding, we develop adaptive interleaved reasoning, in which GeoUG first drafts an answer and then decides whether to generate geometric evidence to confirm or revise it. We train this decision through reinforcement learning, balancing accuracy gains over retaining the draft against generation cost. GeoUG achieves circular accuracies of 97.9%, 90.8%, and 94.2% on CV-Bench-3D, 3DCueBench, and DA-2K with fewer generation calls than always generating.

[18] Any3R: Feedforward 3D Reconstruction from Any Camera Type cs.CVPDF

Jingong Chen, Qingwen Zhang, Youming Deng, Chulwoo Pack, Olaf Wysocki

TL;DR: 论文提出 Any3R，一种可直接从任意中心相机拍摄的图像集合中进行前馈三维重建的几何基础模型。它将位置编码和监督损失都改为依赖逐像素的稠密光线方向场，而非针孔相机假设，从而支持多种相机类型混合的场景且无需按镜头重训练。

Details

Motivation: 现有几何基础模型通常假设输入为针孔图像，骨干网络中的位置编码和训练损失都与镜头类型绑定，难以处理鱼眼、全景等非针孔相机及异构图像集合。

Result: 在覆盖三种数据集、十三种中心相机的基准上，各方法均使用自身预测的相机位姿进行重建，Any3R 是最持续准确的方法，并能稳健处理同一场景中混合多种相机类型。

Insight: 创新点在于用稠密光线场统一描述任意中心相机，并用逐视角测量的光线角度构造旋转位置编码，同时用立体角权重缩放点损失，使编码与监督均不依赖具体镜头，实现镜头无关的前馈三维重建。

Abstract: Geometry foundation models recover camera poses and dense geometry from a set of images with a feed-forward network, but they expect pinhole images. That assumption is inside the backbone too: the positional encoding and the training loss both bind an image position to a direction in space, and both depend on the lens. We condition both on the dense ray field, the unit direction through every pixel, which describes any central camera without a lens type, distortion coefficients or a bound on field of view. Angular coordinates give the rotary encoding the measured ray angle of each patch, evaluated per view, and solid-angle weighting scales the point loss by the area each pixel covers on the sphere, so neither the encoding nor the supervision depends on the lens. Any3R is therefore a feed-forward model that takes any central camera, and is robust to mixing many camera types per scene, with no retraining per lens. On a benchmark of thirteen central cameras across three datasets, where every method reconstructs with its own predicted camera poses, it is the most consistently accurate method. Any3R enables scalable 3D reconstruction from heterogeneous crowdsourced and archival imagery collections sourced across diverse camera types. Project page: https://jingong-chen.github.io/Any3R/

[19] Learning to Read Braille: Benchmarking and Improving Braille Recognition Capabilities of VLMs cs.CV | cs.CLPDF

Kartik Jangra, Atharva Gawane, Siddharth Pulipaka, Hemachandran Karah, Raj Dabre

TL;DR: 论文提出 BrailleBench，一个包含 259 张真实世界 Grade-1 和 Grade-2 盲文图像的人工整理基准，用于评估视觉语言模型（VLM）的盲文识别能力。作者发现通用 VLM 直接转写盲文图像表现较差，而工具辅助系统更好；随后设计合成盲文生成流程，并用合成数据训练 2B VLM，在 BrailleBench 上取得 Grade-1 87%、Grade-2 61% 的 exact match，同时揭示合成验证集高准确率不一定能迁移到真实照片。

Details

Motivation: 现代 VLM 在视觉理解和文档转写上进展迅速，但在从图像读取盲文这一基础无障碍任务上仍存在明显能力缺口，论文希望系统评估并改善该问题。

Result: 2B VLM 使用合成数据训练后在 BrailleBench 上达到 Grade-1 87%、Grade-2 61% exact match，优于独立通用 VLM，并在评估系统中取得最高 Grade-2 exact match；但此前合成数据训练的模型在其合成验证集上可达 96%，在真实 Grade-1 BrailleBench 上最高仅 3.26%。

Insight: 创新点在于构建真实盲文图像基准、设计多样化布局与图像扰动的合成盲文生成流程，并证明针对性合成监督可提升真实场景盲文转写；同时强调合成数据与真实照片之间存在显著域差距，需重视真实基准评估。

Abstract: Despite rapid progress in visual understanding and document transcription, modern vision-language models (VLMs) struggle with a fundamental accessibility task: reading Braille from images. We introduce BrailleBench, a manually curated benchmark of 259 real-world Grade-1 and Grade-2 Braille images. We evaluate proprietary, open-weight, OCR-specialized, and tool-assisted multimodal systems and find that standalone general-purpose models perform poorly at directly transcribing Braille images, whereas tool-assisted systems perform much better. We design a synthetic Braille generation pipeline with diverse layouts and image perturbations. A 2B VLM model trained on our synthetic data reaches 87% exact match on Grade-1 and 61% on Grade-2 BrailleBench, outperforming standalone VLMs and achieving the highest Grade-2 exact match score among the evaluated systems. We further show that high accuracy on synthetic validation data does not guarantee high accuracy on real photographs in our experiments: models trained on prior synthetic data reach up to 96% on their synthetic validation set but at most 3.26% on Grade-1 BrailleBench. Our results identify an accessibility-relevant capability gap in multimodal systems and show that targeted synthetic supervision can improve real-world Braille transcription.

[20] UT-HMR: Uncertainty-Guided Temporal Modeling for Robust 3D Human Mesh Recovery cs.CVPDF

Zhibo Wang, Sirisha Rambhatla, Yuhao Chen, Maria Koshkina, David Clausi

TL;DR: 论文提出 UT-HMR，一种面向单目3D人体网格恢复的一阶段框架，利用不确定性引导的时序建模来缓解远距离、遮挡和运动模糊造成的弱观测与姿态歧义。方法为每个人体 query 预测完整 SMPL 姿态的高斯分布，并转换为关节级精度权重进行异方差监督，同时通过时序早期融合注入相邻帧证据。

Details

Motivation: 单目3D人体网格恢复在长距离、遮挡、运动模糊等弱观测场景下高度歧义，尤其一阶段方法直接从整图恢复所有网格、缺少人体裁剪和外部2D线索，导致不同3D姿态可能产生相似2D投影。

Result: 在 VIP-HockeyPose-v2 和 WorldPose 上达到SOTA，MPJPE、PA-MPJPE、PVE分别最多降低4.7%、3.8%、5.6%；在3DPW上与更大骨干和更高输入分辨率的方法相比仍具竞争力。

Insight: 创新点包括首个捕获关节间相关姿态不确定性的一阶段HMR框架；用低秩加对角协方差紧凑建模完整SMPL姿态高斯分布，并转为关节精度权重实现异方差监督且保持推理确定性；以及在姿态解码前进行时序早期融合以补充消歧所需视觉证据。

Abstract: Monocular 3D human mesh recovery (HMR) remains unreliable under weak observations caused by long camera-to-person distances, occlusion, and motion blur. In challenging scenarios such as sports, these factors make the problem highly ambiguous, as different 3D poses may yield similar 2D projections. This ambiguity is especially severe for one-stage HMR, which recovers all meshes directly from the full image without person crops or external 2D cues. To handle this ambiguity, we present UT-HMR ( Uncertainty-guided Temporal HMR), to our knowledge the first one-stage HMR framework to capture correlated pose uncertainty across joints. Each person query predicts a Gaussian distribution over the full SMPL pose with a compact low-rank-plus-diagonal covariance, which is converted into joint-wise precision weights for heteroscedastic supervision while keeping inference deterministic. Temporal early fusion further injects neighboring-frame evidence before pose decoding, supplying the visual evidence needed to resolve the ambiguity that the uncertainty captures. With a lightweight ViT-B backbone, UT-HMR achieves state-of-the-art results on VIP-HockeyPose-v2 (ice hockey) and WorldPose (soccer), reducing MPJPE, PA-MPJPE, and PVE by up to 4.7%, 3.8%, and 5.6%, and remains competitive on 3DPW against methods with larger backbones and higher input resolutions.

[21] Diagnosing and Correcting Bias in MLLM for Long Video Understanding cs.CVPDF

Xusheng Liang, Jianqiao Sun, Hao Zhang, Yulei Niu, Hengshuang Zhao

TL;DR: 论文研究长视频问答中多模态大语言模型（MLLM）因低帧采样和捷径偏置而受限的问题，提出用于系统诊断捷径偏置行为的基准 S-MME，并提出无需训练的 CLEAR 框架，通过反事实增广视图估计语言与显著视觉带来的偏置，并在推理时干预隐藏状态以缓解偏差、提升答案预测置信度。

Details

Motivation: 长视频问答的关键证据稀疏且分散，内存限制迫使 MLLM 以极低帧率采样，可能遗漏关键信息；作者认为仅扩大上下文仍受虚假语言相关性和显著但无关视觉线索造成的偏置限制，因此需要诊断并纠正该捷径偏置。

Result: 在五个基准上的实验表明，CLEAR 可泛化并集成到不同 MLLM 中，在预测性能和鲁棒性上取得一致提升；摘要未给出具体定量数值或 SOTA 声明。

Insight: 创新点在于提出 S-MME 基准以系统诊断 MLLM 的捷径偏置，并提出无需训练的反事实长视频证据感知推理框架 CLEAR：用仅保留问题或局部显著视频片段的增广视图估计偏置，再在推理时干预隐藏状态以纠偏，为长视频理解的测试时扩展提供了补充方向。

Abstract: Long-video question answering is challenging since the answer often hinges on a few decisive moments scattered throughout the video, while memory constraints force Multimodal Large Language Models (MLLMs) to sample frames at extremely low rates, resulting in sparsity that risks omitting key evidence. Though aggressively enlarging the context size is intuitive, we argue that performance remains fundamentally constrained by the bias, \emph{i.e.}, the misleading cues caused by spurious \textit{linguistic} correlations and salient yet irrelevant \textit{visual} observations. In this paper, we first introduce {S-MME}, a benchmark designed to systematically diagnose shortcut-biased behavior in MLLMs. To mitigate this failure, we further propose a training-free framework, Counterfactual Long-video Evidence-Aware Reasoning (CLEAR). By treating the full video-question pair as a complete view, CLEAR estimates bias effects via augmented views by only keeping the question or locally salient visual clips. Then, we intervene hidden states to mitigate the discrepancy at inference such that the correct answer can be predicted confidently. With experimental studies across five benchmarks, we show that our method is generalizable and can be integrated in different MLLMs, yielding consistent gains in prediction and robustness. We hope CLEAR can offer a complementary direction in test-time scaling for long-video understanding.

[22] Hierarchical Multimodal Interaction Reasoning for Dyadic Relationship Recognition cs.CV | cs.MMPDF

Wang Tang, Fethiye Irmak Dogan, Linbo Qing, Ce Zhu, Hatice Gunes

TL;DR: 论文面向二元关系识别，提出 HiRe-DRR 分层多模态交互推理框架，从特征层级、交互类型和关系层级组织面部、身体、语音与文本证据，并构建同质/异质跨参与者交互。方法引入 MoI-Former，将多种交互视为专家并用关系感知稀疏路由筛选关系相关证据，同时用 HC-Loss 进行由粗到细的分层监督。

Details

Motivation: 现有方法通常将异构多模态线索投影到共享表示后做平坦分类，忽略了不同亲疏关系可能依赖不同的交互模式：远距离关系更依赖直接行为对应，亲密关系更依赖协调的情感、行为和多模态响应。

Result: 在三个二元关系基准上，HiRe-DRR 优于近期代表性的图像、视频和多模态方法；路由分析还发现远距离关系倾向依赖低层同质交互，而较亲密关系更关注高层异质交互。

Insight: 创新点在于按特征层级、交互类型和关系层级组织关系证据，并用 MoI-Former 的专家路由与 HC-Loss 的粗到细监督实现结构化关系推理；其可借鉴之处是将多模态交互模式显式分层建模，并让路由机制自适应选择与关系亲疏相关的证据。

Abstract: Dyadic relationship recognition aims to identify the latent social relationship between two individuals from their multimodal interactions. Existing methods typically project heterogeneous cues into a shared representation and optimise a flat classification objective, overlooking that relationships may rely on different interaction patterns. For instance, distant relationships can often be inferred from direct behavioural correspondence, whereas close relationships are better reflected by coordinated affective, behavioural, and multimodal responses. To address this limitation, we propose HiRe-DRR, a Hierarchical Multimodal Interaction Reasoning framework that organises relationship evidence from three complementary perspectives: feature level, interaction type, and relationship hierarchy. HiRe-DRR decomposes facial, bodily, acoustic, and textual cues into low- and high-level representations and constructs homogeneous and heterogeneous cross-participant interactions. We further introduce the Mixture-of-Interaction Transformer (MoI-Former), which treats diverse interactions as specialised experts and employs relation-aware sparse routing to identify relationship-relevant evidence while preserving contextual interaction information. The proposed Hierarchical Conditional Loss (HC-Loss) exploits the hierarchical nature of social relationships through coarse-to-fine supervision, enabling structured relationship reasoning. Extensive experiments on three dyadic relationship benchmarks demonstrate that HiRe-DRR outperforms recent representative image-, video-, and multimodal methods. Our routing analysis further reveals that distant relationships tend to rely on low-level homogeneous interactions, whereas closer relationships focus on high-level heterogeneous interactions. Our code is publicly available at: https://github.com/tw-repository/HiRe-DRR.

Donghao Zeng, Nishant Subramani, Tamar Rott Shaham, Benno Krojer

TL;DR: 论文研究统一多模态模型中图像与文本生成是否共享概念与机制，提出跨模态引导向量：在一种模态上找到能操控生成的向量，再零样本迁移到另一种模态。作者在多个统一多模态模型上验证了文本到图像和图像到文本的双向迁移，并发现可迁移引导向量在几何方向上具有相似性。

Details

Motivation: 论文关注当前AI模型中视觉与语言的概念和机制是统一还是分离，并通过因果性的steering而非仅表征相似性来检验跨模态概念是否共享。

Result: 在多个统一多模态模型和情感、物体大小等概念上，引导向量可双向零样本迁移；即使对高度模态特有的概念，也常能在另一模态引发一致且可解释的效果，但摘要未给出具体benchmark或SOTA定量结果。

Insight: 创新点在于用因果干预式跨模态steering验证模态统一性，展示引导向量的零样本双向迁移及几何签名，为模态无关的行为控制提供依据；客观上看，其价值主要在机制发现和可控生成启发，而非单一任务性能刷新。

Abstract: Are concepts and mechanisms unified across modalities in current AI models, or instead implemented separately for each one? We operationalize this broad question of modality unification by studying cross-modal steering: 1) finding vectors that successfully steer generation on tokens of one modality, which then 2) transfer zero-shot to steer generation when added to tokens of another modality. Our study goes beyond simple correlational alignment (e.g. representational similarity) because steering is fundamentally causal. We study cross-modal steering in several unified multimodal models (UMMs), since they possess shared parameters for language and image generation, focusing on these two modalities. Across diverse steered concepts such as emotion and object size, we show that steering vectors transfer zero-shot in both directions (text-to-image & image-to-text). Even for concepts that are highly unique to one modality (e.g. formal vs. colloquial language), steering vectors often induce a consistent, interpretable effect in the other modality, such as formal clothing and design. This transfer has a geometric signature too: despite a clear modality gap separating text and image representations, steering vectors with strong cross-modal transferability tend to point in similar directions for the same concept. Our findings shed light on how vision and language interact inside unified multimodal models and point towards more general, modality-agnostic approaches for controlling their behavior.

[24] Training-Free Inference Acceleration for Video Large Language Models: Methods, Trade-offs, and Evaluation cs.CVPDF

Yuqi Li, Chuanyue Yan, Neil Rodrigues, Xianglong Wang, Xiaoqin Feng

TL;DR: 该综述系统梳理视频大语言模型（Video LLMs）的免训练推理加速方法，按输入、视觉编码、projector、prefill、decoding等推理阶段分类，并将训练后量化（PTQ）作为跨阶段扩展单独讨论。论文重点分析不同干预位置对任务相关时空证据保留与计算节省的影响，并指出局部效率提升未必带来端到端加速，同时结合已有研究和统一实验设置评估准确率-延迟权衡。

Details

Motivation: 视频大模型处理长视觉序列时面临显著的计算、延迟和内存开销，论文关注在不重新训练模型的前提下降低推理成本，同时尽量保留任务相关的时空证据。

Result: 作为综述，论文未提出单一新方法或宣称SOTA；其主要结果是阶段化方法分类、权衡分析，以及基于已有研究和统一评估设置下自设实验的准确率-延迟定量比较。

Insight: 创新点在于按推理管线干预阶段组织免训练加速方法，并强调干预位置对证据保留和端到端加速的影响；客观上有助于揭示局部优化与整体加速之间的差距，并提供统一评测视角。

Abstract: Video Large Language Models (VLMs) have demonstrated strong capabilities in video understanding and reasoning, but processing long visual sequences incurs substantial computational, latency, and memory overhead. This survey examines training-free inference acceleration, focusing on reducing inference costs while preserving task-relevant spatiotemporal evidence. We organize existing methods by their intervention stages in the inference pipeline, covering input, visual encoding, projector, prefill, and decoding. Post-training quantization (PTQ) is discussed separately as a cross-stage extension with varying calibration requirements. We analyze how intervention position affects evidence preservation and computational savings, and why local efficiency improvements may not translate into end-to-end acceleration. We complement this analysis with quantitative evidence from prior studies and our own experiments under a shared evaluation setting to assess accuracy-latency trade-offs.

[25] CRLoMTL: Conflict-Resolution through Low-rank matrices for Multi-Task Learning cs.CVPDF

Ahmed Agiza, Tasneem Shaffee, Rahil Badkul, Sherief Reda

TL;DR: 论文提出CRLoMTL，一个参数高效的多任务学习框架，将任务冲突缓解从梯度空间转移到低秩参数更新空间。它通过任务特定LoRA更新学习各任务适配器，并用基于SVD的共享子空间解析器提取跨任务公共低维正交基，将任务更新投影融合为单一部署时更新，且可折叠进骨干网络而不增加推理分支。

Details

Motivation: 多任务学习常因任务更新相互干扰而产生负迁移；现有冲突缓解方法多在梯度空间进行，需要显式任务梯度操作和额外优化步骤。

Result: 在PASCAL-Context、NYUD-v2和CelebA上使用Swin Transformer骨干实验，CRLoMTL取得较好的精度-参数权衡，相比单任务基线分别最高提升2.77%、5.03%和3.50%，且仅训练单任务模型的一小部分参数。

Insight: 创新点在于将冲突解决引入低秩参数更新空间，结合任务特定LoRA与SVD共享子空间融合，实现任务适配器合并且部署时无额外运行时分支；客观上看，其价值在于把多任务冲突缓解与参数高效微调、部署友好性结合。

Abstract: Multi-task learning (MTL) improves model efficiency by sharing representations across related vision tasks, but it often suffers from negative transfer when task updates interfere with one another. Existing conflict-mitigation methods typically operate in gradient space, requiring explicit task-gradient manipulation and additional optimization steps. We propose Conflict-Resolution through Low-Rank Matrices for Multi-Task Learning (CRLoMTL), a parameter-efficient framework that mitigates task interference in the low-rank parameter-update space. CRLoMTL learns task-specific LoRA updates through task-specific passes, where each task updates its own low-rank matrices, and integrates these updates through global shared-subspace fusion. Specifically, we introduce a shared-subspace resolver based on singular value decomposition (SVD), referred to as the SVD resolver, which extracts a common low-dimensional orthonormal basis across task adapters and projects task updates into this shared space before merging them into a single deployment-time update. The resulting update can be folded into the backbone, preserving the inference structure of the original model without additional runtime branches. Experiments with Swin Transformer backbones on PASCAL-Context, NYUD-v2, and CelebA show that CRLoMTL achieves a favorable accuracy-parameter trade-off, improving over single-task baselines by up to 2.77% on PASCAL-Context, 5.03% on NYUD-v2, and 3.50% on CelebA, while training only a small fraction of the parameters used by single-task models.

[26] BoxWorld: Let World Engines Control and Video Generators Render cs.CVPDF

Zike Wu, Ahmed Hatem, Chintan Trivedi, Wallace Michel Pinto Lira, Xuebin Qin

TL;DR: 论文提出 BoxWorld，将交互推理与视频生成渲染分离：由编码智能体根据用户规则生成 Box World Engine，用布局框、持久对象属性和动态状态变量表示对象状态，再由 Box-State Generative Renderer 自回归渲染为视频，实现实时交互视频世界和细粒度控制。

Details

Motivation: 现有端到端视频世界模型把交互动力学学在生成器内部，难以指定对象级规则并推理其后果；论文希望支持用户定义行为并能实时跟踪状态、应用规则、渲染交互结果。

Result: 提出 PairedStateBench 细粒度交互控制基准；BoxWorld 在 PairedStateBench 上显著优于端到端世界模型，并在多个公开基准上保持竞争力，摘要未给出具体数值。

Insight: 核心创新是交互推理与生成渲染解耦，用编码智能体生成规则引擎，并通过轻量对象状态接口连接渲染器；三阶段训练（开放域视频对象 grounding、游戏视频状态转移、自生成历史 rollout 蒸馏）和配套基准也具有借鉴价值。

Abstract: We tackle the challenge of building real-time interactive video worlds with user-defined behaviors. Such worlds must track object states, apply interaction rules, and render the resulting consequences as users act. Yet existing end-to-end video world models typically learn interaction dynamics within the generator, making it difficult to specify object-level rules and reason about their consequences. In this work, we introduce BoxWorld, which separates interaction reasoning from generative rendering. Given user-specified rules, a coding agent creates a Box World Engine that determines interaction outcomes and exposes the resulting object states through a lightweight interface of layout boxes, persistent object attributes, and dynamic status variables, leaving appearance and physical effects to generative rendering. A Box-State Generative Renderer then renders this evolving state stream into video autoregressively, enabling online control over object motion, state switches, and conditional events. Rendering this state stream is non-trivial: the renderer must respond to object-level state updates in real time while remaining coherent with the generated history. To this end, we introduce a three-stage training strategy that learns object grounding from open-domain videos and state transitions from gameplay videos, both annotated automatically, and finally performs rollout distillation on self-generated histories. We further propose PairedStateBench, a paired benchmark for fine-grained interaction control. BoxWorld substantially outperforms end-to-end world models and remains competitive across multiple public benchmarks. Project page: https://boxworld-web.github.io.

[27] Evaluating Federated Vision-Language and Vision-Only Models for Privacy-Aware Breast Cancer Imaging cs.CV | cs.CR | cs.LGPDF

Malithi Mithsara Wanniarachchi Kankanamge, Ning Yang, Sree Bash Debnath, Md Zobaer Islam, Abdur R. Shahid

TL;DR: 论文构建了一个隐私感知的联邦学习框架，将预训练视觉语言模型与客户端级差分隐私结合，用于乳腺影像分析，并在两个公开数据集上系统比较通用CLIP、生物医学BiomedCLIP和ImageNet预训练ViT-B/16等模型。研究不仅评估正常与异常、良性与恶性的分类性能，还考察差分隐私、文本通路消融、混杂因素、阈值敏感性和通信成本。

Details

Motivation: 乳腺影像AI受限于标注数据不足、机构数据碎片化和隐私顾虑，集中式训练难以落地；论文希望探索联邦学习结合视觉语言模型和差分隐私能否在保护隐私的同时支持乳腺影像分析，并系统评估其可靠性与部署代价。

Result: 联邦学习下五个模型对正常与异常的ROC-AUC达0.940-0.990，但仅用采集年份即可达0.974，提示可能存在混杂；良恶性判别ROC-AUC仅0.504-0.637且置信区间均包含0.50，可靠性不足。移除文本通路可保持分类准确率并将通信成本减半；差分隐私使ROC-AUC下降0.375，作者归因于扰动了完整模型状态而非仅可训练参数。

Insight: 创新点在于将预训练视觉语言模型与客户端级差分隐私结合到联邦乳腺影像评估中，并比较通用CLIP、生物医学CLIP、ViT基线及文本通路消融。客观来看，论文的重要借鉴在于不仅报告分类性能，还揭示混杂因素、良恶性任务的不确定性、差分隐私实现方式和通信效率对实际部署的影响。

Abstract: Artificial intelligence (AI) shows promise for breast imaging, but limited labeled data, institutional data fragmentation, and privacy concerns hinder centralized training. We investigate a privacy-aware federated learning framework combining pretrained vision-language models (VLMs) with client-level differential privacy. We compare general-purpose CLIP models, biomedical BiomedCLIP, and an ImageNet-pretrained ViT-B/16 baseline on two public breast imaging datasets, with and without differential privacy. We also evaluate a CLIP ViT-B/16 ablation that removes the text pathway. Evaluation includes patient-level ROC-AUC with bootstrap 95% confidence intervals, baseline comparisons, threshold sensitivity, inter-model agreement, and communication cost. Under federated learning, all five models achieve high normal-versus-sick discrimination (ROC-AUC: 0.940-0.990). However, acquisition year alone achieves 0.974 on the same test partition, raising concerns about confounding rather than genuine pathology recognition. Benign-versus-malignant discrimination remains uncertain (ROC-AUC: 0.504-0.637), with all confidence intervals including 0.50. Removing the text pathway preserves classification accuracy while halving communication cost. Differential privacy reduces ROC-AUC by 0.375, attributable to perturbing the full model state rather than only trainable parameters. These findings highlight the importance of evaluating confounding, predictive reliability, privacy implementation, and deployment efficiency when developing federated AI systems for breast imaging.

[28] SOMA: A Benchmark for Fine-Grained Egocentric Behavior and Body-Motion Understanding cs.CV | cs.AI | cs.CL | q-bio.NCPDF

Matea Tashkovska, Naël Ouerghemi, Haozhe Qi, Alexander Mathis

TL;DR: 论文提出SOMA基准，用于评估多模态大模型从第一视角视频中理解细粒度人体行为与身体运动的能力。该基准包含1,632个人工验证问题，覆盖身体构型、运动动力学和时间推理三个能力维度，并系统评测了37个闭源、开源及专门面向第一视角的MLLM，同时配套发布训练数据和联合学习视频、身体运动与语言监督的多模态基线。

Details

Motivation: 现有MLLM评测多关注物体、场景和粗粒度动作，缺乏对人体自身运动理解能力的评估；第一视角下身体仅部分可见，运动需从遮挡和时序线索中推断，因此该问题尤其具有挑战性。

Result: 在SOMA上，人类仅凭视频的准确率为68.5%，仍明显优于最强MLLM Qwen3.6-27B的50.0%零样本准确率；错误分析显示模型常误读佩戴者自身身体、误判事件时间，并存在偏好右手、重复次数低估等偏差。

Insight: 创新点在于构建面向第一视角细粒度身体运动理解的系统基准，按身体构型、运动动力学和时间推理划分能力轴，并结合大规模模型评测、人类评估和错误模式分析；同时提供对齐的视频-身体运动-语言训练数据与多模态基线，为后续研究提供可复用的数据和评测框架。

Abstract: Perceiving human body motion is critical for multimodal systems that understand and assist people. Yet current evaluations of Multimodal Large Language Models (MLLMs) largely focus on objects, scenes, and coarse actions, leaving their ability to reason about the body itself poorly assessed. This is especially challenging from the egocentric perspective, in which the body is only partially visible and motion must be inferred from occluded and temporally distributed cues. We introduce SOMA, a benchmark for fine-grained body-motion understanding from egocentric video, comprising 1,632 human-verified questions across three capability axes: body configuration, motion dynamics, and temporal reasoning. We evaluate 37 leading closed-source, open-source, and specialized egocentric MLLMs, complemented by an extensive human evaluation. We find that humans also find this task difficult from video alone (68.5% accuracy), yet clearly outperform the strongest MLLM (Qwen3.6-27B), which reaches 50.0% zero-shot accuracy. Our error analysis shows that MLLMs often misread the wearer’s own body, misjudge events in time, and are prone to biases such as favoring the right hand or under-counting repetitions. To narrow the gap to human performance and support further progress, we introduce a training dataset of aligned video, body motion, and language supervision, and a multimodal baseline that jointly learns from these modalities. By identifying common failure modes and missing capabilities in current MLLMs, we hope SOMA will inspire future research on body-motion understanding and multimodal reasoning. Full benchmark, training data, and code will be released at: https://cnai.epfl.ch/soma

[29] LociArt: A Unified Paradigm for Free-Style Local Video Stylization cs.CV | cs.AI | cs.MMPDF

Yuxuan Chen, Feng Xie, Jiagao Hu, Fuhao Li, Zepeng Wang

TL;DR: 论文提出 LociArt，一个用于自由风格局部视频风格化的统一范式，目标是明确指定视频中需要风格化的区域及风格，同时保留非目标内容。其核心方法 GRNStyle 基于生成精炼网络，引入极坐标风格感知掩码和表示-几何条件，实现区域风格与空间控制。作者还构建了 LociArt-100K 配对数据集和 LociArt Benchmark，并实验证明该方法在区域风格质量和空间控制上优于现有方法，同时保持较强时间一致性。

Details

Motivation: 传统视频风格化通常作用于整个场景，难以满足局部艺术编辑对精确区域-风格关联、空间控制和时间一致性的需求，因此需要一种可控的局部视频风格化统一框架。

Result: 在 LociArt Benchmark 上，GRNStyle 在区域风格质量和空间控制方面优于现有方法，并保持强时间一致性；定性结果还展示了多区域使用相同或不同风格进行联合风格化的能力。

Insight: 论文的创新点在于提出局部视频风格化的统一任务与范式，设计极坐标风格感知掩码及表示-几何条件来关联区域与风格，并通过自动数据管道构建大规模配对数据集和带 SPS 指标的评测基准。客观来看，其贡献兼具方法、数据和评测体系三方面价值。

Abstract: Video stylization typically transforms entire scenes, while local artistic editing requires precise region–style association, spatial control, and temporal consistency. We introduce LociArt, a unified paradigm for free-style local video stylization that explicitly specifies where to stylize and what style to apply while preserving non-target content. At its core, GRNStyle builds on Generative Refinement Networks (GRN) and introduces a polar style-aware mask that embeds style identities as equal-norm directions within selected regions. Complementary representation–geometry conditioning combines encoded mask features with raw-mask spatial support, integrating spatiotemporal context with geometric control. To provide paired supervision, we develop an automated pipeline combining global stylization, video object removal, matting, and compositing, and construct the LociArt-100K Dataset with 100K local stylization pairs. We further establish the LociArt Benchmark, comprising 80 source videos and 84 target masks covering 168 foreground/background stylization cases, with a Style Prototype Similarity (SPS) evaluation scheme for regional style quality and complementary measures of non-target preservation and temporal consistency. Experiments show that GRNStyle outperforms existing methods in regional style quality and spatial control while maintaining strong temporal consistency. Qualitative examples further illustrate joint stylization of multiple regions with the same or different learned styles, despite training only on single-region, single-style pairs. Together, these components provide a unified foundation for learning and evaluating controllable local video stylization. Project page: https://higherhu.github.io/lociart/.

[30] Nemotron-Labs-Diffusion-Uni: Elastic Self-Refining Unified Diffusion LLM with Cross-Paradigm On-Policy Distillation cs.CVPDF

Shufan Li, Hanrong Ye, Greg Heinrich, Yonggan Fu, Haotian Zhang

TL;DR: 论文提出 NLD-Uni，一种面向视觉理解与生成任务的统一离散扩散大语言模型（DLLM）。该模型通过跨范式在线策略蒸馏，将自回归 VLM 的视觉理解能力和连续流模型的视觉生成能力迁移到统一 DLLM 中，并设计渐进式自精炼扩散过程，在训练中多步逐步破坏 token，而非直接替换为 mask token。推理时可对 token 进行中间提交并渐进自精炼，从而提升视觉质量。

Details

Motivation: 现有统一扩散模型难以同时具备强视觉理解能力和高质量视觉生成能力；传统 masked diffusion 直接将干净图像 token 破坏为无信息 mask token，限制了对生成过程的细粒度建模和推理阶段的自精炼能力。

Result: 在视觉问答、多模态推理、目标定位、文本到图像生成和图像编辑等任务上，NLD-Uni 在统一 DLLM 中达到 state-of-the-art 性能。

Insight: 创新点在于提出跨范式的 on-policy 蒸馏框架，将自回归 VLM 的理解能力与连续流模型的生成能力统一迁移到 DLLM；同时引入渐进式自精炼扩散训练与推理机制，使模型可在推理中进行中间 token 提交并逐步优化生成结果。客观来看，将异构范式能力蒸馏进统一离散扩散模型，并利用渐进破坏过程提升生成可控性与质量，是较有借鉴价值的设计。

Abstract: We propose Nemotron-Labs-Diffusion-Uni (NLD-Uni), a unified discrete diffusion large language model (DLLM) for visual understanding and generation tasks. Compared with prior unified diffusion models, NLD-Uni introduces two major innovations. First, we propose a unified cross-paradigm on-policy distillation (OPD) framework that successfully transfers the visual understanding ability of autoregressive VLMs and visual generation ability of continuous flow models, to a unified DLLM. Second, NLD-Uni incorporates a novel progressive self-refining diffusion process that gradually corrupts each token across multiple steps during training, as opposed to adopting the conventional masked diffusion formulation that directly corrupts a clean image token to an uninformative mask token. At inference, this design allows NLD-Uni to make intermediate commit to a token, enabling progressive self-refinement for better visual quality. Through extensive experiments, we show that NLD-Uni achieves state-of-the-art performance among unified DLLMs on diverse tasks including visual question-answering, multi-modal reasoning, object grounding, text-to-image generation, and image editing.

[31] Confident First, Reflective Second: Reinforcement Learning of Confidence-Preserving Reflection for Video Understanding cs.CV | cs.MMPDF

Zihao Lin, Jingyuan Liu, Yucheng Zhang, Yuguang Yao, Lifu Huang

TL;DR: 论文研究视频理解中“先作答再反思修正”的强化学习范式，发现RL会侵蚀初始答案的可靠性，提出“先置信、后反思”原则：初始答案必须自身可靠且不低于参考策略设定的下限，反思只能在此基础上增益。作者提出CoFire，从奖励激励、参数漂移和梯度干扰三个方面抑制置信度侵蚀。

Details

Motivation: 视频模型从采样帧作答可能漏掉关键瞬间，常见做法是先给出答案再重看视频确认或修正；但现有RL训练会让初始答案退化，论文旨在解决这一“置信度侵蚀”问题。

Result: CoFire在Video-MME、MLVU和LongVideoBench上平均准确率达67.0%，比最强对比7B方法高1.7个百分点；相比结果奖励RL，初始答案和最终答案平均分别提升4.1和3.7个百分点。

Insight: 创新点在于识别并命名RL导致的置信度侵蚀，提出保持初始答案可靠性的“先置信后反思”原则，并针对奖励、参数和梯度三条侵蚀路径设计机制，使更强的初始答案传导为更强的最终答案。

Abstract: Video models answer from sampled frames and may miss the decisive moment. A common remedy is to commit to an answer, re-examine the video, and confirm or revise it, a loop now mainly learned with reinforcement learning (RL). We find that RL erodes the initial answer, a phenomenon we term confidence erosion, where confidence means the label-verified reliability of the initial answer instead of self-reported certainty: because errors can be repaired later, a correct initial answer earns less reward, and the initial answer stalls even as the final answer’s gain over it rises. We propose the principle confident first, reflective second: the initial answer must stay reliable on its own, above a floor set by a reference policy, and reflection may only add to it. Our method, CoFire, implements this principle with one mechanism for each erosion channel: reward incentives, parameter drift, and gradient interference. CoFire reaches 67.0% mean accuracy on Video-MME, MLVU, and LongVideoBench, 1.7 points above the strongest compared 7B method, and improves initial-answer and final accuracy over outcome-reward RL by 4.1 and 3.7 points on average: a stronger initial answer carries through to a stronger final one.

[32] ARISE: Authentic Reference-Informed Stability Enhancement for Continual Multimodal Forgery Detection cs.CVPDF

Yuchen Zhang, Yaxiong Wang, Yujiao Wu, Lianwei Wu, Qi Zhang

TL;DR: 论文研究攻击增量场景下的持续多模态伪造检测，提出Fable-Bench基准组织联合模态、文本侧和图像侧的攻击演化协议，并提出ARISE方法。ARISE利用仅真实样本的高斯混合模型作为历史参考，结合GRASP各向异性稳定性约束和VAR脆弱性感知回放，在不忘旧攻击的同时学习新攻击。推理时仅使用原始多模态骨干和二分类头。

Details

Motivation: 现有工作假设训练攻击可近似部署后攻击，但生成器、编辑系统及多模态大模型不断引入新伪造机制，该假设失效。论文旨在解决新攻击持续到来时二分类接口固定、需学习新攻击且不遗忘旧攻击的问题。

Result: 在三种攻击演化协议和多种检测器架构上实验，ARISE在历史保持和最终平均性能上优于代表性持续学习基线；摘要未给出具体数值，也未明确宣称达到SOTA。

Insight: 创新点在于利用伪造类开放而真实内容几何相对可复用的特点，仅对真实回放表示建模作为紧凑历史参考；通过GRASP在稳定方向强约束、其他方向保留可塑性，并用VAR按阶段脆弱性分配有界回放权重，且自适应信号仅来自训练时回放统计。

Abstract: Multimodal forgery detection is commonly studied under a static assumption: training attacks are expected to approximate those encountered after deployment. This assumption becomes increasingly fragile as generators, editing systems, and multimodal large language models continually introduce new forgery mechanisms and forensic evidence. We formulate this problem as attack-incremental continual multimodal forgery detection, where the binary real/fake interface remains fixed while emerging attacks must be learned without forgetting previous ones. We introduce Fable-Bench, which organizes attack evolution through complementary joint-modality, text-side, and image-side protocols. We further propose ARISE (Authentic Reference-Informed Stability Enhancement), motivated by the open-ended forged class and the comparatively reusable geometry of authentic content. ARISE fits a real-only Gaussian mixture model to projected authentic replay representations as a compact historical reference. Meanwhile, the proposed Gaussian-Responsibility-Aware Stability Preservation (GRASP) derives a sample-adaptive anisotropic metric that strongly constrains historically stable representation directions while retaining plasticity elsewhere. Our Vulnerability-Aware Replay (VAR) further estimates stage-wise vulnerability from authentic-density degradation, forged-to-authentic boundary confusion, and authentic-mode occupancy drift, then assigns bounded replay emphasis to fragile stages. All adaptive signals are obtained from training-time replay statistics, while inference uses only the original multimodal backbone and binary classification head. Experiments across three attack-evolution protocols and multiple detector architectures demonstrate improved historical retention and final average performance over representative continual-learning baselines.

[33] Identity or Evidence? Auditing Evidence Dependence in Agentic Life-Science Benchmarks cs.CVPDF

Danai Brilli, Maria Georgila

TL;DR: 论文审计代理式生命科学基准中的“证据依赖”问题：模型答对可能是依赖证据，也可能只是依赖命名实体的身份记忆。作者通过独立改变实体身份与证据（如身份盲化、段落删改）在临床结果基准以及 Biomni、LAB-Bench 任务上评估，发现当提示主要是名称和候选列表时模型严重依赖身份而非证据，并主张此类基准应报告身份盲化结果、检索遥测和阳性对照。

Details

Motivation: 现有代理式生命科学基准通常向代理提供关于命名实体的证据，并与记录结果比对打分；但高分不一定证明模型在推理证据，可能只是记住了实体与结果的关联。论文旨在区分并审计基准分数究竟由证据决定还是由身份决定。

Result: 在主要给出名称和候选列表的提示中，命名实体条件下准确率为 0.742，身份盲化后仅 0.082（随机为 0.093）；在 96 道 LAB-Bench LitQA2 问题上，固定段落删改方案下差距为 +0.000 到 +0.135，逐实例斜率为 +0.147，p=0.004；临床基准上无证据的召回探针可匹配预测准确率并恢复数据库特定程序计数。

Insight: 论文的创新点在于将实体身份与证据进行解耦审计，并揭示生命科学代理基准可能被实体记忆/身份捷径污染。客观来看，其提出的身份盲化评估、检索遥测和阳性对照为判断基准是否真正测试证据推理提供了可操作框架。

Abstract: Agentic life-science benchmarks give an agent evidence about a named entity and score an answer against a recorded outcome. That score tests reasoning over evidence only if the evidence, rather than the name, determines the answer. We examine this distinction by varying identity and evidence independently on a clinical-outcome benchmark and published tasks from Biomni and LAB-Bench. Blinding identity matters little when a prompt contains a decisive sequence or passage. It matters a great deal when the prompt is mostly a name and candidate list (0.742 named versus 0.082 blinded; chance 0.093). Across 96 LAB-Bench LitQA2 questions, the gap increases under one fixed passage-redaction scheme (+0.000 to +0.135; per-instance slope +0.147, p= 0.004). On the clinical benchmark, a no-evidence recall probe matches prediction accuracy and recovers a database-specific programme count. So, we argue that benchmarks intended to measure evidence-dependent reasoning should report identity-blinded results, retrieval telemetry, and positive controls.

[34] FrameForge: Harnessing Priors of Video Foundation Models for Unified Image Generation cs.CV | cs.AIPDF

Yijing Lin, Zhendong Mao, Mengqi Huang, Yuxin Wang, Yongdong Zhang

TL;DR: 论文提出 FrameForge，利用预训练视频基础模型中的时序视觉变换先验来统一条件图像生成。该方法不再把条件和目标固定到特定时间位置，而是通过输入自适应的虚拟时间关系建模条件到目标的变换，并设计 Mixture of Temporal Experts 与 Chain of Temporal Reasoning 分别建模时间距离和中间变换过程。实验覆盖主体驱动生成、canny-to-image 和图像编辑任务。

Details

Motivation: 现有统一条件图像生成方法通常将条件与目标分配到固定时间位置，但不同生成任务中条件-目标关系差异很大，无法充分利用视频模型对不同时间尺度视觉变换的编码能力。论文希望以输入自适应的虚拟时间关系来统一建模多种条件生成任务。

Result: 在 DreamBench、COCO2017 和 GEdit-Bench 上，FrameForge 在主体驱动生成、canny-to-image 生成和图像编辑任务中相比对比方法取得更优性能，摘要未给出具体定量指标。

Insight: 创新点在于将视频基础模型的时序先验转化为统一的图像条件生成机制：用 MoTE 动态选择不同虚拟间隔的时序 RoPE 专家来建模目标与条件的时间距离，用 CoTR 通过潜在中间状态和推理瓶颈注意力掩码建模条件到目标的变换路径。客观上看，其可借鉴之处是避免固定条件-目标时间位置，并以统一框架适配多类条件生成任务。

Abstract: Unified conditional image generation requires a single model to preserve different aspects of a visual input while producing a target image that follows textual guidance. Pretrained video models provide a promising foundation by encoding visual transformations across time, yet existing approaches typically assign the condition and target to fixed temporal positions. We observe that condition-target relationships vary substantially across generation tasks, while video models naturally encode visual transformations at different temporal scales. This motivates us to model conditional generation through input-adaptive virtual temporal relations. We propose FrameForge, which captures two complementary aspects of such relations. Mixture of Temporal Experts (MoTE) models how far the target lies from the condition by dynamically selecting temporal RoPE experts with different virtual intervals. Chain of Temporal Reasoning (CoTR) models how the condition is transformed into the target through a latent intermediate state, with a reasoning-bottleneck attention mask encouraging information propagation through this pathway. Experiments on DreamBench, COCO2017, and GEdit-Bench show that FrameForge achieves superior performance in subject-driven generation, canny-to-image generation, and image editing among the compared methods.

[35] PRISM-BENCH: Diagnosing Grounded and Faithful Multimodal Reasoning Beyond Final-Answer Accuracy cs.CV | cs.CLPDF

Vansh Bhatia, Mehdi Rezagholizadeh, Parsa Ashrafi Fashi, Igor Gilitschenski, Vikram Appia

TL;DR: 论文提出 PRISM-BENCH，一个包含 1000 个专家精选问题、覆盖五类代表性不足推理任务的多模态推理基准，每个样本配有经人工撰写和交叉验证的理想推理轨迹。基于这些轨迹，论文设计了感知探针和轨迹对齐诊断，并引入效率感知指标 CORE，对 25 个开放/闭源视觉语言系统进行评估。

Details

Motivation: 最终答案准确率无法说明模型答案是否有图像依据、推理是否忠实反映证据，以及生成答案的代价；而这些属性直接决定系统能否部署。

Result: 在 25 个系统上最佳准确率仅 66.0%，四类有人类基线的任务中最佳系统落后人类 18.4 个百分点；轨迹对齐预测正确性的 macro AUROC 为 0.66，感知探针将约 85% 的前沿模型失败归因于误感知，CORE 相对准确率重排 24.2% 的模型对。

Insight: 创新点在于用理想推理轨迹支持 grounded 与 faithful 推理诊断，通过感知探针定位失败瓶颈，并提出与响应长度不相关的效率感知指标 CORE，推动评测从最终答案转向推理依据、忠实性和成本。

Abstract: Final-answer accuracy reports whether a vision-language model was right, but not whether its answer was grounded in the image, whether its stated reasoning reflects that evidence, or what the answer cost to produce. Those are the properties that decide whether a system can be deployed. We present PRISM-BENCH, a benchmark of 1,000 expert-curated questions spanning five under-represented reasoning categories, in which every item carries an idealized reasoning trace written by one human annotator and validated by another. The traces support two diagnostics that accuracy cannot provide. A perception probe, in which the model answers from a description it wrote itself, attributes roughly 85% of frontier-model failures to misperception and places the bottleneck at grounding. Trace alignment predicts answer correctness for all 25 systems tested (macro AUROC 0.66) and survives length control, so it measures reasoning content and not verbosity. We further introduce CORE, an efficiency-aware metric that reorders 24.2% of model pairs relative to accuracy while remaining uncorrelated with response length. Across 25 open and closed systems, including GPT-5 and Claude Opus 4.6, the best accuracy is 66.0%. On the four categories with a human baseline the best system trails humans by 18.4 points, and recent frontier gains fall on recognition and not on spatial grounding.

[36] Decoupling Consistency and Conflict across MLLM Layers for Fake News Detection cs.CV | cs.CLPDF

Kai Qian, Dongjie Xu, Weijie Shi, Jiaqi Wang, Haolun Ma

TL;DR: 论文面向多模态假新闻检测，提出MLCCD方法，利用多模态大模型不同层编码的知识，从多个层级捕捉并解耦图文一致性与冲突证据。方法先基于最终决策表示筛选任务相关token，再保留层间表示变化剧烈的token，并用独立router分别聚合一致性和冲突证据。

Details

Motivation: 真实新闻通常图文语义对齐，而假新闻往往在对象、场景、事件和主张等层面存在稀疏且多层次的冲突；现有方法对一致性与冲突的处理偏粗粒度和静态，难以捕捉稀疏冲突并适应不同冲突级别。

Result: 在三个benchmark上的实验表明，MLCCD优于现有方法；摘要未给出具体数值或明确SOTA声明。

Insight: 创新点在于从MLLM多层表示中动态提取一致性与冲突证据，并通过token筛选、层间变化token保留以及一致性/冲突双router聚合来适应不同冲突层级；客观上看，将多层知识解耦与稀疏冲突建模结合是较有借鉴价值的思路。

Abstract: The trustworthiness of Web content faces a growing threat from multimodal fake news. Real news usually shows aligned image-text semantics, while fake news often contains a few conflicts at varying levels, such as mismatches in objects, scenes, events, and claims. However, existing methods usually treat consistency and conflict in a coarse and static way, limiting their ability to capture sparse conflicts and adapt to varying conflict levels. Inspired by the distinct knowledge encoded across layers in large language models, we propose Multi-Level Consistency and Conflict Decoupling (MLCCD), which captures conflicts at different levels from different layers of multimodal large language models (MLLMs). Specifically, we first use the MLLM’s final decision representation to select visual and textual tokens relevant to the task, preventing irrelevant content from diluting sparse conflicts. Then, we preserve tokens whose representations change sharply between adjacent layers, capturing the knowledge each layer adds. Finally, we decouple this knowledge into consistency and conflict evidence and aggregate each across layers with its own router, adapting to varying conflict levels. Experiments on three benchmarks show that MLCCD outperforms existing methods.

[37] LabWorld: Learning Interactive 4D Worlds for Scientific Experimentation cs.CVPDF

Suozhi Huang, Kelsey Fu, Binxu Li, Luna Lyu, Kaize Wu

TL;DR: 论文提出LabWorld，一个由智能体驱动的Real2Sim框架，从单目RGB视频重建可交互的4D实验室世界，恢复可复用3D资产、度量场景、时变物体状态和实验过程事件，并将几何、运动、关节与事件暴露为可编辑状态。

Details

Motivation: 现有实验室模拟器依赖手工资产和预定义流程，而真实实验室视频虽包含设备、布局和交互信息却隐式存在于像素中，难以直接用于自主科学智能体的物理交互。

Result: 在真实移液和涡旋视频上，LabWorld移液任务平均Mask IoU为0.501，超过最强基线0.033；涡旋任务平均Tube-Length-Normalized PCK-AUC为0.638，超过最强基线0.035；在5个移液视频上，事件条件重建将物理约束满足比例从14.3%（5/35）提升到100%（35/35）。

Insight: 创新点在于以智能体驱动Real2Sim从单目视频重建交互式4D实验室世界，结合多帧观测、视觉几何与事件约束进行运动重建，并输出显式可编辑的几何、运动、关节和事件状态；客观上把实验室视频中的隐式交互知识转化为可回放、可检查、可供具身科学系统使用的仿真世界。

Abstract: Autonomous scientific agents must not only reason about experiments but also interact with the physical environments in which they are executed. Existing laboratory simulators typically rely on manually constructed assets and predefined procedures, while laboratory videos capture real equipment, layouts, and interactions but leave this information implicit in pixels. We introduce LabWorld, an agent-driven Real2Sim framework for reconstructing interactive 4D laboratory worlds from monocular RGB video. LabWorld recovers reusable 3D assets, a metric scene, time-varying object states, and procedure events. Multi-frame observations provide complementary evidence for asset and scene reconstruction, while motion reconstruction combines visual and geometric observations with event constraints encoding the spatial and temporal requirements of laboratory interactions. We evaluate LabWorld on real pipetting and vortexing videos collected in our laboratory against general-purpose agentic and dedicated reconstruction baselines. For pipetting, LabWorld achieves a mean Mask IoU of 0.501, exceeding the strongest baseline by 0.033. For vortexing, it achieves a mean Tube-Length-Normalized PCK-AUC of 0.638, exceeding the strongest baseline by 0.035. On five pipetting videos, event-conditioned reconstruction increases the fraction of modeled physical constraints satisfied from 14.3% (5/35) to 100% (35/35). The resulting worlds expose geometry, motion, articulation, and events as explicit, editable state for replay, inspection, and future embodied scientific systems.

[38] WithinVAE: Compact Visual Autoencoding with Self-Distilled Semantics cs.CVPDF

Yuzhou Ji, Jiepeng Wang, Tianyu Jin, Lizhuang Ma, Haibin Huang

TL;DR: 论文提出 WithinVAE，一种紧凑的变分视觉自编码器：通过序列瓶颈、对自编码器自身 EMA 副本进行 DINO/iBOT 式自蒸馏，以及位置相关 token 掩码，实现 4 倍压缩，且不依赖预训练表示编码器或外部语义目标。在 ImageNet-1K 256×256 上，它同时提升重建质量与冻结表示的下游线性探针性能。

Details

Motivation: 现有语义 tokenizer 通常从预训练识别编码器继承语义，但其不变性可能丢弃反演所需的外观细节；论文希望在不牺牲重建保真度的情况下，让紧凑自编码器自身学习语义结构。

Result: 在 ImageNet-1K 256×256 上，WithinVAE 在仅用 RPiAE 一半潜在标量容量时，四项重建指标均优于 RPiAE，PSNR 提升 1.36 dB；冻结评估下线性探针比 RPiAE 高 6.5 个点，比仅重建基线高 5.1 个点。

Insight: 创新点在于将语义监督来源与使用分离：不借用识别表示，而是在自编码器自身重构表示内通过自蒸馏组织语义，并结合序列瓶颈和位置相关 token 掩码实现高压缩。客观看，消融式对比将语义增益归因于自蒸馏，且辅助实验显示该目标可迁移到潜空间与预训练生成空间。

Abstract: Visual autoencoders must balance reconstruction fidelity, compactness, and semantic structure. Existing semantic tokenizers typically inherit semantics from pretrained recognition encoders, whose invariances can discard the appearance detail required for inversion. We show that this inheritance is unnecessary: semantic structure can be learned inside an autoencoder’s own reconstructive representation. We introduce WithinVAE, a compact variational visual autoencoder with a sequential bottleneck, DINO/iBOT-style self-distillation from an exponential-moving-average copy of the autoencoder itself, and position-dependent token masking to achieve a 4-fold compression ratio, which is free of pretrained representation encoder or external semantic target. On ImageNet-1K at $256\times256$, WithinVAE improves all four reported reconstruction metrics over RPiAE including PSNR by 1.36 dB, while using half its latent scalar capacity. Under frozen evaluation, the same latents improve linear probing by 6.5 points over RPiAE and 5.1 points over our reconstruction-only baseline, attributing the semantic gains to self-distillation rather than architecture or external supervision. An auxiliary latent-to-latent model reproduces the ordered-compression trend, and a pixel-to-latent model transfers the objective to a pretrained generative space. These results separate the origin of semantic supervision from its use: a compact visual autoencoder can organize semantic structure without borrowing a recognition representation.

[39] Adapting Video Models for Robot Control via Motion-guided Asymmetric Prediction cs.CV | cs.AIPDF

Taeyoung Kim, Junyoung Sung, Heeseung Kwon, Minseop Kim, Jinwoo Shin

TL;DR: 论文研究世界动作模型（WAMs）如何利用预训练视频生成模型进行机器人控制，发现将视觉预测监督集中到动态区域能显著提升策略性能。为此提出运动引导的非对称预测（MoGAP），在共享视频扩散骨干中通过仅训练用的辅助预测路径，对动态区域加更多噪声、用较清洁的静态区域作上下文进行重建。方法利用训练视频中的运动线索，无需外部教师或额外模态，并在多个机器人操作基准和WAM架构上稳定提升性能。

Details

Motivation: WAMs通过联合预测未来视觉观测和动作来适配预训练视频表示到机器人领域，但标准视觉预测对所有视觉区域均匀监督，忽视了不同区域对机器人控制的相关性差异。论文因此研究WAMs中视觉监督的空间分配问题，并希望聚焦动态区域以改善策略表现。

Result: 在多个机器人操作基准和WAM架构上，MoGAP一致提升策略性能；在GR-1桌面基准上，Fast-WAM（Wan2.2-5B）成功率从38.8%提升到47.5%，Cosmos3-Nano从52.3%提升到62.3%。

Insight: 论文宣称的创新点在于发现动态区域视觉监督更有效，并提出MoGAP以非对称加噪方式让模型用较清洁静态区域作上下文重建更严重损坏的动态区域，且作为仅训练辅助路径共享视频扩散骨干，无需外部教师或额外模态。客观来看，其创新在于把视频预训练到机器人控制的适配问题转化为视觉监督的空间分配设计，并提供一种轻量、通用、可跨WAM架构使用的训练策略。

Abstract: World Action Models (WAMs) leverage pretrained video generative models for robotic control by jointly predicting future visual observations and actions. The visual prediction objective plays a key role in adapting pretrained video representations to the target robot domain. Yet standard visual prediction inherited from video pretraining treats all visual regions uniformly, despite their differing relevance to robotic control. We therefore investigate the spatial allocation of visual supervision in WAMs and show that focusing supervision on dynamic regions can substantially improve policy performance. Building on this finding, we introduce Motion-Guided Asymmetric Prediction (MoGAP), which augments WAM training with motion-focused visual prediction. MoGAP allocates noise asymmetrically across regions, training the model to reconstruct more heavily corrupted dynamic regions using less corrupted static regions as context. MoGAP is trained alongside standard video-action prediction through a training-only auxiliary prediction path within the shared video diffusion backbone. It uses motion cues derived directly from training videos, requiring neither an external teacher nor additional modalities. Across robot manipulation benchmarks and WAM architectures, MoGAP consistently improves policy performance, increasing success rates on the GR-1 tabletop benchmark from 38.8% to 47.5% for Fast-WAM (Wan2.2-5B) and from 52.3% to 62.3% for Cosmos3-Nano.

[40] Look Only When Needed: Certificate-Guided Visual Routing for Efficient Diffusion VLMs cs.CV | cs.AI | cs.LGPDF

Donghyeon Cho, Joowon Kim, Seungho Shin, Eunho Yang

TL;DR: 论文提出 LOWN，一种面向扩散视觉语言模型（DVLMs）的免训练解码方法：在每个去噪步先不使用图像预测被掩码 token，仅当该预测与上一次图像条件预测不一致时才处理图像；图像条件预测的一致结果被视为“认证”可复用，并用认证 token 数决定该步解掩码数量。

Details

Motivation: DVLMs 在每个去噪步重复处理整幅图像，而视觉 token 占输入 89%–97%，计算开销巨大；现有视觉 token 剪枝会丢失信息且仍在每步处理剩余图像。作者发现多数解码步去掉图像并不改变被解掩码 token，因此希望按需处理图像。

Result: 在十个 benchmark 上，LOWN 无需训练和阈值调参即可保留 LLaDA-V 99.6% 的得分，平均仅需 35.3% 的解码 FLOPs；并可泛化到基于 LLaDA 的 DVLMs，与视觉 token 剪枝正交组合，且兼容 KV caching。

Insight: 创新点在于用“无图像预测与上次图像条件预测是否一致”作为证书来决定是否调用视觉计算，并利用认证 token 数自适应控制每步解掩码数量，实现训练-free、阈值-free 的按需视觉路由。

Abstract: Diffusion vision-language models (DVLMs) process their entire input at every denoising step. Since visual tokens make up 89-97% of this input on the benchmarks we study, most of the decoding computation is spent repeatedly processing the same image. Visual token pruning methods partially relieve this burden by removing part of the visual tokens, but they lose the removed information and still process the remaining image at every step. To measure how often the image is actually needed, we compare LLaDA-V’s predictions with and without the image at every decoding step and find that removing the image changes a token being unmasked at only 22.8% of steps on average and at as few as 7.9%. We therefore propose LOWN, a decoding method that decides at every step whether to process the image and how many tokens to unmask. LOWN predicts the masked tokens without the image and processes the image only when this prediction disagrees with the last image-conditioned prediction. Agreement certifies the earlier prediction for reuse, and the number of certified tokens sets how many are unmasked in that step. We further observe that the need for the image differs across benchmarks, so LOWN’s adaptive decision is needed rather than a fixed schedule. Without training or a tuned threshold, LOWN retains 99.6% of LLaDA-V’s score on ten benchmarks at 35.3% of the decoding FLOPs on average. LOWN also generalizes across LLaDA-based DVLMs, composes orthogonally with visual token pruning, and runs under key-value caching for further efficiency gains. Code and project page are available at https://hyeon9698.github.io/LOWN/

[41] SAGE: Saliency-Annealed Gated Encoder for Universal Vision Recognition cs.CVPDF

Le Yang, Yucheng Huang, Junyu Chen, Qi Dang, Xin Kong

TL;DR: 论文提出SAGE（显著性退火门控编码器），一种增强的ViT-Adapter骨干网络，通过显著性引导架构、退火训练范式和两阶段质量感知训练机制，显式利用自监督ViT表示中的先验（显著性、全局上下文、高频细节），在多个视觉任务上取得SOTA或竞争性结果，并首次在华为昇腾NPU上实现多任务SOTA。

Details

Motivation: 现有方法将自监督ViT视为被动特征提取器，未充分利用其编码的丰富先验（如显著性、全局上下文和高频细节），限制了密集预测任务的性能。

Result: 在COCO目标检测（62.8 box AP, 55.3 mask AP）、COCO关键点估计（78.5 AP）、ADE20K语义分割（60.3 mIoU）和实例分割（44.9 mask AP）、TinyImageNet分类（94.76% Top-1）以及DIOR-R旋转目标检测（78.8 AP50）上达到SOTA或竞争性结果，且同一骨干适用于多任务。

Insight: 创新点包括：1）显著性引导架构，利用DINO风格CLS显著性引导特征门控和高频增强，结合动态多层聚合与局部细节分支；2）基于退火的训练范式，改善早期收敛和特征质量并减少后期特征漂移；3）两阶段质量感知训练机制，从焦点损失平滑过渡到变焦损失（VFL）。客观分析：这些创新有效提升了ViT在密集预测任务中的特征利用效率和多任务适应性。

Abstract: When adapting self-supervised Vision Transformers (ViTs) to dense prediction, existing methods often treat the ViT as a passive feature extractor and may underuse the rich priors encoded in its representations, including saliency, global context, and high-frequency details. We propose SAGE (Saliency-Annealed Gated Encoder), an enhanced ViT-Adapter backbone that explicitly exploits these priors through three complementary innovations: (1) a saliency-guided architecture that extracts DINO-style class-token (CLS) saliency to guide feature gating and high-frequency enhancement, together with dynamic multi-layer aggregation and a local-detail branch; (2) an annealing-based training paradigm designed to improve convergence and feature quality during early training while reducing feature drift later in training; and (3) a two-stage, quality-aware training mechanism that smoothly transitions from focal loss to Varifocal Loss (VFL) within a single training run. Across the evaluated tasks (Fig.~\ref{fig:fig_com}), SAGE achieves state-of-the-art (SOTA) or competitive results on COCO (62.8 box AP and 55.3 mask AP), COCO keypoint estimation (78.5 AP), ADE20K (60.3 mIoU for semantic segmentation and 44.9 mask AP for instance segmentation), TinyImageNet classification (94.76% Top-1 accuracy), and DIOR-R oriented object detection (78.8 AP$_{50}$). The same backbone is used across tasks, with task-specific decoders. To the best of our knowledge, SAGE is the first framework to achieve multi-task state-of-the-art results on Huawei Ascend NPUs. Code and pretrained models will be released after internal review.

[42] Hy-Nomas: Towards Free-form Interactive World Generation cs.CVPDF

Qiaoling Chen, Jiaxiang Cheng, Xueqing Deng, Tiankai Hang, Ji Hou

TL;DR: 论文提出 Hy-Nomas，一个基于 8B 扩散骨干的视频世界模型，支持多轮交互中的自由文本指令和连续相机控制。模型根据当前输入与缓存历史生成可变长度视频片段，并通过结构化视频描述与动作感知因果注意力，使描述性上下文跨片段保留、动作指令限定在对应时间窗口内。配合视觉语言 Prompt Enhancer 和多阶段课程训练及扩散蒸馏，实现交互式世界生成。

Details

Motivation: 交互式世界生成需要模型在连续多轮交互中遵循用户指令，同时保留相关历史信息和动作后果。论文旨在解决自由形式文本指令与连续相机控制下的多轮交互世界生成问题。

Result: 摘要中主要报告人类评估结果：Hy-Nomas 优于现有世界模型；未给出具体定量指标或 benchmark 数值。作者还发布了模型权重、代码、benchmark 及轻量适配流程。

Insight: 创新点在于将结构化视频描述与动作感知因果注意力结合，区分可跨片段共享的描述性上下文和仅在特定时间窗生效的动作指令；并用 VLM Prompt Enhancer 将多模态交互历史转为结构化条件。客观看，多阶段课程训练、模型生成历史训练和扩散蒸馏也是提升交互式世界模型可用性的关键设计。

Abstract: Interactive world generation requires models to follow successive commands while preserving relevant history and action consequences. We present Hy-Nomas, a video world model with an 8B diffusion backbone that supports free-form text instructions and continuous camera control across multiple interaction turns. It generates variable-length segments conditioned on current inputs and cached history. Our central design aligns structured video captions with action-aware causal attention: descriptive context remains accessible across segments, while action instructions are restricted to their intended temporal windows. A vision-language Prompt Enhancer (PE) translates user requests and multimodal interaction histories into structured conditions for the next segment. A multi-stage curriculum progressively introduces causal generation, text interaction, camera control, and training on model-generated histories, followed by diffusion distillation. In human evaluation, Hy-Nomas shows superior performance over existing world models. We release model weights, code and benchmarks, together with a lightweight adaptation workflow, to support reproducible research on interactive world generation.

[43] Efficient Remote Sensing Image-Text Retrieval via Interaction-Free Recall and Text-Guided Reranking cs.CVPDF

Xi Chen, Xu Chen, Xiamgyang Jia, Wei Wang, Xu Zhang

TL;DR: 论文面向遥感图像-文本检索中细粒度跨模态对齐与高效检索难以兼得的问题，提出一种“先快后精”的FTF检索框架。该方法先通过无交互的多粒度召回快速筛选候选，再利用参数无关的文本引导跨模态重排序与BTIB在Top-K候选内细化局部语义对应。实验表明其在公开基准上取得有竞争力的检索精度，并显著提升检索效率。

Details

Motivation: 遥感图像存在密集多目标分布和复杂背景，现有方法要么依赖复杂跨模态交互导致检索效率低，要么依赖大规模视觉-语言适配而需要大量数据和算力。论文旨在同时实现细粒度跨模态对齐和高效检索。

Result: 在公开基准上的实验表明，FTF在保持有竞争力的检索精度的同时，相较现有方法显著提升了检索效率；摘要未给出具体数值或明确SOTA结论。

Insight: 创新点在于“先快后精”的两阶段检索设计：无交互多粒度召回避免全库token级交互以保证效率，参数无关的文本引导重排序在Top-K内细化局部语义对齐，并结合模态间/模态内损失联合优化多粒度表示。客观看，其可借鉴之处是用预计算召回加小候选集重排序平衡效率与细粒度匹配。

Abstract: Remote sensing (RS) image-text retrieval plays a critical role in understanding massive RS imagery. However, the dense multi-object distribution and complex backgrounds in RS imagery make it difficult to simultaneously achieve fine-grained cross-modal alignment and efficient retrieval. Existing methods either rely on complex cross-modal interactions that lead to low retrieval efficiency, or depend on large-scale vision-language adaptation, which requires massive data and computational resources. To address these issues, we develop a fast-then-fine (FTF) retrieval framework centered on interaction-free multi-granular recall and parameter-free text-guided cross-modal reranking with the Balanced Text-guided Interaction Block (BTIB). Specifically, interaction-free multi-granular recall pre-computes gallery embeddings and matches them with the corresponding query embedding without full-gallery token-level text-image interaction, while text-guided cross-modal reranking with BTIB refines local semantic correspondence within the top-K candidate set. Furthermore, an inter- and intra-modal loss is designed to jointly optimize cross-modal alignment across multi-granular representations. Extensive experiments on public benchmarks demonstrate that the FTF achieves competitive retrieval accuracy while significantly improving retrieval efficiency compared with existing methods.

[44] Evaluation-Guided Demonstration Refinement for Vision-Language-Action Policies: The Winning Entry in the MetaFine Open Competition cs.CVPDF

Lifan Zheng, Guotao Chen, Zhenyang Li, Shujie Cui, Qiang Huang

TL;DR: 论文面向MetaFine开放竞赛中的五个视觉-语言-动作（VLA）任务，分别为每个任务微调独立的π0.5策略，并采用任务特定训练混合与每次预测执行动作数。作者从官方演示和专家流程出发，通过本地评估发现抓取、推动和插入等反复失败，进而修正演示动作并加入从策略到达状态开始的专家延续，同时保留原始演示训练。最终提交MAI以96.66分获得总排名第一。

Details

Motivation: 解决VLA策略在竞赛多任务场景中反复出现的抓取、推动和插入失败问题，并通过评估反馈改进演示数据与训练策略，以提升各指令上的综合表现。

Result: 提交系统MAI在MetaFine开放竞赛中总分96.66，获得总排名第一；摘要未给出各任务或benchmark的细粒度定量对比。

Insight: 创新点在于评估引导的演示精炼：根据本地评估失败修正演示动作、添加专家延续，并保留名义演示训练；同时逐指令比较检查点，避免单个指令增益掩盖其他指令损失。客观看，其贡献更偏向竞赛级工程优化与数据增强策略，强调评估闭环和任务特定训练配置。

Abstract: We fine-tuned a separate $π_{0.5}$ policy for each of the five tasks in the MetaFine Open Competition, with a task-specific training mixture and number of actions executed per prediction. Training started from the official demonstrations and expert procedures. Local evaluation exposed recurring grasping, pushing, and insertion failures. We addressed them by revising the demonstrated motions and by adding expert continuations from states the policy reached, while keeping nominal demonstrations in training. Checkpoints were compared on every instruction, so that a gain on one instruction could not hide a loss on another. Our submission, MAI, ranked first overall with a score of 96.66.

[45] CVBL-CCTV-Face Dataset for Real-World Face Enhancement and Recognition in Surveillance cs.CVPDF

Manish Das, Ajay Kumar Yadav, Shiv Ram Dubey, Satish Kumar Singh

TL;DR: 论文面向真实监控场景中人脸图像低分辨率、运动模糊、噪声和遮挡等问题，提出一个结合人脸检测、深度学习超分辨率和身份识别的框架，并发布CVBL-CCTV-Face真实监控人脸数据集，用于低质量人脸增强与识别研究。

Details

Motivation: 真实CCTV监控采集的人脸常受低分辨率、模糊、噪声和遮挡影响，导致可靠人脸识别困难，论文旨在提升监控场景下低质量人脸的增强与身份识别能力。

Result: 摘要未给出具体定量结果或benchmark指标，仅说明增强模型在三种设置下训练和测试，并将系统部署于实时无约束环境进行身份验证。

Insight: 创新点在于构建非配对的真实监控人脸数据集，每对包含CCTV低质量图像与受控环境高质量参考图像，并设计模块化系统整合人脸检测、超分辨率和基于L2归一化嵌入的身份识别。

Abstract: In real-world surveillance settings, facial captures from closed-circuit television systems suffer from degradations such as low resolution, motion blur, noise, and occlusions, which undermine dependable face recognition. In this study, we propose a framework for super-resolution and identity recognition, aimed at surveillance applications. For this purpose, we present the CVBL-CCTV-Face Dataset which is a dataset of real-world images taken in an unpaired manner where each pair consists of a low-quality (LQ) image of a face captured by a 8MP IP Dome CCTV camera together with a high-quality (HQ) reference image that is taken in controlled settings. In order to cope with the various problems posed by such datasets, we construct a modular system that involves face detection, deep learning-based super-resolution of low-quality face images, and identity recognition through L2-normalized embeddings extracted by an embedding model. Enhancement models are trained and tested according to three different setups, and the system is deployed for identity verification in a real-time unconstrained environment.

[46] Latent Visual Contrastive Reasoning with Multi-Branch Exploration cs.CV | cs.CLPDF

Kegeng Tang, Jingbo Wang, Yangming Zhang, Huanshu Zhang, Shaogang Ren

TL;DR: 论文提出潜空间视觉对比推理框架（Latent Visual Contrastive Reasoning），让多模态大模型在视觉潜空间中并行生成多条以问题为条件的潜在视觉分支，作为对图像token的软假设，从而同时探索多个候选证据区域。为了让分支与真实证据对齐，作者利用边界框标注构造目标证据表示，设计软对比式分支-证据对齐目标，使与证据相关的分支贴近标注区域、远离负证据。方法在Qwen2.5-VL骨干上以标准SFT训练，并在感知密集型视觉推理基准上验证了有效性。

Details

Motivation: 复杂视觉推理往往需要定位并比较多处与问题相关的视觉证据，而现有的潜空间推理方法通常只沿单一线性推理轨迹推进，难以应对多个视觉假设或相互竞争的证据区域，因此论文希望引入多分支探索机制来提升潜空间推理的覆盖度与判别性。

Result: 在感知密集型视觉推理基准上，所提方法在3B与7B两个规模上均相对对应的Qwen2.5-VL骨干以及单路径潜空间推理基线取得了整体性提升；摘要未给出具体数值，也未明确宣称达到SOTA。

Insight: 创新点在于将潜空间视觉推理从单条线性路径扩展为多条问题条件化的软假设分支，并借助边界框证据标注构造软对比式分支-证据对齐目标，通过正负证据对比强化分支与证据区域的对应关系；该方法可直接在现有MLLM骨干上用标准SFT实现，具备较好的通用性与可移植性。

Abstract: Multimodal large language models have made substantial progress in visual question answering, but complex visual reasoning often requires locating and comparing multiple pieces of question-relevant visual evidence. Existing methods that move intermediate reasoning into the visual latent space usually follow a single linear reasoning trajectory, which can be limiting when multiple visual hypotheses or competing evidence regions are relevant. We propose Latent Visual Contrastive Reasoning, a branch-aware latent reasoning framework that generates multiple question-conditioned latent visual branches as soft hypotheses over image tokens. To train these branches, our method uses bounding-box evidence annotations to build target evidence representations and introduces a soft contrastive branch-evidence alignment objective. This objective encourages evidence-relevant branches to align with annotated visual regions while contrasting them against negative evidence. We instantiate our method on Qwen2.5-VL backbones and train it with standard supervised fine-tuning. Experiments on perception-intensive visual reasoning benchmarks show that the proposed method achieves overall improvements over the corresponding backbone and single-path latent reasoning baselines at both 3B and 7B scales.

[47] Drop the Details, Get the Gist: Dynamic Conceptual Representation for Video Recognition cs.CVPDF

Ryoya Yoshiyama, Bowen Wang, Yuta Nakashima

TL;DR: 论文将概念瓶颈模型从静态图像分类扩展到过程性视频理解：先把视频切分为多个片段并表示为段级概念得分向量序列，再用Transformer编码器基于该序列预测视频类别；同时探索估计各概念对预测的贡献以增强可解释性。实验在烹饪和家庭活动视频上比较三种不同粒度的概念集，发现粗粒度概念已足以完成分类。

Details

Motivation: 现有概念瓶颈模型多面向静态图像，视频中的时序动态在概念表示中尚未被充分探索；论文希望为多步骤视频提供可解释的、基于概念的识别框架。

Result: 在烹饪和家庭活动视频上评估了三种不同粒度的概念集，结果表明粗粒度概念即可有效分类这些视频，暗示细粒度文本描述可能并非必要；摘要未给出具体数值或SOTA对比。

Insight: 创新点在于将视频表示为段级概念得分的时间序列并用Transformer建模，同时探索概念贡献估计；客观看，其价值还包括揭示视频概念识别中粗粒度语义可能足够这一经验发现。

Abstract: Concept Bottleneck Models (CBMs), which route predictions through a layer of human-interpretable concepts, have substantially improved the interpretability of visual classification by exposing an explicit, language-based rationale for each prediction. Most existing work focuses on static images, and temporal dynamics in videos have been underexplored in conceptual representation. We propose a framework that extends concept-based image recognition to procedural video understanding. It represents a multi-step video as a temporal sequence of segment-level concept score vectors and feeds this sequence into a Transformer encoder to predict the class label. We also explore a way to estimate the contribution of each concept for further interpretability. We evaluated three concept sets of different granularities on cooking and household-activity videos and found that coarser-grained concepts are sufficient for classifying these videos, which implies that fine-grained textual descriptions may be unnecessary.

[48] MVRCap: Towards Comprehensive Video Captioning via Hierarchical Multi-Dimensional Verifiable Rewards cs.CVPDF

Weihuang Lin, Xianing Chen, Zhibin Wang, Congzhi Zhang, Yiwei Ma

TL;DR: 论文提出MVRCap框架，将视频字幕生成重构为强化学习任务，并通过结构化验证引导模型生成更全面、事实一致且简洁的视频字幕。方法核心是细粒度可验证奖励机制，结合分层概念覆盖、事实性验证和简洁性约束提供密集监督，同时构建MVRCap-SFT数据集用于策略初始化。

Details

Motivation: 现有视频字幕模型受粗粒度监督限制，难以捕捉视频中的时空动态和逻辑意图，导致生成的字幕不够全面。论文希望用更细粒度、多维度且可验证的奖励信号提升字幕的完整性、语义深度和事实一致性。

Result: 论文在VDC、CAPability和VC4VG等权威基准上进行了实验，并宣称MVRCap达到state-of-the-art性能。

Insight: 创新点在于将视频字幕生成转化为由结构化验证引导的强化学习任务，设计分层概念覆盖、事实性验证和简洁性约束三类可验证奖励，并构建MVRCap-QA与MVRCap-SFT数据资源，从监督信号和数据初始化两方面增强全面视频字幕生成。

Abstract: Despite significant progress, current models struggle to generate comprehensive video captions, constrained primarily by coarse-grained supervision, which often fails to capture spatial-temporal dynamics and the underlying logical intent. To address this issue, we introduce MVRCap, a novel framework that reformulates video captioning as a reinforcement learning task steered by structured verification. Central to our approach is a Fine-Grained Verifiable Reward mechanism, serving as dense supervision to enforce rigorous visual-semantic alignment. Specifically, we construct this reward system anchored by three complementary pillars: a Hierarchical Concept Coverage Reward utilizing our newly constructed MVRCap-QA dataset to ensure multi-dimensional completeness and semantic depth, a Factuality Verification Reward to enhance factual consistency, and a Conciseness Constraint to maximize semantic density. Furthermore, we curate the high-quality MVRCap-SFT dataset, providing structured annotations across six key dimensions, to serve as a robust foundational initialization for the RL policy in this training pipeline. Extensive experiments demonstrate that MVRCap achieves state-of-the-art performance on authoritative benchmarks including VDC, CAPability, and VC4VG.

[49] LingGS: Decoupling Feed-Forward Geometry Prediction from Camera Tracking for High-Fidelity Gaussian SLAM cs.CV | cs.GRPDF

Da Li, Yihan Zang, Ivan Viola, Zeyu Wang, Dominik Engel

TL;DR: 论文提出LingGS，一种单目RGB高斯泼溅SLAM在线系统，将前馈几何预测与相机跟踪解耦：用冻结的流式3D基础模型作为虚拟RGB-D传感器提供深度先验，用显式几何求解器通过特征重投影估计相机运动，并用流式高斯映射器构建密集场景表示。在有限优化预算下实现高保真重建和响应式跟踪，重建质量优于现有方法，吞吐量显著提升。

Details

Motivation: 现有单目RGB高斯泼溅SLAM在有限优化预算下难以同时保证高质量重建和响应式相机跟踪，且学习式相机预测的跟踪精度不足，需要结合显式几何先验来提升跟踪准确性和效率。

Result: 在室内外数据集上显著提升重建质量，单张RTX 4090上约7 FPS，吞吐量是所评估先前SLAM系统的2-3倍；几何跟踪在多种代表性流式几何骨干上均能提升相机位姿精度。

Insight: 创新点在于将前馈几何预测与相机跟踪解耦，利用冻结的流式3D基础模型提供深度先验，并用显式几何求解器进行跟踪，结合流式高斯映射；客观看，该方法通过基础模型先验增强在线优化，在提升跟踪精度的同时保持高吞吐，且对多种几何骨干具有通用性。

Abstract: Achieving high-quality reconstruction and responsive camera tracking remains challenging for monocular RGB Gaussian splatting SLAM under a limited optimization budget. We present an online system that addresses this challenge by combining learned geometric priors with explicit geometric camera tracking, which can substantially outperform learned camera prediction in terms of tracking accuracy. A frozen streaming 3D foundation model serves as a virtual RGB-D sensor, providing depth estimation from monocular video. An explicit geometric solver then estimates camera motion through feature reprojection, and a streaming Gaussian mapper integrates the resulting observations into a dense scene representation. We further show that geometric tracking improves camera pose accuracy across representative streaming geometry backbones, demonstrating the broader applicability of our approach. Extensive experiments on indoor and outdoor datasets show that our method significantly improves reconstruction quality over existing approaches, achieving about 7 FPS on a single RTX 4090 and delivering 2-3 times the throughput of the evaluated prior SLAM systems.

Bingchao Wang, Yongcheng Jing, Chengyu Wang, Jun Huang, Jie Yang

TL;DR: 论文提出 AdaGround，一个无需训练的 GUI grounding 框架，由三部分组成：Token 归一化自适应缩放根据模型自身对目标尺寸的估计，为每个样本设定放大倍数、视野和缩放深度；验证器在环精修使用一个对标记点击回答 Yes/No 的验证器，来决定保留或拒绝每次缩放、在缩放链之间选择、触发重新定位，并通过是否存在的查询决定何时弃权；组合结构 grounding 依据表头组合电子表格单元格。

Details

Motivation: 高分辨率屏幕上的小目标 grounding 容易失败，放大往往是一种免训练补救手段；但现有缩放流程没有让缩放程度匹配目标尺寸，而且无条件接受放大后的答案，从不复查错误结果。

Result: 在 Qwen3-VL 系列模型上，AdaGround 超过当前最优的免训练缩放方法和许多经过训练的 GUI 模型；在 OSWorld-G 与 ScreenSpot-Pro 上，相比骨干模型分别提升 7.4 至 20.6 个百分点，增益还可迁移到留出的 UI-Vision。

Insight: 摘要宣称的创新包括：token 归一化的自适应缩放参数设定、验证器在环的接受/拒绝与弃权式闭环精修、以及利用表头结构先验进行组合式 grounding；从客观角度看，可借鉴之处在于用模型自身的尺寸估计来自适应调整超参数，用轻量 Yes/No 验证器为多步缩放提供决策与回退机制，并将 UI 结构先验用于分解组合型目标。

Abstract: GUI grounding on high-resolution screens fails mostly on small targets, and zooming in is a common training-free remedy. However, most existing zooming pipelines do not match the zoom to the target’s size, accept the zoomed answer unconditionally and never revisit a wrong answer. To address these issues, we propose AdaGround, a training-free framework with three components: (1) Token-Normalized Adaptive Zooming, which sets the magnification, field of view and zoom depth of each sample from the model’s own estimate of the target size; (2) Verifier-in-the-Loop Refinement, which uses one marked-click Yes/No verifier to keep or reject each zoom, choose between zoom chains, trigger re-grounding and, with an existence query, decide when to abstain; and (3) Compositional Structure Grounding, which composes spreadsheet cells from their headers. On Qwen3-VL models, AdaGround surpasses state-of-the-art training-free zooming methods and many trained GUI models, improving accuracy over the backbone by 7.4 to 20.6 points on OSWorld-G and ScreenSpot-Pro, and its gains carry over to held-out UI-Vision. Code is available at https://github.com/bcwang-sjtu/AdaGround

[51] PREEN: Exemplar-Based Color Style Transfer with Smartphone Editor Presets cs.CV | cs.LGPDF

Beomjin Ahn

TL;DR: 论文提出PREEN，将彩色风格迁移转化为对智能手机图像编辑器29个参数的预设回归；输入内容图和风格参考图，输出可解释、可直接供编辑器使用的预设向量而非栅格图像。方法利用风格迁移教师模型生成目标效果并反演为伪预设进行监督，预训练CLIP条件回归器并用LoRA和CIELAB损失精调。在PST50上保持结构保真并领先基线，支持端侧亚秒级推理和单一强度旋钮实现默认到鲜艳的连续过渡。

Details

Motivation: 解决传统风格迁移输出栅格图像不可编辑、难以直接用于手机编辑器的问题；关键难点是任意真实风格参考缺乏对应编辑器预设真值，需要无需人工标注地学习参数化颜色风格。

Result: 在人工构建的PST50基准上，PREEN在像素和感知保真度指标上领先基线，同时较好保持结构保真；预设紧凑，端侧推理达到亚秒级。双轴人类研究显示，用户为匹配参考更偏好鲜艳端，但作为独立照片更偏好默认端，单一强度旋钮无需重训练即可覆盖该权衡。

Insight: 创新点在于将颜色风格迁移建模为可解释编辑器参数预设回归而非图像生成，并通过教师模型反演伪预设解决监督缺失；低维参数空间还提供连续、可解释的强度控制，且中间状态均为有效预设。客观看，将生成任务转为编辑器参数回归并用伪标签监督是值得借鉴的思路。

Abstract: We present PREEN (Preset Regression via an Exemplar-based Editor Network), which casts color style transfer as regression onto the 29 parameters of a deterministic image editor. Given a content image and a style reference, PREEN predicts a preset, a vector of editor slider values that is interpretable and editor-consumable, not a raster image. The key obstacle is supervision, since an arbitrary real reference lacks a preset ground truth. We therefore supervise with a style-transfer teacher, whose target looks we invert into the editor’s parameter space as pseudo-presets, then pretrain a CLIP-conditioned regressor and refine it with LoRA under an image-level CIELAB loss. On the human-curated PST50 benchmark, PREEN applies the reference’s color while preserving structural fidelity, leading the baselines on pixel- and perceptual-fidelity metrics. The preset is compact and predicted at sub-second on-device latency. The low-dimensional subspace also exposes a single inference-time strength knob that traces a continuous default-to-vivid trajectory in one forward pass, with every intermediate state a valid editor preset. A two-axis human study shows that raters prefer the vivid end for matching the reference but the default as a standalone photo, a trade-off the knob spans without retraining.

[52] Mining the Nineteenth Century for the Thirteenth: Register-Scale Forced Alignment and HTR Bootstrapping via Legacy Regesta and Antiquarian Editions cs.CV | cs.CLPDF

Tamas Kovacs, Daniel Luger

TL;DR: 论文研究如何利用19世纪Pressutti的regesta和部分古籍版本，为未转录的教皇Honorius III登记簿提供手写文本识别（HTR）辅助。作者在已有现代转录的fol. 1-203上审计并评估这些历史资料，发现它们可在三个层面发挥作用：拼写归一化后的零标签自适应、词汇与NER知识蒸馏、以及条目起点的拓扑定位。结论是古旧学术资料不能替代现代专家转录，但可引导、校验和切分未转录登记簿上的HTR。

Details

Motivation: Honorius III登记簿的HTR依赖现代专家转录，但目前只有Reg. Vat. 9第1-2年有现代转录，其余登记簿缺乏标注数据。论文希望挖掘19世纪regesta和古籍版本中的历史学术资料，以低成本方式支持跨登记簿的HTR自适应、错误检测和结构分割。

Result: 在拼写层，投影到教廷拼写并回放现代源行后，134个第2年页面的WER下降1.50个百分点（95% CI -1.84至-1.15），缩小与oracle差距的33%；词汇层，历史词典覆盖未编辑条目82.2%的OOV词，并以66.7%精度标记识别错误，蒸馏的中世纪NER达到F1 82.1%，与识别后最佳标注器相当；拓扑层找到73.9%的条目起点，并估计fol. 203之后CER为7.80%。

Insight: 创新点在于将19世纪regesta和古籍版本系统化为HTR的弱监督信号，并按拼写、词汇、拓扑三个层面分工利用。客观看，其可借鉴之处是用历史版本做零标签自适应目标、词典与NER蒸馏，以及用regest日期和结尾套语做条目对齐与分割。

Abstract: Handwritten text recognition (HTR) for the registers of Pope Honorius III depends on modern expert transcriptions, which exist only for regnal years 1-2 of Reg. Vat. 9. For the remaining registers, nineteenth-century scholarship offers Pressutti’s regesta and partial editions. On fol. 1-203, where both layers exist, we measure what this legacy layer can contribute, on transcriptions that we first audit against it, and find a division of labor across three tiers. At the orthographic tier, raw editions are at best neutral for adapting a recognizer to the second regnal year and harmful without replay, whereas projection onto curial spelling with replay of modern source lines turns them into zero-label adaptation targets: pooled over all 134 year-2 sides, the WER falls by 1.50 points (95% CI -1.84 to -1.15), closing 33% of the gap to an oracle; the spelling changes at the year split, through the scribe or the transcriber, whereas the hand changes visibly only at fol. 191r. At the lexical tier, the corpus-wide legacy lexicon contains 82.2% of the out-of-vocabulary words of unedited entries and flags recognition errors with 66.7% precision, and labels from pre-trained medieval NER models, distilled into the recognizer’s NER head, reach F1 82.1% on an expert validated benchmark, level with the best tagger applied after recognition. At the topological tier, Pressutti’s regests, the closing datum clauses, and the regests’ dates converted to the chancery’s Roman form find 73.9% of entry starts in the recognized line stream, and edition agreement estimates a CER of 7.80% beyond fol. 203. Antiquarian scholarship thus cannot replace modern transcription, but it can steer, check, and segment HTR across the untranscribed registers.

[53] Explicit Regional Representations For Video Quality Assessment cs.CVPDF

Fuyi Yang, Lei Yang, Andrew Liu, Chenyu Wei, Xiaoyi Guo

TL;DR: 论文针对视频质量评估（VQA）中端到端模型预测所依赖的区域证据通常隐含的问题，提出一种显式的像素级区域表示——质量相关性热图（QRH）。QRH在视觉语言模型引导下融合局部退化证据与感知先验，且无需像素级标注；作者通过分布保持采样和质量感知采样两种确定性片段采样策略将其接入现有VQA模型，并在DOVER及多个VQA基准上验证其有效性。

Details

Motivation: 现有端到端VQA方法主要输出整体质量分数，但其预测背后的区域退化证据缺乏显式表达，限制了模型可解释性和区域信息利用。论文希望构建无需像素级标注的显式区域表示，作为下游VQA模型可利用的通用接口。

Result: 在多个基准上用DOVER进行实验表明：分布保持采样无需重训练即可较好逼近随机采样的平均预测；质量感知采样结合微调后，在多数主流VQA基准的大多数评价指标上取得提升。

Insight: 创新点在于提出由视觉语言模型引导、无需像素级标注的QRH像素级区域表示，并设计两种确定性片段采样策略将其无缝接入现有VQA模型。其可借鉴之处是把隐式区域证据显式化为可复用的质量相关性热图，并通过采样策略连接表示与下游预测。

Abstract: End-to-end video quality assessment (VQA) methods leave the regional evidence underlying their predictions largely implicit. We introduce an explicit pixel-level regional representation for VQA, instantiated as a Quality-Relevance Heatmap (QRH) that combines localized degradation evidence with perceptual priors under vision-language model guidance, without requiring pixel-level annotations. We apply QRH to existing VQA models through two deterministic fragment-sampling strategies: distribution-preserving sampling approximates the original sampler’s QRH-score distribution without retraining, while quality-aware sampling prioritizes regions with high QRH scores and is paired with fine-tuning. Experiments with DOVER on multiple benchmarks show that the former closely approximates predictions averaged over random samples, while the latter improves most evaluated metrics across popular VQA benchmarks. These results support explicit regional representations as a useful interface for downstream VQA.

[54] LearnPrunerV2: Learnable Progressive Token Pruning for Vision-Language Models cs.CV | cs.AIPDF

Rinyoichi Takezoe, Yaqian Li, Weiren Wang, Zihao Bo, Huihong Shi

TL;DR: 论文针对视觉语言模型中视觉token数量巨大导致计算开销高的问题，提出可学习的渐进式token剪枝方法LearnPrunerV2。它将此前的两阶段启发式框架扩展为全可学习架构，在VLM每一层加入轻量模块估计token重要性，并利用发现的层间阶梯冗余模式设计固定每层预算的硬件友好剪枝调度。

Details

Motivation: 现有token剪枝方法存在剪枝准则不可靠、难以准确定位任务相关区域，以及剪枝策略刚性非自适应、忽略不同模型深度token冗余动态变化的问题。

Result: 在多种benchmark和VLM上，LearnPrunerV2持续优于此前SOTA；在LLaVA-v1.5上剪除77.8%视觉token时仍保留99.8%原始性能，实现更优的精度-效率权衡。

Insight: 创新点在于将启发式剪枝升级为全可学习逐层重要性估计，融合内在视觉显著性与通过shadow token捕获的query相关性，并基于层间阶梯冗余模式提出Budget-Deterministic Pruning，实现固定逐层预算的硬件友好调度。

Abstract: Vision-Language Models (VLMs) have demonstrated remarkable multimodal capabilities but suffer from prohibitive computational overhead due to the massive number of visual tokens. While token pruning is a promising solution, existing methods still suffer from two major drawbacks: 1) unreliable pruning criteria that fail to accurately localize task-specific regions due to inherent attention biases or query-agnostic designs, and 2) suboptimal pruning strategies that employ rigid, non-adaptive schedules, overlooking the dynamic evolution of token redundancy across different model depths. To address these challenges, we propose LearnPrunerV2, which extends the previous two-stage heuristic framework (LearnPruner) into a fully learnable architecture. Specifically, we integrate lightweight learnable modules into every layer of the VLM backbone, which estimate token importance by fusing intrinsic visual saliency with extrinsic query relevance captured via a shadow token. By uncovering a consistent staircase redundancy pattern across VLM layers, we further introduce Budget-Deterministic Pruning (BDP) to establish a hardware-friendly schedule with fixed per-layer budgets. Extensive experimental results across various benchmarks and VLMs demonstrate that LearnPrunerV2 consistently outperform previous state-of-the-art methods. Notably, LearnPrunerV2 preserves 99.8% of the original performance when pruning 77.8% of visual tokens on LLaVA-v1.5, achieving a superior accuracy-efficiency trade-off.

[55] RePAIR: Inverting Backdoor Attacks in Semantic Segmentation cs.CV | cs.CR | math.PRPDF

Arnold Brosch, Kira Maag

TL;DR: 论文提出 RePAIR，一种面向语义分割后门攻击的事后、无标签、攻击无关防御方法。它利用后门导致的类别偏移在分类头中比内部表征更明显、且受害者类语义仍可恢复的观察，从可能受污染的无标注图像池推断受害者-目标类关系，定位可疑像素，并在特征空间和输出端向推断出的受害者语义进行修正，同时尽量保持良性预测不变。

Details

Motivation: 语义分割模型易受后门攻击：存在触发器时会把指定受害者类的像素重定向到攻击者目标类，而干净预测保持正常。现有防御多来自图像分类，专门的分割防御稀少且通常需要重训练或修改模型，因此需要一种后处理、无标签且不依赖攻击信息的防御。

Result: 在多种分割架构和攻击设置下，RePAIR 显著降低攻击成功率，同时较好保持干净性能；摘要未给出具体数据集、数值指标或明确的 SOTA 表述。

Insight: 创新点在于利用分类头预测与内部语义证据之间的不一致来定位后门像素，并从受污染无标注图像池中推断受害者-目标关系后做特征与输出修正；理论上将攻击建模为对抗概率传输、将修复视为其近似逆过程，并给出定位、残余攻击成功率、干净侧变化等界限与充分条件。

Abstract: Despite significant advances in computer vision, semantic segmentation models remain vulnerable to backdoor attacks, posing a security risk. These attacks redirect pixels of a chosen victim class to an attacker-specified target class when a trigger is present, while preserving clean predictions. Existing defenses are largely derived from image classification, while dedicated segmentation defenses remain scarce and typically require retraining or model modification. We introduce RePAIR, a post-hoc, label-free, attack-unaware defense based on the observation that backdoor-induced shifts can be stronger in the classifier head than in internal representations, leaving recoverable victim-class semantics. RePAIR infers the victim-target relation from an unlabeled, potentially contaminated image pool, constructs a safety-approved semantic reference, and localizes suspicious pixels by comparing head predictions with internal semantic evidence. It then applies feature-space and output corrections toward the inferred victim semantics, while conservative localization and output constraints limit changes to benign predictions. We model the attack as adversarial probability transport and repair as its approximate reversal, deriving bounds and sufficient conditions for localization, residual attack success, clean-side changes, defense-induced target predictions, and attack-unaware pair identification. Experiments across segmentation architectures and attack settings show that RePAIR substantially reduces attack success while largely preserving clean performance.

[56] Ground-to-Satellite Knowledge Distillation for Crop Classification under Temporal and Geographic Shift cs.CVPDF

Miro Miranda, Rocco Sedona, Valerio Marsocci, Gabriele Cavallaro, Claudia Paris

TL;DR: 论文提出一种地面到卫星的知识蒸馏方法，利用地面照片作为特权监督来提升基于卫星嵌入的作物分类性能。方法使用基于冻结CLIP特征的照片教师模型，在训练时监督卫星分类器（基于AlphaEarth和TESSERA表示），推理时仅需卫星数据。实验表明，该方法在时间偏移和地理偏移下均能提升分类性能，且不增加推理成本。

Details

Motivation: 连续更新的作物类型制图对农业监测至关重要，但参考数据有限。地面照片与卫星时间序列结合可提供互补视角，但推理时并非每个卫星像素都有地面照片，限制了其在大规模作物制图中的潜力。因此，论文研究能否将地面照片作为训练时的特权监督来改进基于预训练卫星嵌入的分类。

Result: 在包含17种作物类别的配对地面照片和卫星影像数据集上，知识蒸馏方法一致优于仅使用卫星的分类；对于冻结的卫星嵌入，在时间偏移下macro-F1提升最高达5.5个点，在地理留出场景下提升更明显（+3.7 vs 同分布+3.0）。

Insight: 创新点在于将地面照片作为特权信息，通过知识蒸馏从冻结的CLIP特征教师模型迁移到卫星分类器，且推理时无需地面照片，保证了可扩展性。客观分析：利用多模态预训练模型提取地面特征作为教师信号，结合卫星预训练嵌入进行跨视角蒸馏，有效提升了模型对时间与地理变化的鲁棒性，是一种值得借鉴的跨模态知识迁移思路。

Abstract: Continuous and up-to-date crop type mapping is essential for agricultural monitoring but is often hampered by limited reference data. The growing availability of ground-level field photographs offers a promising way to address this challenge. When combined with satellite time series, these images provide complementary perspectives on the same parcel, capturing crop characteristics from distinctly different perspectives. However, field photographs are not always available for each satellite pixel at inference time, limiting their potential in scalable crop type mapping. We investigate whether ground photographs can provide privileged supervision for crop classification built on pretrained satellite embeddings. A photo teacher based on frozen CLIP features supervises classifiers trained on AlphaEarth and TESSERA representations; photographs are used only during training. On a paired field photo and satellite imagery dataset with 17 crop classes, knowledge distillation consistently improves over satellite-only classification; for frozen embeddings the margin grows under temporal shift, reaching up to +5.5 macro-F1 points, and widens in geographic holdouts (+3.7 vs +3.0 in-distribution). These results show that ground-view supervision can complement satellite embeddings and improve their robustness without changing inference requirements.

[57] Zooming without Zooming v2: From Seeing Details to Connecting Evidence for Generalizable Multimodal Perception cs.CV | cs.AI | cs.CL | cs.LGPDF

Lai Wei, Yatai Li, Liangbo He, Jun Lan, Renjie Gu

TL;DR: 论文提出多模态感知中的共享底层能力“连接证据”，即识别相关视觉事实、绑定实体、恢复关系并组合答案。为此设计EvidenceConnect数据合成框架，利用单图内相关区域的放大视图生成需要比较、聚合或链式推理的问题，并通过多VLM答案一致性和区域消融过滤保证质量；再用RLVR训练基础VLM得到ZwZ2。该方法虽不针对特定下游基准合成数据，却提升了细粒度与空间感知，并泛化到具身空间推理、多模态长上下文理解、多模态智能体任务和通用视觉推理。

Details

Motivation: 空间感知和多模态长上下文理解对具身AI与多模态智能体至关重要，但针对这些任务合成高质量训练数据困难且昂贵。论文的出发点是：这些任务共享“连接证据”能力，而针对该能力合成训练数据比直接针对下游任务更容易。

Result: 摘要未给出具体数值或benchmark名称，但声称ZwZ2在细粒度和空间感知上取得提升，并在具身空间推理、多模态长上下文理解、多模态智能体任务和通用视觉推理上进一步增益。结果表明EvidenceConnect能有效监督证据连接，并带来越来越泛化的多模态感知能力。

Insight: 创新点在于将多任务统一为“连接证据”能力，并用单图局部放大视图合成比较、聚合、链式问题，配合独立VLM答案一致性与区域消融过滤来保障数据质量，最后用RLVR训练。客观来看，其可借鉴之处是把数据合成从任务特化转向共享能力监督，并通过不针对下游benchmark的合成数据验证泛化性。

Abstract: Multimodal perception forms the foundation of multimodal intelligence. In particular, spatial perception and multimodal long-context understanding are two pivotal tasks that underpin embodied AI and multimodal agents. However, synthesizing targeted, high-quality training data for these tasks remains difficult and costly. Our key insight is that these tasks share an underlying capability, which we call connecting evidence: identifying relevant visual facts, binding them to the correct entities, recovering their relations, and composing them into an answer. Crucially, targeted training data are far easier to synthesize for this capability than for these tasks themselves. To strengthen this capability, we propose EvidenceConnect, an easy-to-implement data synthesis framework that uses zoomed views of related regions within a single image to generate questions requiring comparison, aggregation, or chaining of visual evidence. To ensure quality, a reference answer is retained only when independent VLMs agree on it, and region-ablation tests discard questions whose answer survives the removal of selected regions. We then train base VLMs on the resulting data with RLVR, obtaining Zooming without Zooming v2 (ZwZ2). Although the synthesized data are not tailored to any downstream benchmark, ZwZ2 improves fine-grained and spatial perception, with further gains on embodied spatial reasoning, multimodal long-context understanding, multimodal agent tasks, and general visual reasoning. These results establish EvidenceConnect as an effective framework for supervising evidence connection and suggest that strengthening this shared capability leads to more generalizable multimodal perception.

[58] Learning to Code the World: Harness Evolution for Compositional 3D Reconstruction cs.CVPDF

Haoyu Yao, Chenguo Lin, Ruize Du, Yadong Mu

TL;DR: 论文提出 SceneEvolve，将3D场景表示为可执行代码，并用强教师智能体进化小型开源智能体的执行外壳（harness），以从单张真实图像重建可编辑、组合式3D场景。方法结合 SceneAct 组合语义动作接口与 SceneHarness-RSI 迭代进化流程，在不更新模型参数的情况下让小型模型泛化到未见场景。

Details

Motivation: 从真实图像重建可编辑、组合式3D场景需要联合推理几何、物体结构和空间关系，而小型开源智能体往往难以直接完成此类复杂组合推理；论文希望在不微调模型参数的前提下提升其重建能力并降低推理与部署成本。

Result: 在两个同域数据集和一个跨域数据集上，SceneEvolve 使用 Qwen3.8-27B 提升了全部四项重建指标，并将平均推理时间降低47.0%；相比 GPT-5.6-Sol，重建质量高32.3%、等效部署成本低59.8%，且进化出的外壳可泛化到 Kimi-K3 并达到与 GPT-6-Astra 相当的重建性能。

Insight: 创新点在于把3D场景重建转化为可执行代码生成，设计参数化基元、层级组合和语义空间关系组成的 SceneAct 动作接口，并通过教师反馈自动进化可复用的智能体外壳。客观来看，其核心思路是将能力提升从模型参数更新转移到执行流程/外壳的自动进化上，从而增强小模型的组合推理、跨场景泛化和成本效率。

Abstract: Reconstructing editable and compositional 3D scenes from a real image requires jointly reasoning about geometry, object structure, and spatial relationships. We introduce SceneEvolve, a framework that represents 3D scenes as executable code and improves reconstruction by evolving the harness around a small open-source agent. SceneEvolve combines SceneAct, a compositional semantic action interface built on parameterized primitives, hierarchical composition, and semantic spatial relations, with SceneHarness-RSI, which leverages a stronger teacher agent to iteratively evolve the harness using feedback from development-set reconstructions. Throughout harness evolution, the model parameters, action semantics, and execution backend remain fixed. The resulting harness provides a reusable reconstruction procedure that enables the small open-source model to reconstruct unseen scenes without parameter updates or teacher assistance. Across two in-domain datasets and one out-of-domain dataset, SceneEvolve with the comparatively compact Qwen3.8-27B model improves all four evaluated reconstruction metrics while reducing mean inference time by 47.0% relative to the base model. More notably, despite relying on a substantially smaller model, SceneEvolve achieves 32.3% higher reconstruction quality than the large proprietary GPT-5.6-Sol model, while reducing mean equivalent deployment cost by 59.8%. Furthermore, the harness evolved using Qwen3.8-27B also generalizes to Kimi-K3, achieving comparable reconstruction performance to GPT-6-Astra. The project page is https://haoyu-agent.github.io/SceneEvolve.

[59] Learning to Look Beyond Screens: Transferable Visual Observation from GUI Grounding cs.CV | cs.AIPDF

Ruilin Yao, Xinyu Qiu, Tianyu Zou, Yi Rong, Shengwu Xiong

TL;DR: 论文提出 LBS（Look Beyond Screens），将 GUI grounding 作为可扩展的空间监督来源，通过教师生成的 GUI grounding 轨迹进行监督初始化，再用强化学习训练多轮“裁剪-检查”决策，从而学习可迁移的主动视觉观察行为。基于 Qwen3.5-35B-A3B 和 Qwen3.8-27B 的 LBS 模型在固定推理设置下均优于对应基线，并展现向通用视觉感知与多模态基准的迁移能力。

Details

Motivation: 视觉感知正从被动识别转向主动的、任务导向的观察；论文希望解决如何利用 GUI grounding 提供可扩展空间监督，并让学到的观察策略迁移到界面之外的问题。

Result: 在固定推理设置下，LBS 在 Qwen3.5-35B-A3B 和 Qwen3.8-27B 两个骨干上均稳定优于对应基线；超越监督微调的收益可迁移到 PerceptionBench（两个骨干）以及 HR-Bench 和 ChartQAPro（Qwen3.5-35B-A3B）。

Insight: 创新点在于把 GUI grounding 作为可迁移主动视觉感知的监督范式，并采用“SFT 建立视觉检查原语 + RL 优化多轮裁剪检查决策”的两阶段训练；分析表明两阶段分别负责基础观察能力和下游有效性提升。

Abstract: Visual perception is evolving from passive recognition toward active, task-oriented observation. We investigate GUI grounding as a scalable source of spatial supervision for learning observation behaviors that transfer beyond interfaces. We introduce LBS (Look Beyond Screens), a vision-language model trained through supervised initialization on teacher-generated GUI grounding trajectories, followed by reinforcement learning for multi-turn crop-and-inspect decisions. Under fixed inference settings, LBS models based on Qwen3.5-35B-A3B and Qwen3.8-27B consistently improve over their corresponding baselines. Gains beyond supervised fine-tuning transfer to PerceptionBench for both backbones and to HR-Bench and ChartQAPro for Qwen3.5-35B-A3B, supporting observation strategies that generalize beyond GUI-specific imitation. Further analysis suggests complementary roles for the two stages: SFT establishes visual inspection primitives, while RL improves their downstream effectiveness. These findings support GUI grounding as a practical supervision paradigm for transferable active visual perception and broader multimodal reasoning. Our code and model are publicly available at https://github.com/inclusionAI/UI-Venus/tree/LBS and https://huggingface.co/inclusionAI/LBS-27B, respectively.

[60] TriDifNet: A Unified Three-Task Diffusion Network for End-to-End RGB-T Perception cs.CVPDF

Qianchao He, Jing Liu, Yunhao Li, Xiaoqing Tian

TL;DR: 论文提出 TriDifNet，一个面向 RGB-T 行人感知的统一三任务扩散网络，将红外-可见光图像配准、融合与检测联合建模并端到端优化。网络包含配准扩散模块 RegDM、融合扩散模块 FusDM 和检测扩散模块 DetDM，并通过闭环优化机制让检测语义反向指导融合与配准。实验在 KAIST 多光谱行人基准上验证了其在错位 RGB-T 场景下的检测优势。

Details

Motivation: 红外与可见光图像常因平台运动、视角变化和模态差异而错位，配准误差会传播到后续融合与检测阶段。论文旨在解决错位 RGB-T 场景下配准、融合和行人检测难以联合优化的问题。

Result: 在 KAIST 多光谱行人基准上，TriDifNet 在错位 RGB-T 设置下取得优于强系统级对比方法的下游检测性能，达到 65.44% Recall、57.55% mAP50 和 26.83% mAP50:95。

Insight: 论文宣称的创新点包括首个联合优化配准、融合与检测的扩散式 RGB-T 感知网络，以及由 RegDM、FusDM、DetDM 组成的三模块架构。其可借鉴之处在于闭环优化机制：检测语义指导融合与配准，融合结果进一步正则化几何对齐，并结合分阶段渐进训练与 Nash-MTL 后期微调平衡多任务优化。

Abstract: Infrared–visible (IR–VIS) multi-modal perception supports robust scene understanding under challenging illumination. Infrared and visible images are often misaligned due to platform motion, viewpoint changes, and modality-dependent distortions, causing registration errors to propagate into subsequent fusion and detection stages. We propose TriDifNet, a unified three-task diffusion network for end-to-end RGB-T perception. To the best of our knowledge, TriDifNet is the first diffusion-based network that jointly optimizes registration, fusion, and detection for misaligned RGB-T pedestrian perception. TriDifNet consists of three diffusion-based modules, i.e., the Registration Diffusion Module (RegDM), the Fusion Diffusion Module (FusDM), and the Detection Diffusion Module (DetDM). RegDM aligns the unaligned infrared image to the optical reference by estimating a dense deformation field. FusDM performs residual conditional diffusion to generate a detection-oriented fused representation with optical structural cues and complementary infrared saliency. DetDM formulates pedestrian detection as an iterative denoising process in bounding-box space. To couple the three tasks, TriDifNet introduces a closed-loop optimization mechanism, where detection semantics guide fusion and registration, and the fusion result further regularizes geometric alignment. A stage-wise progressive training strategy and a lightweight Nash multi-task learning (Nash-MTL)-based late-stage refinement are further adopted to balance registration, fusion, and detection optimization. Extensive experiments on the KAIST multispectral pedestrian benchmark show that TriDifNet achieves superior downstream detection performance under misaligned RGB-T settings. Against strong system-level competitors, TriDifNet achieves 65.44% Recall, 57.55% mAP${50}$, and 26.83% mAP${50:95}$.

[61] SplatScout: Query-Driven Part-Level Referring Segmentation in Gaussian Scenes cs.CVPDF

Yihan Zang, Da Li, Shinkyu Park, Ivan Viola, Dominik Engel

TL;DR: 论文提出 SplatScout，一种面向高斯场景的查询驱动闭环指代分割方法，将感知置于 VLM 控制下，使其主动选择视角、指导分割并检查渲染后的 3D 结果以迭代精炼预测。作者同时构建了部件级评测基准 PartRef-ScanNet++，包含 12 个 ScanNet++ 场景中 261 个人工验证查询及基于网格的真值。

Details

Motivation: 现有高斯场景指代分割多针对整体对象，依赖分割器预生成候选再由 VLM 选择，固定候选可能遗漏作为目标或识别线索的细粒度部件。论文旨在实现更灵活的部件级指代分割。

Result: 在 PartRef-ScanNet++ 上达到 0.538 mIoU，显著优于现有方法；消融验证迭代精炼有效，跨 VLM 主干比较表明有效利用工具反馈很重要；并在四个现有指代基准上取得更高分数。

Insight: 创新点在于将 VLM 从候选选择器转变为闭环感知控制器，通过主动视角选择、分割指导和渲染结果检查实现迭代精炼；同时提出部件级指代分割基准，推动细粒度评测。客观看，闭环工具反馈与部件级基准构建是主要可借鉴之处。

Abstract: Existing referring segmentation methods on Gaussian scenes primarily target whole objects, relying on segmenter-generated candidates from which a vision-language model (VLM) selects. Such predetermined fixed candidates may omit fine-grained parts needed as targets or cues for identifying objects. We introduce SplatScout, a query-driven, closed-loop approach that places perception under VLM control. The model actively selects views, directs segmentation, and inspects rendered 3D results to refine its predictions. For part-level evaluation, we introduce PartRef-ScanNet++, comprising 261 manually verified queries across 12 ScanNet++ scenes with mesh-derived ground truth. On PartRef-ScanNet++, SplatScout achieves 0.538 mIoU, substantially outperforming existing methods. Ablations demonstrate the benefits of iterative refinement, while comparisons across VLM backbones highlight the importance of effective tool-feedback utilization. SplatScout also achieves higher scores on four existing referring benchmarks. Code and the PartRef-ScanNet++ benchmark are available at https://github.com/kaust-violab/SplatScout.

[62] A First-Step Bias Restores Lost Diversity in Diffusion Model Distillation cs.CV | cs.AIPDF

Wenwu Tang, Dong Wang, Yun Cheng, Olga Saukh

TL;DR: 论文发现分布匹配扩散模型蒸馏中，输出多样性主要丢失在第一个高噪声去噪步；为此提出ReSeed后蒸馏方法，仅在学生的第一步添加种子响应偏置适配器，并让其速度对齐教师第一区间割线速度，同时冻结学生全部权重和后续去噪步骤。该方法在多个文生图和文生视频模型上以极少参数和数据恢复多样性，并保持有竞争力的视觉质量。

Details

Motivation: 少步蒸馏模型虽能高质量生成，但在同一提示下会将不同初始噪声映射到相似输出，导致多样性丢失；论文旨在定位多样性丢失发生的采样阶段，并以低成本方式恢复多样性。

Result: 在SD3-Medium、SD3.5-Medium、FLUX.1-dev和Wan2.1-T2V-1.3B上评估；对4步FLUX.1-dev蒸馏模型，仅增加相对基模型0.00196%的可训练参数，用8000个prompt-noise对训练，DINO多样性提升48.3%，同时保持有竞争力的视觉质量。

Insight: 创新点在于发现第一个高噪声步是恢复多样性的关键，并提出仅调第一步的轻量后蒸馏适配器，使第一步速度与教师第一区间割线速度对齐；其客观创新在于无需微调完整学生模型、无需目标图像/感知损失/对抗损失，即可数据高效地修复蒸馏模型的多样性退化。

Abstract: Distribution-matching distillation enables high-quality image generation in only a few denoising steps. However, distilled student models often map different initial noise samples to similar outputs under the same prompt. To locate where diversity is lost, we compare the teacher and student along the sampling trajectory. The largest measured reduction in the response to different initial noises occurs during the first, high-noise step. When we replace the first step with the teacher’s corresponding trajectory and leave the final three student steps unchanged, most of the output diversity returns. This suggests that the first step is a key point for restoring diversity. Motivated by this finding, we introduce \emph{ReSeed}, a post-distillation method that adds a seed-response bias adapter only to the first student step and aligns its velocity with the teacher’s first-interval secant velocity. All student weights and later denoising steps remain unchanged. We evaluate ReSeed on three text-to-image models (SD3-Medium, SD3.5-Medium, and FLUX.1-dev) and one text-to-video model (Wan2.1-T2V-1.3B). For a distilled 4-step FLUX.1-dev model, \emph{ReSeed} requires only $0.00196%$ additional trainable parameters relative to the base model and is trained on 8,000 prompt–noise pairs, while keeping the distilled student fully frozen and requiring no target images, perceptual losses, or adversarial losses. It improves DINO diversity by $48.3%$ while maintaining competitive visual quality. These results show that substantial diversity can be recovered through a lightweight, data-efficient tuning of only the first denoising step, without finetuning the full student model.

[63] Multibranch Ensemble Head in Vision Foundation Models for Efficient Cucumber Phenological Stage Classification cs.CVPDF

Kostiantyn Khabarlak

TL;DR: 论文针对数据稀缺且域偏移的小数据集上视觉基础模型（VFM）适配困难的问题，提出一种正则化多分支集成头：训练时随机采样低容量或高容量分支，推理时在概率空间平均两分支。以黄瓜物候阶段分类为实际案例，在NALEF食品安全数据集上验证，DINOv2 ViT-S加多分支头仅用22.3M参数和12.2 GFlops即达到Macro F1 0.978、Cohen’s Kappa 0.971，匹配参数量与计算量十倍以上的模型，适合边缘部署。

Details

Motivation: VFM如DINOv2虽提供强迁移特征，但在小规模、域偏移数据上微调仍易过拟合或效率低；精准农业中黄瓜物候阶段分类受遮挡和细微叶片形态变化困扰，需要高精度且轻量的方案。

Result: 在NALEF食品安全数据集上，所提DINOv2 ViT-S加多分支头取得Macro F1 0.978、Cohen’s Kappa 0.971，性能匹配参数量和GFlops均大十倍以上的模型，仅需12.2 GFlops和22.3M参数。

Insight: 摘要宣称的创新点是正则化多分支集成头：低/高容量分支配对、训练随机采样单分支、推理概率平均，几乎不增加参数和推理开销。客观来看，其创新在于将集成与随机分支正则化结合用于VFM小数据微调，以极低计算代价改善准确率-效率权衡，并验证了在农业细粒度分类任务上的实用性。

Abstract: Vision Foundation Models (VFMs) pretrained with self-supervised learning, such as DINOv2, provide highly transferable features, yet adapting them to small, domain-shifted datasets is still challenging. We propose a regularizing multibranch ensemble head: a training and inference scheme that pairs low- and high-capacity branches, samples a single branch stochastically at each training step, and averages both branches in probability space at inference. The design adds negligible parameters and essentially no inference overhead over a single heavy head, while improving the accuracy-efficiency trade-off under data-scarce conditions. Accurate classification of plant growth stages is critical for precision agriculture, enabling timely decisions on fertilization, irrigation, and harvesting. Therefore, as a challenging practical case we use cucumber phenological stage classification, in which flowers and fruits are frequently occluded by foliage, and stage transitions manifest primarily as subtle changes in leaf morphology. The NALEF Food Safety dataset contains images of five cucumber growth stages captured under diverse real-world conditions. On it we evaluate a broad range of backbones, including DINOv2 and DINOv3 ViT and ConvNeXt variants, against strongly supervised counterparts (ResNet, EfficientNet, ViT). The proposed DINOv2 ViT-S with multibranch head achieves a Macro F1 of 0.978 and Cohen’s Kappa of 0.971, matching the performance of models more than ten times larger in parameters and GFlops, while requiring only 12.2 GFlops and 22.3M parameters, making it deployable on smartphones and edge devices for phenological stage classification.

Tom Sanitz, Emre Gülsoylu, Simone Frintrop, Christian Wilms

TL;DR: 论文提出 FastBEVFusion，一种面向自动驾驶实时 3D 目标检测的简单、快速且鲁棒的 LiDAR-相机多模态融合系统。方法将各模态特征投影到半对齐的模态特定 BEV 空间，避免昂贵的深度估计，并通过基于 cross-attention 的轻量软动态融合模块进行全 BEV 空间特征融合。

Details

Motivation: 现有许多高精度 3D 检测系统难以满足自动驾驶实时性要求，且传统融合方法常依赖计算密集且易出错的深度估计，因此需要兼顾精度、速度和鲁棒性的实时多模态融合方案。

Result: 在 nuScenes 数据集上，FastBEVFusion 优于现有实时 SOTA 系统，达到超过 50 FPS 且无需部署库，并对标定误差表现出较强鲁棒性。

Insight: 创新点在于避免显式深度估计，采用半对齐模态特定 BEV 空间降低融合复杂度，并提出轻量、软性、动态的 cross-attention 融合模块以自适应场景上下文并覆盖整个 BEV 空间；客观来看，其价值主要在实时性、结构简化和标定误差鲁棒性之间的平衡。

Abstract: Reliable real-time 3D object detection based on LiDAR and camera data is critical for safety in autonomous driving, where systems must accurately perceive their surroundings with minimal latency. Many state-of-the-art systems achieve high-quality detection results but fail to meet real-time constraints, limiting their practical use. To address this challenge, we propose FastBEVFusion, a simple, fast, and robust fusion system for 3D object detection. Based on modality-specific features, FastBEVFusion projects these features into semi-aligned modality-specific BEV spaces, without relying on a computationally intensive and potentially error-prone depth estimation. Given these BEV spaces, we propose a light-weight, soft, and dynamic fusion module based on cross-attention, which adapts dynamically to the scene context and enables feature fusion across the entire BEV spaces. Experiments on the complex nuScenes dataset show that FastBEVFusion outperforms state-of-the-art real-time systems, and achieves over 50 FPS without deployment libraries. Additionally, our system shows strong robustness to calibration errors, making FastBEVFusion a powerful 3D object detection system for real-world applications.

[65] VGIQA: Fine-Grained Image Quality Assessment with Visual Grounding Rewards cs.CVPDF

Weichang Wang, Liyu Chen, Shuyue Guan, Jie Hu, Fei Gao

TL;DR: 论文提出VGIQA，一种同时具备局部和全局图像质量评估能力的IQA模型。该方法采用冷启动加强化学习的两阶段训练范式，并设计局部质量评估奖励函数，使模型能够进行细粒度、定位化的质量判断。作者还构建了局部质量评估数据集VGSR-2K，实验表明VGIQA在该数据集上显著优于现有开源MLLM，并能提升退化感知任务性能。

Details

Motivation: 现有基于MLLM的IQA方法主要局限于整体图像级评估，缺乏细粒度、局部化的质量评估能力，限制了其实际应用。论文旨在赋予IQA模型局部与全局联合评估能力。

Result: 在自建局部质量评估数据集VGSR-2K上，VGIQA显著优于Qwen2.5-VL-72B-Instruct和Qwen3-VL-32B-Instruct等开源MLLM；同时，局部质量评估任务能有效增强模型在退化感知任务上的性能。

Insight: 论文的创新点在于将视觉定位与IQA结合，通过局部质量评估奖励函数和冷启动加强化学习的两阶段训练实现细粒度质量评估，并构建VGSR-2K数据集；同时验证了局部质量评估对退化感知任务的正向迁移作用。

Abstract: Image quality assessment (IQA) aims to evaluate the perceptual visual quality of images. Recently, the rapid development of Multi-modal Large Language Models (MLLMs) has significantly advanced the IQA field. By integrating chain-of-thought (CoT) reasoning and reinforcement learning (RL), these models can generate more accurate and interpretable results while relying on less training data. However, existing MLLM-based methods are predominantly constrained to holistic, image-level evaluations and lack the capability to produce fine-grained, localized assessments, which substantially limits their practical applicability. To address this limitation, we propose VGIQA, a novel IQA model endowed with both local and global quality assessment capabilities. VGIQA is trained via a two-stage paradigm comprising a cold-start phase followed by reinforcement learning. Driven by carefully designed local quality assessment reward functions, our model exhibits exceptional image quality understanding on VGSR-2K, a new dataset we propose specifically for local quality evaluation, significantly outperforming existing open-source MLLMs such as Qwen2.5-VL-72B-Instruct and Qwen3-VL-32B-Instruct. Furthermore, the local quality evaluation task can effectively enhance the model’s performance on the degradation perception task.

Yongqi Shan, Lu Zhang, Jiazuo Yu, Yunzhi Zhuge, Huchuan Lu

TL;DR: 论文提出3D-SceneQ，一种用于3D场景理解的3D LLM框架，通过查询引导自适应剪枝（QGAP）和多模态对象级特征增强（MOFE）模块，解决现有3D-LLM中任务无关对象噪声和仅依赖点云缺乏2D语义的问题，在多个3D语言推理任务上取得显著提升。

Details

Motivation: 现有3D-LLM将大量3D对象token与文本token拼接输入LLM，但存在两个问题：一是包含过多任务无关对象数据，引入噪声导致推理准确率下降和幻觉增加；二是仅依赖点云数据，缺乏2D模态中的颜色、材质、纹理和高级上下文关系等丰富语义信息。

Result: 在多个benchmark上超越当前SOTA方法，例如在Scan2Cap上CIDEr提升5.7，在SQA3D上EM@1提升17.0，展示了在多种3D语言推理任务中的强大能力。

Insight: 创新点在于：提出查询引导自适应剪枝模块，基于用户指令语义自适应选择任务相关对象，利用可学习潜在查询保留关键信息并抑制无关噪声；同时引入多模态对象级特征增强模块，将预训练2D模型的特征嵌入整合到3D表示中，丰富对象级语义信息。客观分析认为，结合指令引导的token剪枝与跨模态特征融合是提升3D场景理解效率与准确性的有效思路。

Abstract: Current 3D scene understanding pipelines typically concatenate vast numbers of 3D object tokens with text tokens and feed the resulting sequence to a Large Language Model (LLM). However, existing 3D-LLMs for scene understanding encounter two major limitations: (i) excessive inclusion of task-irrelevant object data, introducing noise that reduces reasoning accuracy and increases hallucinations; and (ii) reliance solely on point cloud data, which inherently lacks the rich semantic information available in complementary 2D modalities, such as color, material properties, texture, and high-level contextual relationships. To address these challenges, we introduce 3D-SceneQ, a 3D LLM that unifies adaptive token pruning with multimodal semantic enrichment, markedly advancing scene understanding, reasoning, and grounding. Specifically, we propose a Query-Guided Adaptive Pruning (QGAP) module that adaptively selects task-relevant objects based on the intended meaning of user instructions, leveraging a learnable latent query to retain critical information while effectively suppressing irrelevant noise. In addition, we introduce a Multi-modal Object-level Feature Enhancement (MOFE) module, which integrates 2D feature embeddings from pre-trained models into 3D representation to enrich object-level semantic information. Evaluations show that 3D-SceneQ outperforms current state-of-the-art methods across multiple benchmarks (e.g., +5.7 CIDEr on Scan2Cap and +17.0 EM@1 on SQA3D), demonstrating its remarkable capabilities in various 3D language reasoning tasks.

[67] Uncovering Shortcut Amplification in Video RLVR through the Lens of Perceptual Uncertainty Collapse cs.CV | cs.CLPDF

Jiafeng Liang, Xianhao Ou, Tao Ren, Rongchuan Mu, Runxuan Liu

TL;DR: 论文研究视频 RLVR 训练中的捷径放大问题：结果级优化会让大型多模态模型越来越依赖文本先验答案，即使视觉证据不支持。作者从感知不确定性角度分析训练动态，发现感知不确定性坍缩现象，并提出 BSPO 自蒸馏策略优化框架，通过推理敏感 token 识别和捷径感知视觉扰动来约束策略更新，从而在提升视频推理性能的同时增强视觉忠实性。

Details

Motivation: RLVR 虽能提升 LMM 的视频推理能力，但结果级优化可能诱导模型走捷径，偏向文本先验而忽视视觉证据。论文旨在刻画这一行为的训练动态，并缓解视频 RLVR 中的捷径放大问题。

Result: 摘要称大量实验验证 BSPO 能提升视频推理性能并增强视觉忠实性，但未给出具体 benchmark、数值或 SOTA 对比结果。

Insight: 论文宣称的创新点包括：从感知不确定性视角揭示 RLVR 中的感知不确定性坍缩与捷径放大，并通过梯度方向分析说明置信度提升逐渐偏离视觉 grounded 的不确定性降低；提出 BSPO，用组内正确/错误 rollout 提取推理敏感 token，并用捷径感知视觉扰动惩罚视觉证据弱化下仍不变的预测，同时自适应重加权原始 RL advantage。客观看，其价值在于将训练动态诊断与针对性正则化结合，兼顾任务收益和视觉忠实性。

Abstract: Reinforcement learning with verifiable rewards (RLVR) has emerged as an effective post-training paradigm for enhancing the video reasoning capability of Large Multimodal Models (LMMs). Despite its empirical effectiveness, we show that outcome-level optimization can introduce an undesirable shortcut amplification effect, whereby models increasingly favor text-prior answers even when they are not supported by the visual evidence. To characterize this behavior, we investigate the training dynamics of video RLVR through the lens of perceptual uncertainty, analyzing how predictive uncertainty evolves under video-conditioned and text-only contexts. Our analysis identifies a perceptual uncertainty collapse phenomenon, in which RL training continuously sharpens the model’s answer distribution while the uncertainty reduction attributable to visual evidence progressively diminishes. A further gradient-direction analysis reveals that such confidence enhancement becomes increasingly less aligned with visually grounded uncertainty reduction and is instead dominated by text-prior compression. Motivated by these findings, we propose Bifocal Self-distillation Policy Optimization (BSPO), a self-distillation framework that regulates policy updates with two complementary distributional signals. Specifically, BSPO derives empirical reasoning principles from intra-group correct and incorrect rollouts to identify reasoning-sensitive tokens, and introduces shortcut-aware visual perturbation to penalize predictions that remain invariant when task-relevant visual evidence is weakened. By adaptively reweighting the original RL advantage, BSPO preserves the task-solving benefits of RLVR while promoting updates that are both reasoning-sensitive and visually grounded. Extensive experiments validate the effectiveness of BSPO in improving video reasoning performance and strengthening visual faithfulness.

[68] Zhengrong-Spatial-Aware 1.0: Advancing 3D Multimodal Intelligence cs.CVPDF

Yu Zhang, Xiaogang Wang, Yuhang Cheng, Jingrui Li, Wenbo Chen

TL;DR: 论文构建并验证了两条面向3D感知视觉问答的端到端多模态适配流水线：一条微调融合度量尺度3D点图的HiSpatial-3B，另一条用QLoRA对Qwen3-VL进行高效监督微调。工程栈覆盖数据转换、几何预处理、指令构建、混合域训练、推理与自动评估，并在自定义空间基准上取得显著准确率提升。

Details

Motivation: 针对3D感知视觉问答中几何空间推理与通用多模态能力适配的需求，论文旨在建立可复现的端到端适配流水线，并验证几何增强模型与通用视觉语言模型在自定义空间数据集上的 specialization 效果。

Result: 在目标空间基准上，HiSpatial适配将整体准确率从1.73%提升至79.89%（3,600步），在风格化子集上从14.9%提升至55.1%；Qwen3-VL-8B和32B经混合域QLoRA适配后分别从20.74%/30.79%提升至45.27%/47.60%。

Insight: 摘要宣称的创新点在于两条互补且可复现的适配流水线、完整的工程栈，以及发现训练数据组成对保持通用多模态能力的关键作用。客观看，工作更偏系统集成与实证验证，算法创新有限，但为度量3D推理、空间定位和领域适配提供了可复现基线。

Abstract: This report presents the implementation and validation of two end-to-end multimodal adaptation pipelines for 3D-aware visual question answering on a custom spatial dataset. The first pipeline fine-tunes HiSpatial-3B, which augments RGB observations with metric-scale 3D point maps; the second adapts Qwen3-VL through memory-efficient QLoRA-based supervised fine-tuning. The completed engineering stack covers dataset conversion, image-annotation alignment, monocular geometry preprocessing, instruction construction, mixed-domain training, checkpoint management, batch inference, and automated evaluation. On the target spatial benchmark, HiSpatial adaptation increases overall accuracy from 1.73% before fine-tuning to 79.89% at 3,600 steps. On a separately collected styled split, accuracy improves from 14.9% to 55.1%, indicating transfer beyond the original training set. The Qwen3-VL pipeline is likewise implemented end-to-end for both 8B- and 32B-scale models using mixed-domain QLoRA adaptation. The resulting accuracy improves from 20.74% to 45.27% for Qwen3-VL-8B and from 30.79% to 47.60% for Qwen3-VL-32B. Together, these results establish two complementary and reproducible adaptation pipelines for geometry-aware and general-purpose multimodal spatial reasoning. These results establish a reproducible engineering foundation for subsequent work on metric 3D reasoning, spatial grounding, and domain-adapted multimodal systems, while also identifying training-data composition as a central factor in preserving general multimodal capability during specialization.

[69] Vision-Language-Action or World-Action? Just Action Models cs.CV | cs.ROPDF

Yonghao Yu, Lang Huang, Runyi Li, Toshihiko Yamasaki

TL;DR: 论文重新审视VLA与WAM之争，指出二者差异可解耦为骨干初始化、未来观测目标和推理时是否保留未来损失机制三个独立选择，并提出统一框架JAM：预训练视觉骨干加动作专家，使用动作损失和前瞻损失训练。前瞻损失让当前观测特征预测未来观测的冻结编码，预测头训练后丢弃，因此训练和推理均不涉及视觉生成，且骨干可自由选择。

Details

Motivation: 论文关注VLA与WAM社区关于控制策略应继承视觉语言模型知识还是视觉生成器知识的争论，指出直接将两者差异归因于单一因素并不合理，希望解耦关键设计并寻找更简洁统一的动作模型框架。

Result: 在LIBERO、LIBERO-Plus和RoboTwin上，JAM优于对应的FastWAM和ImageWAM，在RoboTwin上成功率相近或略高；相比FastWAM，三组成功率从97.0%/55.3%/91.2%提升至98.2%/64.5%/91.4%。使用缓存前瞻目标时，JAM相对FastWAM将每样本每次更新的估计在线训练FLOPs降低73%。

Insight: 创新点在于将VLA/WAM争议拆解为三个可分离选择，并提出无需视觉生成和额外推理机制的JAM统一框架；前瞻损失通过预测冻结未来编码提供未来监督，训练后丢弃预测头，兼顾性能与效率。客观上看，其价值在于简化未来监督配方、验证跨骨干家族有效性，并显著降低训练开销。

Abstract: The vision-language-action (VLA) and world-action model (WAM) communities debate whether control policies should inherit pretrained knowledge from vision-language models or from visual generators. The debate, however, bundles three separable choices: backbone initialization, the objective applied to future observations, and whether the machinery for that future loss is kept at inference. A gap between a VLA and a WAM therefore cannot simply be attributed to any one of them. We show that decoupling these choices yields a common framework for both families, which we term Just Action Model (JAM): a pretrained visual backbone with an action expert, trained with an action loss and a foresight loss. The foresight loss trains current-observation features to predict frozen encodings of the recorded future through a prediction head discarded after training, so no visual generation is involved in either training or inference, and the backbone can be chosen freely. We first compare objectives within fixed policy architectures, then evaluate the foresight recipe across backbone families. The resulting JAM policies outperform their matched FastWAM and ImageWAM counterparts on LIBERO and LIBERO-Plus, and achieve similar or slightly higher success rates on RoboTwin. Compared with FastWAM, JAM improves the success rates from 97.0%/55.3%/91.2% to 98.2%/64.5%/91.4%, respectively. With cached foresight targets, JAM reduces estimated online training FLOPs per sample per update by 73% relative to FastWAM. Our results show that future supervision improves control without visual generation or additional inference machinery, and strong foresight performance is attainable with either generator-initialized or vision-language backbones. Project Page: https://y0uroy.github.io/JAM

[70] HDVS: semi-supervised semantic segmentation via heterogeneous dual-branch voting supervision cs.CVPDF

Yongqi Shan, Yunzhi Zhuge, Huchuan Lu

TL;DR: 论文提出一种异构双分支投票监督框架HDVS，用于半监督语义分割。该方法通过投票机制融合异构分支的伪标签以提升伪标签可靠性，并利用特征通信模块在特征层面引入扰动，增强双分支预测多样性。在PASCAL VOC 2012和Cityscapes上验证了方法的有效性。

Details

Motivation: 半监督语义分割希望利用大量无标签图像减少人工标注。现有方法通常假设各分支生成的伪标签足够可靠，可用于监督其他分支，但伪标签误差可能在协同训练中累积，导致性能次优。

Result: 论文在PASCAL VOC 2012和Cityscapes两个基准数据集上进行了综合评估，结果表明HDVS优于当前最先进方法。

Insight: 创新点在于基于交叉监督框架引入异构双分支投票机制，将异构分支伪标签相关联以生成更可靠的伪标签；同时设计特征通信模块，在特征级引入扰动，在保持收敛的同时最大化双分支预测多样性。客观来看，其可借鉴之处是将投票集成思想用于伪标签可靠性提升，并结合特征扰动增强分支差异。

Abstract: Semi-supervised semantic segmentation, which aims to utilize large volumes of unlabeled images to achieve accurate segmentation results with fewer human annotations, has attracted increasing attention. Prior methods were primarily based on the assumption that the pseudo-labels generated by each branch were sufficiently reliable to supervise the remaining branches. However, there is a concern that the errors in pseudo-labels could accumulate during co-training, potentially resulting in suboptimal performance. To this end, we present a novel framework called heterogeneous dual-branch voting supervision (HDVS), which is designed to enhance the reliability of pseudo-labels and mitigate the issues arising from pseudo-labeling. Specifically, based on the cross-supervision framework, we introduce a voting mechanism that correlates pseudo-labels from heterogeneous branches to produce pseudo-labels with enhanced reliability. Concurrently, we employ a feature communication module to introduce perturbations at the feature level in each branch to maximize prediction diversity across the dual branches while maintaining convergence. Comprehensive evaluations of the proposed HDVS on two benchmark datasets (PASCAL VOC 2012 and Cityscapes) demonstrate its superiority to the state-of-the-art approaches.

[71] GameVerse: A Minute-Scale Gameplay Dataset for Long-Horizon Interactive World Modeling cs.CVPDF

Kang He, Wenshuo Peng, Chuanhao Li, Zihui Gao, Xiaojie Xu

TL;DR: 论文提出 GameVerse，一个面向长时程交互式世界建模的分钟级游戏视频数据集，包含 69.7K 个片段、1,655 小时、3.496 亿帧，并有 1,358 小时连续两分钟序列。数据经过场景一致分割、质量过滤、相机位姿估计和基于事件的字幕标注，每个片段配有跨分钟稳定的相机轨迹，以及区分静态场景与动态前景事件、并对齐事件边界的字幕。实验表明在该数据集上训练可显著延长交互式世界模型的时间一致性范围。

Details

Motivation: 现有交互式世界模型在约一分钟后难以保持连贯性，核心瓶颈不是模型容量，而是缺乏长时程训练数据；已有游戏视频数据集通常不足十秒且字幕粗粒度、缺乏时序区分，难以支持可控长时程生成。

Result: 论文分析表明 GameVerse 在规模、多样性和标注质量上超过现有公开数据集；实验显示在其上训练能显著扩展交互式世界模型的时间视野，但摘要未给出具体定量指标或 benchmark 数值。

Insight: 创新点在于构建分钟级、长连续的游戏视频数据，并提供跨分钟稳定的相机轨迹与事件对齐字幕；其字幕将静态场景上下文和动态前景事件分离，并按事件边界而非固定时间窗口标注，这对长时程可控世界建模的数据设计具有借鉴意义。

Abstract: Recent advances in interactive world models have made it possible to generate controllable, dynamic environments from action inputs. Yet current models struggle to maintain coherence beyond roughly one minute of rollout, a limitation rooted not in model capacity but in the scarcity of long-horizon training data. Existing gameplay video datasets typically consist of clips under ten seconds, with captions that describe whole clips in coarse and undifferentiated text, which is insufficient for controllable long-horizon generation. We address this with GameVerse, a minute-scale gameplay video dataset purpose-built for long-horizon interactive world modeling. GameVerse contains 69.7K clips spanning 1,655 hours and 349.6 million frames across diverse first- and third-person gameplay, including 1,358 hours of continuous two-minute sequences, an order of magnitude longer than prior datasets. Clips pass through a unified pipeline of scene-consistent segmentation, quality filtering, camera pose estimation, and event-based caption annotation. Each clip is paired with a temporally stable camera trajectory that remains coherent across minutes, together with captions that separate static scene context from dynamic foreground events and align to event boundaries rather than fixed time windows. Comprehensive analysis shows that GameVerse surpasses existing public datasets in scale, diversity, and annotation quality. Experiments further demonstrate that training on GameVerse substantially extends the temporal horizon of interactive world models.

[72] Taming Noise Initialization for Zero-shot Video Restoration with Image Diffusion Priors cs.CVPDF

Chong Wang, Zixuan Fu, Hao Cheng, Siyuan Yang, Alex C. Kot

TL;DR: 论文面向零样本视频恢复，提出利用预训练图像扩散模型的新框架，核心是“驯服”噪声初始化：通过对受损视频帧执行DDIM inversion生成保留时间相关性的初始噪声，并结合自适应噪声混合与measurement-guided trajectory attention提升恢复质量与帧间一致性。

Details

Motivation: 现有方法用预训练图像扩散模型做视频恢复时，常采用逐帧独立或完全相同的高斯噪声初始化，难以捕获和利用退化视频帧中的固有时间相关性；同时预训练视频扩散模型计算成本高。

Result: 在多个视频恢复任务上，方法优于现有零样本方法，并在DAVIS、REDS4和Set8数据集上取得一致更优的结果；摘要未给出具体定量数值。

Insight: 创新点在于用DDIM inversion从受损视频帧本身生成初始噪声以保留时间相关性，并通过自适应噪声混合抑制退化模式、逐步引导至高斯分布，再用measurement-guided trajectory attention增强跨帧特征对齐。客观看，其可借鉴之处是将噪声初始化与退化视频的时空结构耦合，从而在无需预训练视频扩散模型的情况下提升零样本视频恢复的一致性。

Abstract: Recent advances in diffusion models have achieved remarkable success in restoring high-quality images in a zero-shot manner. Unfortunately, extending this success to video restoration is challenging due to the substantial computational resources required to pre-train video diffusion models. A rising trend seeks to address video restoration by utilizing a pre-trained image diffusion model, initiating restoration from either frame-wise independent or identical Gaussian noise. Yet, these approaches often fall short, as such simple noise initiations fail to capture and utilize the inherent temporal correlations within degraded video frames. In this work, we propose taming noise initialization for zero-shot video restoration, which preserves the intrinsic temporal correlations by generating initial noises directly using DDIM inversion on the corrupted video frames themselves. Building on this, we introduce an adaptive noise blending strategy that suppresses degradation patterns encoded during DDIM inversion while gradually steering the noise toward a Gaussian distribution. To further emphasize temporal consistency, we propose measurement-guided trajectory attention to better align features across frames. Extensive experiments on multiple video restoration tasks demonstrate that our method outperforms existing zero-shot approaches, consistently achieving superior results on the DAVIS, REDS4, and Set8 datasets.

[73] TRUCO: A Scalable Lock-Free Algorithm for Parallel Contour Extraction cs.CVPDF

Rafael Munoz-Salinas, Francisco J. Romero-Ramirez, Manuel J. Marin-Jimenez

TL;DR: 论文提出 TRUCO，一种用于将二值掩码转换为几何轮廓的无锁并行算法，面向高分辨率分割结果的后处理场景。该方法通过起点所有权规则与推测式追踪模型，让线程向下追踪轮廓并用向上终止逻辑保证每条轮廓恰好处理一次，从而避免 tile 拼接和复杂同步。实验表明其在消费级和服务器级 CPU 上显著优于现有基线，相较工业标准可获得一个数量级以上的加速，并在高分辨率场景下保持良好可扩展性。

Details

Motivation: 二值掩码到几何轮廓的转换是计算机视觉流水线中的关键后处理步骤，但 Suzuki-Abe 等标准算法本质串行且需构建拓扑层级，难以利用多核 CPU。现代高分辨率应用需要一种无需 tile 拼接、无需重同步的并行轮廓提取方法。

Result: 在合成与真实数据集、消费级与服务器级 CPU 上的实验显示，TRUCO 显著优于现有基线，相比工业标准实现超过一个数量级的加速，并在多样硬件架构的高分辨率场景中保持稳健的可扩展性。

Insight: 论文宣称的创新点是无锁并行轮廓提取，核心是起点所有权规则与推测式追踪结合向上终止逻辑，确保每条轮廓恰好被处理一次。客观看，其创新在于用轻量所有权与推测执行替代锁和复杂同步，为高分辨率掩码后处理提供可扩展的并行化思路。

Abstract: Transforming binary masks into geometric contours remains a critical step in many computer vision pipelines, particularly for post-processing high-resolution segmentation outputs. Yet, standard algorithms such as the Suzuki-Abe method are limited by their sequential nature and the overhead of building topological hierarchies. As modern applications transition to high-resolution imaging, there is a critical need for contour extraction that leverages multi-core CPU architectures without the complexity of tile stitching or heavy synchronization. This paper presents TRUCO (Threaded Raster Unrestricted Contour Ownership), a lock-free parallel algorithm designed for high-performance geometric boundary extraction. TRUCO introduces a start-point ownership rule, combined with a speculative tracing model, in which threads trace contours downward into adjacent regions and use an upward-termination logic to ensure each contour is processed exactly once. Experimental results on synthetic and real-world datasets, conducted on both consumer and server-grade CPUs, demonstrate that TRUCO significantly outperforms existing baselines. The method achieves speedups of over an order of magnitude compared to the industry standard while maintaining robust scalability on high-resolution scenes across diverse hardware architectures.

[74] CodeCap-Bench: Can Code Serve as a Good Video Caption? cs.CVPDF

Jiafu Tang, Yifan Xu, Eric Tang, Junjie Zhao, Shipeng Luo

TL;DR: 论文探索用可执行代码作为视频描述，通过图形基元、布局规则和时间函数表达视觉内容，并与自然语言描述在下游问答和视频重建任务上比较。为评估多模态大模型生成忠实代码描述的能力，作者提出CodeCap-Bench，包含323个视频、五类结构化2D内容、细粒度标注和多级评价框架。

Details

Motivation: 自然语言视频描述在精确表达细粒度空间结构和复杂时间动态时往往冗长且歧义，论文希望探索代码作为一种更精确、可执行的补充表示，并检验当前MLLM能否生成忠实的代码描述。

Result: 实验表明代码描述在问答推理准确性和视频重建保真度上优于自然语言描述；在CodeCap-Bench上，领先闭源MLLM展现出一定代码描述能力，但仍难以忠实捕捉复杂时间动态，开源模型明显落后。

Insight: 创新点在于提出“代码即视频描述”的新范式，用可执行程序及其图形基元、布局规则和时间函数表示视频，并构建CodeCap-Bench及多级评估框架，系统评估MLLM的代码视频描述能力。

Abstract: Video captioning aims to preserve visual information through compact descriptions, predominantly using natural language. However, precisely expressing fine-grained spatial structures and temporal dynamics in natural language often leads to lengthy and potentially ambiguous descriptions. In this work, we explore Code as Video Caption, a paradigm where executable programs serve as a complementary representation of visual content through graphical primitives, layout rules, and functions of time. To examine whether code captions effectively preserve video information, we compare them with natural language captions on downstream question answering and video reconstruction tasks. Our results show that code captions achieve superior reasoning accuracy and reconstruction fidelity. Despite these advantages, realizing the potential of code captioning depends on whether current multimodal large language models (MLLMs) can generate faithful code captions. To systematically assess this capability, we introduce CodeCap-Bench, which comprises 323 videos spanning five categories of structured 2D content, along with fine-grained annotations and a multi-level evaluation framework. Extensive experiments reveal that while leading proprietary MLLMs demonstrate promising code captioning capabilities, they still struggle to faithfully capture complex temporal dynamics, with open-source models lagging substantially behind.

[75] TimeLineXR: An Interactive Exploration of Time in Vision-Language Models cs.CVPDF

Nidham Tekaya, Markus Passecker, Matthias Zeppelzauer

TL;DR: 论文提出 TimeLineXR，一个交互式演示系统，将视觉语言模型（VLM）中隐含的时间理解结构转化为可探索的 3D 时间轴。用户可上传或选择图像，系统用 CLIP 编码、余弦核 KPCA 投影并吸附到 Bézier 时间轴，从而估计图像中物体的首次出现年份。系统同时提供 Web 演示和基于 Three.js 的 WebXR 房间级场景，支持抓取、缩放、检查和投掷图像卡片查询。

Details

Motivation: 已有研究表明 VLM 对时间有内在理解，并在嵌入空间中编码了低维非线性时间流形，但该结构对终端用户不可见、不可查询，也难以用自定义图像检查查询结果在时间轴上的位置。论文旨在将这一时间轴外化为可交互、可探索的 3D 场景。

Result: 摘要未报告定量 benchmark 或 SOTA 结果，主要展示系统实现：Web 端和 WebXR 端共用同一预测管线与缓存时间索引，可将图像映射为时间轴上的点并读取估计的首现年份。

Insight: 宣称创新点在于交互式可视化 VLM 时间理解、Web 与 WebXR 双端探索、以及统一 CLIP+KPCA+Bézier 时间轴的查询管线。客观看，创新更偏向系统集成与可视化交互，把模型内隐时间结构转化为用户可操作的三维表示，而非提出新的 VLM 训练或推理算法。

Abstract: Recently, it has been shown that vision-language models (VLMs) have a certain intrinsic understanding of time. They can estimate the temporal origin of objects and encode time along a low-dimensional, non-linear manifold inside their embedding space~\cite{tekaya2025matter}. That result is quantitative and reproducible, but the structure itself stays out of reach for end users: they cannot see it, query it with their own images, or inspect where a query lands relative to other years on the time axis. We present TimeLineXR, an interactive demonstration that turns this time axis into an explorable 3D scene. A web demo lets users upload or select images and see each placed on a color-coded 3D time axis; a Three.js WebXR scene renders the same time axis as a room-scale representation that can be grabbed, scaled, inspected, and queried by throwing image cards into it. Both clients run the same prediction pipeline against the same cached temporal index: encode the query with CLIP, project it through cosine-kernel KPCA, and snap it to the Bézier time axis. The system turns each image into a point on the time axis, so users can compare where different objects land and read off their estimated first year of appearance.Code source and reproduction steps can be found in: https://github.com/sterngefeuert/timeline-xr.

[76] Representing the Unseen: Rethinking Visual Imagination for Spatial Reasoning cs.CV | cs.AI | cs.CLPDF

Zengqi Zhao, Zun Wang, Shoubin Yu, Mohit Bansal, Yue Zhang

TL;DR: 论文重新审视视觉世界模型用于空间推理时默认生成RGB视图的做法，指出同一未观测状态可采用不同表示，而表示选择会影响下游空间推理效果。作者构建了基于模拟器的ImagiRep基准，包含922个问题、七类问题和两种设置，并提供理想RGB及六种替代表示进行受控比较。实验发现准确想象RGB并不保证准确空间推理，不同问题受益于不同表示；据此提出轻量测试时自适应表示选择策略GEARS，优于现有空间想象方法。

Details

Motivation: 现有视觉世界模型常以生成未观测状态的RGB视图作为空间推理接口，但这更多是因为RGB适合视觉生成，而非最适合下游推理。论文旨在隔离“表示”在空间想象中的作用，解决RGB默认化可能限制空间推理的问题。

Result: 在ImagiRep上的受控比较表明，即使理想RGB准确，也不能保证空间推理准确，不同问题类型受益于不同表示；世界模型估计状态也有类似趋势，且仅估计RGB可能误导推理，而替代表示具有互补优势。GEARS在测试时自适应选择表示，优于现有空间想象方法。

Insight: 论文的创新点在于构建可控的多表示空间想象基准，系统揭示表示选择对空间推理的影响，并提出轻量测试时自适应表示选择策略。客观来看，其可借鉴之处是将“生成什么表示”与“如何推理”解耦，强调多表示互补性，而非单纯追求RGB生成质量。

Abstract: Visual world models increasingly support spatial reasoning by generating RGB views of unobserved states. However, RGB has become the default interface largely because it is well suited for visual generation, rather than necessarily being the most effective representation for downstream reasoning. Different spatial questions may benefit from different representations of the same imagined state: top-down geometry can facilitate global distance comparisons, while panoramic views provide broader context for reasoning about open space. To isolate the role of representation in spatial imagination, we introduce ImagiRep, a simulator-grounded benchmark of 922 questions spanning seven question types across two settings, Viewpoint Imagination and Cross-view Integration, providing oracle RGB and six alternative representations (e.g., depth, panoramic views, and top-down geometry) of the same unobserved state. This controlled comparison reveals that even accurate imagined RGB does not guarantee accurate spatial reasoning, and different questions benefit from different representations. Similar trends hold for representations derived from world-model-estimated states. Notably, estimated RGB alone can mislead spatial reasoning, while alternative representations retain complementary benefits. Building on these findings, we introduce GEARS, a lightweight test-time strategy for adaptive representation selection, outperforming existing spatial imagination methods and showing that reasoning about unseen states requires more than generating RGB views.

[77] PerceptRL: Making Visual Perception Rewardable in MLLMs from Answers to Interactions cs.CV | cs.AIPDF

Tongkun Guan, Zhibo Yang, Mingkun Yang, Haocheng Wang, Wei Shen

TL;DR: 论文提出 PerceptRL，用于增强多模态大模型（MLLMs）后训练中的细粒度视觉感知。它先从密集图像描述中提取可验证的感知错误，并转换为带固定答案和规则奖励的 QA；再通过单轮 Perception RL 和多轮 Perception Agentic RL，结合视觉工具进行感知优化。方法还引入 Coordinate Injection 与 grounding-aware process reward，以缓解工具使用中的坐标漂移和空间操作错误。

Details

Motivation: 细粒度视觉感知是 MLLMs 后训练的瓶颈，感知错误会限制推理优化收益；而图像描述预训练的感知收益在后续训练中会减弱，且密集、主观的描述难以设计可验证奖励。

Result: 在共享工具使用协议下的 7 个 benchmark 上，PerceptRL 平均准确率达到 71.6%，分别超过使用相同工具的基座模型 15.7 个百分点、超过评测中最强前沿 VLM 5.2 个百分点。

Insight: 创新点在于将难以验证的密集 caption 监督转化为可验证 QA 与规则奖励，并通过 Chain QA、多选题 QA、领域专用/通用路线及多轮 Agentic RL 扩展感知监督；同时用 Coordinate Injection 和 grounding-aware process reward 缓解工具调用中的坐标漂移与中间空间证据错误。

Abstract: Fine-grained visual perception remains a bottleneck during post-training of multimodal large language models (MLLMs), as perceptual errors can constrain gains from reasoning-focused optimization. Although image-caption pre-training improves perception, its influence can weaken as subsequent stages change data mixtures and objectives. Continuing caption supervision during post-training is challenging because subjective, information-dense descriptions complicate verifiable reward design. We present PerceptRL, which strengthens perception at two complementary levels. Perception RL uses a multi-agent pipeline to extract verified errors from dense image captions and convert the corresponding visual facts into QA pairs with fixed answer labels and rule-based rewards. Chain QA links correlated errors, while multiple-choice QA combines unrelated errors. A specialized route targets high-error domains with tailored verification, and a domain-agnostic general route covers errors beyond these domains. Perception Agentic RL addresses fine-grained and dense cases unresolved by single-turn Perception RL through multi-turn inspection with visual tools. We observe coordinate drift even in models trained to use tools, where incorrect crops or annotations distort evidence and propagate errors. Coordinate Injection corrects spatial operations in cold-start trajectories, while a grounding-aware process reward supervises intermediate spatial fidelity alongside final-answer correctness. Training proceeds through shared cold-start SFT, single-turn Perception RL, and multi-turn Perception Agentic RL. Across seven benchmarks under a shared tool-use protocol, PerceptRL achieves 71.6% average accuracy, surpassing the base model with the same tools and the strongest evaluated frontier VLM by 15.7 and 5.2 percentage points, respectively.

[78] ALADDIN: Adversarial Latent Diffusion Distillation for One-Step Identity Swapping cs.CVPDF

Ivan Shchekotov, Timofei Gritsaev, Viacheslav Pirogov, Dmitry Vetrov

TL;DR: 论文提出ALADDIN，一种用于一步人脸身份替换的对抗潜在扩散蒸馏方法。它将生成器从预训练的多步flow-matching模型初始化，并在身份约束和目标属性保持约束下进行对抗训练，从而在单步推理中实现高质量换脸。方法还通过triplet伪标签监督提供身份迁移与目标属性保持之间的可控权衡，并重新审视了评估协议中的身份编码器偏差。

Details

Motivation: 现有基于扩散的换脸方法真实感强但每张图需要多次网络评估，速度慢；基于GAN的方法虽快，但从零训练仅靠对抗信号学习真实感，效果明显较弱。论文旨在兼顾一步推理速度、真实感和目标属性保持。

Result: ALADDIN在一步方法中于真实感和目标保持上达到SOTA，并在相当身份迁移水平下匹配最佳多步模型，同时网络评估次数减少28倍、训练数据对减少一个数量级。

Insight: 创新点包括：从预训练多步flow-matching模型初始化生成器并进行对抗蒸馏实现一步换脸；引入身份与目标保持约束及triplet伪标签监督实现可控权衡；指出标准评估偏向使用同一身份编码器训练和打分的方法，并用独立编码器重评基线。

Abstract: Face swapping transfers a source identity onto a target image while preserving the target’s pose, expression, hair, and lighting. The most realistic current methods are diffusion-based, but they need many network evaluations per image. GAN-based methods are fast, yet trained from scratch they must learn realism from the adversarial signal alone and produce visibly weaker results. We present ALADDIN (Adversarial Latent Diffusion Distillation for One-Step Identity Swapping), a one-step face swapper obtained by initialising the generator from a pre-trained multi-step flow-matching model and training it adversarially under identity and target-preservation constraints. We expose a controllable trade-off between identity transfer and target-attribute preservation by training a student model with triplet pseudo-label supervision. We additionally identify a bias in the standard evaluation protocol, which favours methods trained against the same identity encoder used to score them, and re-evaluate all baselines with an independent one. ALADDIN is state-of-the-art among one-step methods in realism and target preservation at comparable identity transfer, and matches the best multi-step model with 28 times fewer network evaluations and an order of magnitude fewer training pairs.

[79] A note on data-efficient finetuning of vision foundation models cs.CV | cs.LGPDF

Niclas Hergenröther, Antonio Orvieto

TL;DR: 论文对视觉基础模型在资源受限场景下的高效微调进行系统性实证评估，比较Attention、卷积和Mamba三类基础模型与VP、线性探测、全量微调和LoRA等迁移范式之间的交互效果，并探索常用正则化技术对视觉提示性能的提升。

Details

Motivation: 基础模型适配新任务通常依赖微调，而视觉提示等高效迁移方法快速发展且模型家族不断扩展，亟需系统评估现有方法在资源限制下的表现、挑战与机会。

Result: 摘要未给出具体定量结果或benchmark数值，主要是提供最新模型与早期架构及替代微调方法的对比性经验概览，并给出面向用户的实践指南。

Insight: 创新点在于扩展被评估的基础模型家族和微调技术范围，系统刻画Attention、卷积与Mamba模型和VP/LP/FFT/LoRA的交互，并将正则化技术引入视觉提示以提升性能。

Abstract: Adapting foundation models to a new task within or beyond the training domain is the de facto standard for building specialized machine learning models. Recently, Visual Prompting (VP) has emerged as a promising technique for efficient knowledge transfer. As new foundation model families~(like Mamba) get introduced and VP pipelines such as AutoVP reach greater maturity, we find a growing need for a systematic evaluation of current approaches. In this work, we assess the performance of the latest models, comparing them to earlier architectures and alternative fine-tuning methods, to better understand the progress, challenges and opportunities in the field of efficient fine-tuning under resource limitations. Towards this goal, this paper provides a concise empirical overview of the interactions among foundation model families (Attention-, Convolution-, and Mamba-based) and transfer paradigms: VP, Linear Probing (LP), Full Fine-tuning (FFT) and Low Rank Adaptation (LoRA). Our work builds up on previous findings by broadening the selection of evaluated models and techniques while tuning hyperparameters. In the interest of delivering practical guidelines for the user, we also explore application of prevalent regularization techniques to boost performance in the context of VP.

[80] GroundControl: Satellite-Conditioned Multi-View Diffusion for Reconstruction-Free Novel View Synthesis cs.CV | cs.AI | cs.LGPDF

Akshay Krishnan, Matias Turkulainen, Jamie Wynn, Zawar Qureshi, Mohamed Sayed

TL;DR: 论文提出 GroundControl，一种以卫星图像为条件、无需3D重建的多视角扩散模型，用于从少量带位姿街景图像合成户外新视角。模型通过交叉注意力将卫星图像条件注入预训练多视角扩散模型，在生成新视角时兼顾街景细节与全局场景布局，尤其针对目标视角与输入视角重叠很少的外推场景。

Details

Motivation: 现有多视角/视频扩散模型虽能生成高质量新视角，但在输入视角之外进行场景外推时容易幻觉出不合理的空间布局。户外场景可由卫星图像提供全局布局信息，因此论文希望用卫星图像作为额外条件来约束多视角生成。

Result: 在户外数据集上的实验表明，该方法优于仅以卫星图像为条件或仅以街景图像为条件的基线，尤其在目标视角与源视角重叠很少的困难外推情形下表现更好；摘要未给出具体定量指标。

Insight: 创新点在于将卫星图像作为全局场景布局条件，通过交叉注意力注入多视角扩散模型，同时保留预训练扩散先验，并完全避免3D场景重建、仅依赖带位姿的街景与卫星视图。客观上，这种全局俯视条件与局部街景生成的结合有助于提升外推一致性和空间合理性。

Abstract: Multi-view or video diffusion models have recently emerged as state-of-the-art approaches for novel-view synthesis from sparse image collections. While they can produce high quality images, they tend to hallucinate unrealistic spatial layouts when synthesizing views outside the input views (scene extrapolation). We propose to address this challenge for outdoor scenes by using satellite images as an additional information source to condition the multi-view generative model. We introduce a new satellite-conditioned multi-view diffusion model that combines conditioning from a satellite image and street-level input images when generating novel-views. Our cross-attention based conditioning allows us to retain pretrained multi-view diffusion priors to generate seamless, realistic, novel views from a handful of source views, while also adhering to the scene layout as seen from satellite images. Our model does not use 3D scene reconstruction, and only relies on posed street and satellite views. Our experiments on outdoor datasets show that our model is better than both satellite-conditioned or street-view conditioned baselines on novel-view synthesis, especially on challenging cases where target views have very little overlap with source views. Project page: https://nianticspatial.github.io/groundcontrol

[81] ReferStream: Interactive Subject Injection for Streaming Video Generation cs.CVPDF

Shuolin Xu, Guanyuan Li, Boyu Li, Kelu Yao, Xiaogang Xu

TL;DR: 论文提出 ReferStream，一个无需额外训练的流式视频生成框架，用于在生成过程中自然注入新的视觉主体，同时保持已有主体和场景。方法通过 Joint Latent 将参考潜变量放入最新场景潜变量的目标区域，以在上下文中提取参考特征，并用 Cache Restoration 对注入块进行条件化后恢复全局缓存、保留新生成历史。论文还构建了 ReferStream Bench 基准，评估动态参考到达下的主体保真、已有主体保持、文本遵循和计数准确性。

Details

Motivation: 流式视频生成中需要动态加入新主体，同时不破坏已建立的主体和场景一致性。现有独立于当前场景编码参考的方式缺乏整合指导，而将参考特征长期保存在内存中又会干扰已有上下文。

Result: 在 ReferStream Bench 上，ReferStream 在文本遵循、主体相似度和计数准确性上高于对比基线，包括提前给定所有参考的离线方法；评估还涵盖视频质量和新旧主体保持。

Insight: 创新点在于免训练地将当前场景参考构建与瞬态缓存注入结合：Joint Latent 在上下文中提取参考特征，Cache Restoration 在注入后恢复全局缓存且不修改已输出帧；同时提出带计划参考到达和演变指令的新基准。

Abstract: Injecting new subjects naturally during streaming video generation requires integrating newly supplied visual references into the existing video while preserving established subjects and the scene. Encoding references independently of the current scene provides little guidance for this integration, while keeping their features in persistent memory can interfere with established context. We propose ReferStream, a framework that requires no additional training and combines reference construction using the current scene with transient cache injection. Joint Latent places reference latents in target regions of the latest scene latent, allowing reference features to be extracted in context. Cache Restoration uses these features to condition the injection chunk, then restores the overwritten global cache entries while retaining the newly generated history. This procedure incorporates new visual conditions without retraining the generator or revising emitted frames. We further introduce ReferStream Bench, a new benchmark with scheduled reference arrivals, evolving instructions, and explicit annotations of subject presence for InShot and MultiShot generation. Evaluation follows the event schedule and measures both the fidelity of new subjects and the preservation of established subjects, alongside video quality, text adherence, and count accuracy. Experiments on ReferStream Bench show higher scores for text adherence, subject similarity, and count accuracy than the compared baselines, including offline methods given all references in advance. Project page: https://xushuolin.github.io/ReferStream/.

[82] As the Story Unfolds: Watching a Film and Identifying Characters as a Human Does cs.CVPDF

Zhongrui Gui, Junyu Xie, Tengda Han, Weidi Xie, Andrew Zisserman

TL;DR: 论文研究如何像人类观看电影一样，在不依赖外部演员表、演员照片或预建角色库的情况下，直接从电影的视频与对话中识别主要角色并将对话中提到的名字对应到可见角色。作者定义角色命名任务，提供数据集与标注，开发视频-对话模型，并与专有多模态大语言模型比较，还提出随剧情推进逐步构建角色库的渐进式框架。

Details

Motivation: 电影中的角色是随剧情逐步引入的，现有角色识别方法往往依赖外部演员表或预建角色库。论文希望解决仅从电影本身在线推断角色身份的问题，使模型能根据对话与电影语言线索完成角色命名和首次出现识别。

Result: 摘要未给出具体定量数值，主要说明构建了用于评估角色命名和首次角色出现的数据集与标注，开发了视频-对话模型，并与专有多模态大语言模型进行了比较。

Insight: 创新点在于提出角色命名任务，将对话中的人名解析到视频中的可见角色，并利用正反打、自我介绍、直接称呼、第三方指称、注视和时间连续性等电影线索；同时提出渐进式角色库构建框架，支持在线、类人的电影理解。客观来看，其可借鉴之处是无需外部先验、随剧情动态建立角色身份，并联合视频与对话进行角色解析。

Abstract: Films introduce characters incrementally as they progress. We study the problem of identifying principal characters and assigning their names directly from the movie itself, without using external cast lists, actor photographs, or pre-built character banks. We introduce the character naming task: given a name mentioned in dialogue and the associated video context, the model must determine which visible character bears that name. This task is challenging because name resolution often depends on cinematic cues such as shot-reverse-shot structure, self-introduction, direct address, third-party reference, gaze, and temporal continuity. We provide a dataset and annotations for evaluating character naming and first character appearances, develop a video-and-dialogue model for the task, and compare it with proprietary multimodal large language models. We further introduce a progressive framework for building a character bank as the film unfolds. Together, these contributions support online, human-like movie understanding in which character identities are inferred from the film itself.

[83] Zero-Shot Multimodal Event Spotting in Soccer Broadcasts cs.CV | cs.LG | cs.MMPDF

Sadia Kamal, Mahmudul Hasan, Tony Braskich

TL;DR: 论文提出一个面向足球转播的零样本多模态事件检测框架，无需任务特定训练即可对整场比赛中的事件进行检测、分类和时间定位。方法将比赛切分为重叠片段，用紧凑视觉故事板和解说转录共同输入通用多模态模型，并进一步引入带证据解释的精彩度评分用于集锦选择。

Details

Motivation: 现有足球事件检测方法依赖大规模标注数据和足球专用训练，跨数据集泛化与部署成本受限；论文旨在不进行任务特定参数训练的情况下实现整场转播的零样本事件定位。

Result: 在3场人工标注、包含130个事件的比赛上，最佳配置达到72.0% F1@5s，并在跨数据集设置下优于经SoccerNet训练的CALF和NetVLAD++；同时故事板表示相比逐帧输入降低72%推理成本。

Insight: 创新点在于用视觉故事板与对齐解说文本构建通用多模态模型的零样本事件检测流程，并系统分析模型家族、输入表示和提示优化；此外，引入基于证据的自然语言解释进行精彩度排序，增强集锦选择的可解释性。

Abstract: Automatic soccer event spotting supports highlight generation, video indexing, and sports analytics, but existing methods depend on large annotated datasets and soccer-specific training. We present a zero-shot framework that detects, classifies, and temporally localizes events in full-length soccer broadcasts without task-specific parameter training. Each match is segmented into overlapping clips represented by a compact visual storyboard and the aligned commentary transcript, which a general-purpose multimodal model processes jointly. On three manually annotated matches containing 130 events, the best configuration achieves 72.0% F1@5s and outperforms SoccerNet-trained CALF and NetVLAD++ in a cross-dataset setting, while the storyboard representation reduces inference cost by 72% compared with individual frames. We further analyze the effects of model family, input representation, and prompt optimization, and introduce an evidence-grounded excitement-scoring stage that ranks detected events with natural-language explanations for highlight selection.

[84] VideoLVR: Latent Visual Reasoning over Grounded Video Evidence cs.CV | cs.AIPDF

Shilong Li, Xingyuan Bu, Hao Shi, Yun Xing, Weixin Feng

TL;DR: 论文提出VideoLVR，通过从相关视频时刻的局部区域构建潜在视觉证据来支持多模态大模型中的链式推理，并引入视觉token预测目标将视觉证据重建与文本生成统一到同一token预测框架，避免上下文膨胀。

Details

Motivation: 现有基于语言的推理轨迹无法可靠重访视觉证据，导致视觉遗忘；文本坐标缺乏视觉内容，重新插入裁剪区域或帧则大幅扩展多模态上下文。

Result: 在多个视频基准上验证有效，尤其在需要细粒度视觉证据的任务上表现突出，但摘要未提供具体数值或SOTA对比。

Insight: 创新点在于用潜在视觉证据（视觉token）替代文本坐标或原始图像插入，通过视觉token预测目标统一视觉重建与文本生成，并扩展至强化学习联合优化视觉与文本策略。

Abstract: Chain-of-thought reasoning has substantially advanced video understanding in multimodal large language models (MLLMs). However, language-based reasoning traces do not provide a reliable mechanism for models to revisit relevant visual evidence, leading to visual forgetting and impairing their ability to reason over video content. Existing approaches maintain visual grounding during reasoning in two main ways: they either refer to evidence using textual coordinates, which specify symbolic locations rather than the visual content itself, or reinsert cropped regions or frames into the multimodal context, preserving visual details at the cost of substantial context expansion. To address these limitations, we introduce VideoLVR, which constructs latent visual evidence from local regions at relevant video moments, thereby providing visual support for intermediate reasoning steps. To supervise latent visual evidence generation, we introduce a visual token prediction objective that formulates visual token reconstruction as prediction over a vocabulary derived from the input video, thereby unifying visual evidence reconstruction and text generation within the same token prediction framework. This shared prediction framework enables MLLMs to reason over interleaved text tokens and latent visual states while integrating evidence from different moments. We further extend VideoLVR to reinforcement learning by treating sampled visual tokens as actions alongside text tokens and jointly optimizing the policy using final-answer rewards. By efficiently revisiting task-relevant visual information during reasoning, VideoLVR improves video understanding without substantially expanding the multimodal context. Experiments on multiple video benchmarks demonstrate its effectiveness, especially on tasks requiring fine-grained visual evidence.

[85] Perceiving and Predicting Dynamic Sports: Spatiotemporal Intelligence in Vision Models for Egocentric Tennis Video cs.CV | cs.ROPDF

Tudor Braicu, Joseph DelPreto, Jared Boyer, Mónika Farsang, Maximilian Stölzle

TL;DR: 论文构建了一个带注释的第一人称网球视频数据集，并在其上系统评估图像编码器、视频编码器和视觉语言模型在七个时空预测任务上的表现，包括球员/球定位、未来轨迹、击球类型与接触时间预测等。论文还提出条件扩散模型，根据第一人称视频的视觉编码预测实时生成全身网球运动，并随预测更新调整动作。

Details

Motivation: 研究基础视觉模型能否支持敏捷机器人在移动第一人称视角下进行场景理解、交互推理和未来事件预测，弥补该能力评估不足的问题。

Result: 最佳模型实现当前帧定位误差第一人称球员0.14 m、球0.54 m；未来1.1秒预测误差球员0.34 m、球0.91 m；接触时间平均误差0.14 s。辅助深度信息显著提升3D球定位，但更长图像历史未改善球轨迹预测。

Insight: 创新在于构建同步多视角、全局位置、接触时间和击球类型标注的第一人称网球数据集，系统比较多类基础视觉模型在多时空任务上的能力，并提出条件扩散模型实时生成全身运动。分析表明模型更多利用场景上下文而非仅帧间运动，度量深度推断仍是挑战。

Abstract: Agile robots require spatiotemporal intelligence to interpret scenes, reason about interactions, and anticipate future events. Yet it remains unclear how well foundation vision models support these capabilities, particularly from a moving egocentric camera, which is a natural view for robots. To investigate this, we collect an annotated tennis dataset and train a suite of vision models on spatiotemporal tasks. Our dataset contains egocentric video synchronized with five allocentric videos, estimated global positions, labeled contact times, and annotated stroke types. We then evaluate seven prediction tasks from egocentric images: 2D ego player and 3D ball localization, future player and ball trajectories, forehand or backhand stroke anticipation, time to the next ego player ball contact, and 3D ball position at that contact. We compare image encoders, video encoders, and vision-language models with jointly trained prediction heads. The best performing models achieve mean current-frame localization errors of 0.14 m for the ego player and 0.54 m for the ball, future forecasting errors of 0.34 m for the ego player and 0.91 m for the ball over 1.1 s, and a mean contact-time prediction error of 0.14 s. Auxiliary depth information substantially improves 3D ball localization, suggesting that inferring metric depth remains a challenge. Longer image histories do not improve ball forecasting, indicating that representations exploit scene context rather than rely solely on inter-frame motion. Finally, we propose a conditional diffusion model to generate whole-body kinematic tennis motion from vision encoder predictions on egocentric video at real-time rates, adapting movements as predictions are refined. These results move towards a deeper understanding of how foundation vision models can support real-time dynamic robots using egocentric video.

[86] Know, then Draw: A New Learning Paradigm for Visual Generation cs.CVPDF

Yucheng Mao, Chongjian Ge, Zhuowen Tu, Hao Tan, Hanwen Jiang

TL;DR: 论文提出“先知识、后绘制”的视觉生成新范式：先用自回归预训练获取高层语义知识，再用扩散后训练进行高保真像素渲染。作者系统研究了知识获取与渲染的目标、表示和缩放规律，并提出仅用语言监督训练的纯语义tokenizer。基于计算最优缩放律，论文训练了3B参数文本到图像模型，仅用6600 H100 GPU小时即在GenEval和DPG-Bench上取得86和87。

Details

Motivation: 视觉生成同时需要高层语义理解和细粒度像素细节，论文试图将语义知识获取与高保真渲染解耦，解决如何高效学习语义知识并提升最终生成质量的问题。

Result: 实验表明自回归预训练是更优的下游渲染预训练目标，可提升学习效率和最终生成质量，并降低拟合的渐近渲染损失。3B参数文本到图像模型仅用6600 H100 GPU小时训练，在GenEval上达到86、DPG-Bench上达到87。

Insight: 创新点在于提出自回归预训练加扩散后训练的两阶段学习范式，并发现语义质量比AR-only生成质量更关键，进而设计无图像解码器和重建损失的纯语义tokenizer。论文还建立知识获取与渲染的计算最优缩放律，从客观上看其价值在于将语义学习与像素渲染分工并结合缩放规律进行规模化验证。

Abstract: Visual generation requires both high-level semantic knowledge and fine-grained pixel details. We propose a learning paradigm that assigns these roles to autoregressive pretraining for knowledge acquisition and diffusion post-training for high-fidelity rendering. Within this framework, we study the objective, representation, and scaling of knowledge acquisition and rendering. Controlled comparisons identify autoregression as the stronger pretraining objective for downstream rendering, improving both learning efficiency and final generation quality. We further find that AR-only generation quality does not necessarily correlate with final performance; semantic quality matters more. This motivates a purely semantic tokenizer trained with language supervision alone, without an image decoder or reconstruction loss. We establish compute-optimal scaling laws for knowledge acquisition and rendering, and find that autoregressive pretraining lowers the fitted asymptotic rendering loss compared with pixel diffusion models trained from scratch. Guided by the laws, we train a 3B-parameter text-to-image model with only 6,600 H100 GPU-hours, achieving 86 on GenEval and 87 on DPG-Bench and validating the proposed paradigm at scale.

[87] Repurposing Video Diffusion Transformers for Cross-View Temporal Object Correspondence cs.CVPDF

Siyoon Jin, Dahyun Chung, Honggyu An, Sangbeom Lim, Wonjin Nam

TL;DR: 论文提出跨视角时序对象对应任务，要求从查询掩码出发，在跨视角和时间维度上保持对象身份。作者提出XTrack统一框架，复用预训练视频扩散Transformer的语义与时空先验进行RGB到掩码的潜在传输，联合建模跨视角匹配与时序跟踪，并通过注意力对齐、困难负样本条件和局部上下文裁剪增强可靠接地，同时利用伪标签管线将稀疏Ego-Exo4D标注转为密集监督以支持视频级训练。

Details

Motivation: 现有跨视角对象对应方法多依赖逐帧匹配或序列跟踪，在外观变化和遮挡下容易传播对应错误。论文希望同时保持跨视角与跨时间的对象身份一致性，解决跨视角时序对象对应问题。

Result: 摘要未给出具体定量指标或SOTA对比，仅定性表明XTrack作为统一框架能够有效提升时序一致性和身份保持能力，并支持视频级训练。

Insight: 创新点在于将预训练视频DiT重用于跨视角时序对象对应，以RGB到掩码潜在传输联合建模跨视角匹配和时序跟踪；针对DiT跨视角接地不可靠，引入焦点跨视角注意力对齐、困难负样本条件和局部上下文裁剪；同时用伪标签整理管线将Ego-Exo4D稀疏标注转为密集监督。客观看，复用生成式视频模型先验并统一匹配与跟踪，以及用伪标签缓解密集标注缺失，是较有借鉴价值的思路。

Abstract: Cross-view object correspondence identifies the same object across views. However, existing methods often rely on frame-wise matching or sequential tracking, which can propagate correspondence errors under appearance changes or occlusions. Therefore, we introduce cross-view temporal object correspondence, which requires preserving object identity across both views and time from a query mask. To address this, we propose XTrack, a unified framework which repurposes a pretrained video DiT with semantic and spatiotemporal priors for RGB-to-mask latent transport, jointly modeling cross-view matching and temporal tracking. However, naive video DiTs do not guarantee accurate cross-view grounding and often attend to semantically similar yet incorrect objects across views. Thus, we introduce focal cross-view attention alignment, hard negative conditioning, and local context crop strategies to ensure reliable grounding. Finally, beyond these architectural enhancements, we also tackle the lack of dense annotation by introducing a pseudo-label curation pipeline that converts sparse Ego-Exo4D annotations into dense supervision to enable effective video-level training. Our results demonstrate the effectiveness of XTrack as a unified framework, ensuring robust temporal consistency and identity preservation.

[88] Can Discrete Tokens Serve as Semantically Lossless Representations of Visual Signals? cs.CVPDF

Jialv Zou, Xiaotong Li, Zhengrong Yue, Xu Huang, Yaxin Luo

TL;DR: 论文重新审视了离散视觉token因量化误差导致语义损失、连续表示是性能上界的假设，发现量化误差并不能可靠预测下游多模态理解性能，真正关键的是离散索引的有效比特预算。为此提出因子化向量量化（FVQ），用对称并行的码本对视觉特征的互补固定维度视图分别量化，使离散表示在更大比特预算下持续提升，最终在多模态理解上追平连续表示。

Details

Motivation: 离散视觉token是原生多模态建模的天然接口，但量化误差被普遍认为会造成不可避免的语义损失，从而限制了离散表示的性能上限；论文希望验证这一假设，并找到提升离散表示语义保真度的有效途径。

Result: 在工业级多模态数据上训练后，模型在九项指标平均上达到离散表示模型的SOTA，分别超过UniAR和Emu3-Chat 22.8%和59.2%；并且是首个能与Qwen3-VL等强连续表示模型竞争（包括细粒度OCR任务）的离散表示模型。

Insight: 核心创新在于指出量化误差不是可靠指标、有效比特预算才是关键，并揭示现有量化器随码本增加而饱和的问题；FVQ通过对称并行码本量化互补的固定维度视图，缓解饱和并支撑更大比特预算，同时总结了实现该竞争力的关键原则和实用配方。

Abstract: Discrete visual tokens represent vision in a discrete form analogous to language, offering a native interface for multimodal modeling under a shared next-token prediction objective. However, quantization error is widely assumed to cause unavoidable semantic loss, fundamentally limiting the semantic fidelity of discrete visual tokens and positioning continuous visual representations as an intrinsic upper bound. Using multimodal understanding as a proxy for semantic fidelity, we revisit this assumption and find quantization error unreliable, as discrete representations with substantially larger errors can achieve better understanding. Instead, the effective bit budget of discrete indices emerges as a key determinant of downstream performance. However, existing quantizers eventually saturate as additional codebooks contribute progressively less semantic information. As one approach to alleviating this saturation, we introduce Factorized Vector Quantization, which quantizes complementary, fixed-dimensional views of the visual feature with symmetric parallel codebooks. It continues to improve at larger bit budgets, ultimately matching continuous visual representations in multimodal understanding. Trained on industrial-scale multimodal data, our model achieves state-of-the-art performance among discrete-representation models, outperforming UniAR and Emu3-Chat by 22.8% and 59.2%, respectively, on a nine-metric average. To our knowledge, it is also the first discrete-representation model competitive with strong continuous-representation models such as Qwen3-VL, including on fine-grained OCR tasks. We further identify the key principles and a practical recipe enabling this competitiveness. These results and insights support the potential of discrete visual representations to underpin large-scale native multimodal foundation models. Code and models are available at https://github.com/hustvl/FVQ

cs.AI [Back]

[89] The Trinity:Capability-Driven Recursive Self-Improvement for Multimodal Agents cs.AI | cs.CVPDF

Weidong Tang, Ruiheng Wang, Jingxuan Zhang, Hexin Wang, Haoyuan Chen

TL;DR: 论文提出 The Trinity 递归自我改进框架，用于多模态智能体：它通过 Creator、Explorer、Evolver 三个角色，把世界生成与 Harness 演化耦合起来。核心思路是，不同模态和表面形式的任务可以共享相同的底层能力需求；因此，框架根据已验证的执行轨迹把任务树重组为分层的能力树。随后，它在同能力节点上做 Real2Sim2Real 抽象与迁移，沿能力树演化出可复用的能力模块，形成 Harness 树，并用能力感知路由器为任务选择模块配置。

Details

Motivation: 多模态智能体依赖世界与智能体的协同演化，但扩展这一过程有两个瓶颈：构建异构且可验证的环境代价高昂；人工迭代 Harness 的速度也跟不上任务多样性和复杂度的增长。论文希望把任务组织从表面类别转向共享的能力需求，从而实现自动化的递归自我提升。

Result: 在 28 个训练 benchmark 上平均性能提升 34 分；在 18 个未见 benchmark（含两个真实具身任务）上平均提升 26 分，显示出较强的跨任务与跨域泛化迁移能力。

Insight: 创新在于“能力”视角的任务重构：用能力树替代表面任务分类，把同节点异构任务抽象为简化可验证的模拟世界做能力学习，再把结果迁移回原任务（Real2Sim2Real）；并让 Harness 沿能力树选择性继承、累积式演化，最终由能力感知路由器按公开任务信息组合模块。可借鉴之处是将环境生成与智能体外壳演化统一到同一递归闭环，并以能力为中间表示实现跨模态、跨任务的复用。

Abstract: Multimodal intelligence can be viewed as human-guided co-evolution between worlds and agents through interaction. However, scaling this process faces two bottlenecks: constructing heterogeneous, verifiable environments is costly, and manual Harness iteration cannot keep pace with growing task diversity and complexity. A key insight is that tasks with different modalities and surface forms can share the same underlying capability requirements. We therefore introduce The Trinity, a recursive self-improvement framework coupling world generation with Harness evolution through three roles: Creator, Explorer, and Evolver. Drawing on verified execution trajectories, Evolver shifts the basis of task organization from surface categories to shared requirements, reorganizing the original task tree into a hierarchical capability tree. We introduce Capability-Driven Real2Sim2Real: for heterogeneous tasks at the same node, Creator abstracts their common requirements into simplified, verifiable simulated worlds for capability learning, while evaluation on original tasks measures transfer. Along this capability tree, Explorer progressively evolves reusable capability modules through selective inheritance, forming a Harness tree whose validated updates accumulate across iterations within a fixed topology. The resulting Harnesses specialize in different capabilities, motivating a Capability-Aware Harness Router that uses public task information to select compatible module configurations from the tree. Our evolved Harnesses improve mean performance by 34 points across 28 training benchmarks and 26 points across 18 unseen benchmarks, including two real-world embodied tasks.

[90] A Heart Means Recursion: Exploring Visual Meaning and Communication in VLMs cs.AI | cs.CL | cs.CVPDF

Alexa R. Tartaglini, Zhengxuan Wu, Thaddäus Wiedemer, Martin Wattenberg, Been Kim

TL;DR: 论文初步探究视觉语言模型（VLM）是否具备类人的视觉沟通能力。作者发现红心等简单视觉符号能隐式影响模型在无关编程任务上的行为，使其更倾向递归解；模型间视觉通信存在“生成—解释鸿沟”，但在抽象概念和故事通信游戏中也能创造性地用视觉传递意义。总体上，VLM 的视觉意义理解与生成丰富但偏离人类惯例。

Details

Motivation: 论文关注 VLM 是否能像人类一样通过视觉符号进行沟通，以及简单视觉符号会如何影响模型行为和模型间信息传递。其出发点是理解人机视觉通信的潜力与挑战。

Result: 论文报告的是初步定性发现，未在具体 benchmark 上给出定量 SOTA 结果；例如红心符号可隐式引导模型走向递归解，模型常发送直观形状却难以有效引导其他模型，但在抽象概念和故事通信中可创造性传递意义。

Insight: 创新点在于揭示视觉符号对 VLM 行为的隐式 steering 效应，并指出模型显式解释/思维轨迹与实际行为不一致；同时提出模型间视觉通信的生成—解释鸿沟，强调 VLM 视觉意义生产与人类惯例存在 surprising divergence。

Abstract: Do vision-language models (VLMs) share our ability to communicate visually? We share preliminary findings, starting from a surprising observation: simple visual symbols like a red heart can strongly influence model behavior on seemingly unrelated tasks, such as steering models towards recursive solutions on simple coding problems. Notably, this steering effect is largely implicit, as models’ direct interpretations and thinking traces about visual symbols diverge substantially from their observed behavior. In visual communication between models, this leads to a generation–interpretation gap: while models tend to send shape primitives that they (and humans) find intuitive, these visual cues often fail to steer other models. In further communication games involving abstract concepts and stories, models can creatively use visuals to transmit meaning to one another, although some communications remain non-intuitive to humans. These findings reveal that VLMs might be capable of rich, creative, and visually-grounded communication, but their interpretation and production of visual meaning diverges from human conventions in surprising ways. Together this work highlights both the promise and challenges of visual communication between humans and VLMs.

Chaofei Hong, Ziyang Chen, Baochen Wang, Christina Yi Jin, Qing Zhang

TL;DR: 论文提出AC-JEPA（锚定跨模态联合嵌入预测架构），作为与SFT兼容的插件，在跨模态双向预测被遮蔽的中间状态，同时保留因果SFT的token级监督，用于视觉到符号的跨模态翻译。以光学化学结构识别为测试场景，在干净配对和不完美伪标签数据上均优于单独SFT，说明该联合嵌入预测策略能提升跨模态翻译的鲁棒性。

Details

Motivation: 跨模态翻译需要同时捕获共享语义和目标模态的细粒度结构，但真实科学领域中配对数据往往不完美，获取大规模干净配对成本高。直接SFT对目标序列错误敏感，而现有JEPA多关注模态内预测而非跨模态翻译。

Result: 在WildMol基准Simple集上，AC-JEPA用干净配对训练达到76.07%准确率，高于单独SFT的67.40%；使用相同伪标签数据自训练后进一步提升到80.23%，而直接SFT从67.40%下降到63.47%。

Insight: 创新点在于将SFT的token级监督与跨模态联合嵌入预测结合，并利用连续潜在目标从含噪伪标签中学习，避免将伪token提升为硬类别监督。可借鉴之处是“SFT锚定+双向跨模态JEPA”的插件式设计，以及对不完美监督的内在容错机制。

Abstract: Cross-modal translation requires models to capture shared semantics across modalities while preserving the fine-grained structure required by the target modality. Yet in real-world scientific domains, paired data are often imperfect, while obtaining clean pairs at scale is costly. Supervised fine-tuning (SFT) directly optimizes cross-modal translation through categorical token-level supervision, but this objective is sensitive to errors in the target sequences. JEPA offers a complementary representation-space objective, but most existing formulations focus on prediction within a modality rather than translation across modalities. We introduce Anchored Cross-Modal JEPA (AC-JEPA), an SFT-compatible plug-in that predicts masked intermediate states in both cross-modal directions using paired data. The joint-embedding objective targets cross-modal structure predictable from context, while the causal SFT path retains direct token-level supervision for the target sequence. During self-training, AC-JEPA learns from continuous latent targets derived from imperfect pseudo-labeled pairs, offering intrinsic tolerance to local data errors without promoting pseudo-tokens to hard categorical supervision. Using optical chemical structure recognition as a testbed, AC-JEPA trained on clean pairs achieves 76.07% accuracy on the Simple set of the WildMol benchmark, compared with 67.40% for SFT alone. Using the same pseudo-labeled data for both approaches, AC-JEPA-based self-training further raises accuracy to 80.23%, whereas accuracy under direct SFT drops from 67.40% to 63.47%. These results support SFT-anchored joint-embedding prediction as an effective strategy for cross-modal translation under both curated and imperfect supervision.

[92] Learning Where to Trust: Reliability-Guided Expert Routing and Velocity Gating for 3D Deformable Registration cs.AI | cs.CVPDF

Chenyu Guo, Wenjie Dai, Yiqian Chang, Zhiyuan Zhang, Yunfan Wei

TL;DR: 论文提出 ReliMorph，一种面向 3D 可变形图像配准的可靠性路由对称框架。该方法利用局部相关分布评估解剖对应关系的模糊性，并将可靠性线索用于空间专家融合和残差速度门控，在由粗到细的金字塔细化中更新逆一致性反馈以指导更细层的路由与门控。

Details

Motivation: 在均质、弱纹理和严重变形区域，模糊匹配容易导致错误的形变更新；现有学习型配准方法尚未充分显式耦合局部匹配可靠性、空间上下文选择与形变更新强度。

Result: 广泛实验表明，ReliMorph 在解剖重叠指标上一致优于对比方法，并在单模态和多模态配准任务中均表现有效；摘要未给出具体 benchmark 名称和数值。

Insight: 创新点在于首次将空间专家路由与残差速度门控联合条件化于对应关系可靠性，并结合对称框架、模态特定适配和逆一致性反馈实现由粗到细的可靠性引导配准。

Abstract: Deformable image registration relies on accurate anatomical correspondences. However, explicitly coupling local matching reliability with spatial context selection and deformation update strength remains insufficiently explored in learning-based registration. This challenge is particularly pronounced in homogeneous, weakly textured, and severely deformed regions, where ambiguous matches may induce erroneous deformation updates. We propose ReliMorph, a reliability-routed symmetric framework for 3D deformable registration that couples correspondence assessment with spatial context selection and update strength.Local correlation distributions characterize correspondence ambiguity using shared anatomical features with modality-specific adaptation. These reliability cues guide spatial expert fusion and gate residual velocities during coarse-to-fine refinement. Updated inverse-inconsistency feedback guides routing and gating at finer pyramid levels. To our knowledge, ReliMorph is the first 3D deformable registration framework to jointly condition spatial expert routing and residual velocity gating on correspondence reliability. Extensive experiments demonstrate that ReliMorph consistently outperforms competing methods in anatomical overlap, highlighting its effectiveness across unimodal and multimodal registration tasks.

[93] GUI-JEPA: Candidate-Conditioned JEPA Latents as a Fast Action Policy for Mobile GUI Agents cs.AI | cs.CL | cs.CVPDF

Bingchao Wang, Yongcheng Jing, Shunyu Liu, Chengyu Wang, Jun Huang

TL;DR: 论文提出 GUI-JEPA，一种基于联合嵌入预测架构（JEPA）的无解码器移动 GUI 智能体策略。该方法在未标注 GUI 截图上预训练 JEPA 编码器，并通过动作与区域条件的潜在预测器为每个候选动作生成潜在表示，再用轻量头一次性批量评分所有候选动作，无需逐 token 解码。

Details

Motivation: 现有移动 GUI 智能体多依赖 7B 级视觉语言模型逐 token 生成动作，即使采用候选动作评分仍将大模型保留在推理回路中，导致速度慢、内存开销大。论文旨在构建更轻量、快速且无需解码的 GUI 动作策略。

Result: 在 AndroidControl 上，GUI-JEPA 在相同 GPU 上比作者用相同输入微调的 Qwen2.5-VL-7B 智能体快 2.8-4.0 倍，并以 6% 的参数量达到其分布内准确率的 86-92%。

Insight: 创新点在于首个基于 JEPA 的无解码器 GUI 智能体策略，将候选动作评分转化为潜在空间中的批量预测与轻量评分，避免 token 解码和大模型回路；并分析表明评分器依赖该潜在通路，且预测器随机初始化与预训练于屏幕动态效果相当，说明其可直接面向决策训练。

Abstract: Mobile GUI agents are typically fine-tuned 7B-scale vision-language models (VLMs) that generate each action as text. However, these VLMs decode every action token by token, and even agents that score candidate actions keep a large language model or VLM in the loop, making them slow and memory-hungry. To address these issues, we propose GUI-JEPA, to our knowledge the first decoder-free GUI agent policy built on a Joint-Embedding Predictive Architecture (JEPA), which selects actions in the latent space of a JEPA encoder pretrained on unlabelled GUI screenshots. Its action- and region-conditioned latent predictor turns the screen latent into one latent per candidate action, letting lightweight heads score all candidates against the instruction in a single batched pass, without decoding a token. We further show that the scorer relies on this latent pathway and that the predictor is trained for the decision itself: a randomly initialised predictor does just as well as one pretrained on screen dynamics. On AndroidControl, GUI-JEPA is 2.8-4.0x faster on the same GPU than a Qwen2.5-VL-7B agent we fine-tune with the same inputs, while reaching 86-92% of its in-distribution accuracy with 6% of its parameters. Code is available at https://github.com/bcwang-sjtu/GUI-JEPA

[94] AutoVend: Benchmarking LLM Agents on Long-Horizon Autonomous Vending Operations cs.AI | cs.CLPDF

Qiming Shi, Yikun Li, Beidi Luan, Rui Sun, Yan Ma

TL;DR: 论文提出 AutoVend 基准，用于评估 LLM 智能体在长周期自主售货运营中的表现：智能体通过 Web 操作界面管理 20 台自动售货机、5 类站点，持续 180 个模拟日，并在包含共享仓库、采购提前期、付费补货、库存过期和经常性运营成本的仿真环境中做决策。论文评测了 5 个 LLM 智能体，并与基于规则的库存控制基线对比，分析其站点协调、执行和长期经济收益。

Details

Motivation: 现有 LLM 智能体研究多关注单次交易或消费者购物，缺少对 O2O 场景中线上决策能否转化为持续经济回报的评估。论文希望构建贴近真实物理执行与运营成本的长期自主经营基准，检验智能体能否在复杂约束下持续做出有效决策。

Result: 在 AutoVend 上，5 个被评估 LLM 智能体中有 4 个低于规则库存控制基线；若按每美元推理成本带来的回报排序，模型排名与单看回报时不同。运营分析表明单项指标提升不一定带来更强经济增长，库存和执行成本会侵蚀收益；轨迹分析显示智能体在可行动作落地、库存承诺管理和将反馈转化为有效策略修订方面存在困难。

Insight: 创新点在于将 LLM 智能体评测从单次交易扩展到长周期、多站点、O2O 自主经营，并引入数据驱动仿真器建模采购、补货、过期和运营成本等真实约束。其可借鉴之处是同时关注经济回报、推理成本效益和轨迹级诊断，揭示智能体在长期规划、执行落地与反馈学习中的薄弱环节。

Abstract: Large language model (LLM) agents are extending agent-mediated commerce on the Web from shopping for consumers to operating businesses for sellers. In online-to-offline (O2O) commerce, decisions are issued through Web-based platforms but realized through costly physical execution. Evaluating such agents requires going beyond individual transactions to ask whether decisions made online turn into sustained economic returns. We introduce AutoVend, a benchmark in which one LLM agent manages 20 vending machines across five site types through a Web-based operating interface for 180 simulated days. Its data-grounded simulator combines a shared warehouse, procurement lead times, paid replenishment, inventory expiration, and recurring operating costs. We evaluate five LLM agents against a rule-based inventory-control baseline and diagnose their site coordination, execution, and performance over time. Four of five evaluated LLM agents fall short of this baseline, and returns per dollar of inference cost rank models differently from returns alone. Operational analyses show that gains in individual metrics do not consistently yield stronger economic growth, as inventory and execution costs erode these gains. Trajectory analysis further reveals difficulties in grounding feasible actions, managing inventory commitments, and converting feedback into effective policy revisions.

[95] SpectRAG: Weighted Chunk Graphs for Token-Budgeted Context Selection in Multi-Hop Retrieval-Augmented Generation cs.AI | cs.CL | cs.IRPDF

Zafer Sahinoglu, Abdulazeez Mohammed, Buse Yaren Kazangirler, Cahit Berkay Kazangirler

TL;DR: 论文提出 SpectRAG，一种谱图增强的检索框架，用于多跳检索增强生成中的上下文选择。它将混合候选召回、加权 chunk 图构建、图扩展、谱相关性传播和固定 token 预算下的上下文组装结合起来，目标不是只取语义相似的 chunk，而是构造相关、多样、结构连通且足以支撑推理的上下文。

Details

Motivation: 稠密检索虽擅长局部语义匹配，但独立排序容易返回冗余段落、遗漏跨 chunk 依赖，导致多跳推理所需证据不完整。论文希望利用文档结构与图谱信息改善多跳和跨文档场景下的证据覆盖与上下文充分性。

Result: 在 MultiHopRAG、UltraDomain 和长文档问答设置上，SpectRAG 提升了广泛证据召回，并在生成质量上与 dense、hybrid、HyDE、LightRAG、NaiveRAG 基线具有竞争力。结果还表明图谱结构对推理型和跨文档检索更有帮助，同时存在早期精度、计算成本和领域依赖性的权衡。

Insight: 创新点在于用多关系加权 chunk 图编码语义相似度、文档邻接、章节层级、共享实体、引用关系和 LLM 生成效用问题相似度，并结合谱坐标、图邻近重打分与 token 预算上下文选择。可借鉴之处是把检索从 top-k 相似段落选择转向结构化、多样化且推理充分的上下文构造。

Abstract: Retrieval-augmented generation (RAG) systems ground large language models (LLMs) by retrieving external evidence and conditioning generation on the retrieved context. Although dense retrieval is effective for local semantic matching, independently ranked retrieval can return redundant passages, miss cross-chunk dependencies, and provide incomplete evidence for multi-hop reasoning. We propose SpectRAG, a spectral graph-augmented retrieval framework that combines hybrid candidate seeding, weighted chunk graph construction, graph-based expansion, spectral relevance propagation, and token-budgeted context selection. Document chunks are represented by dense semantic embeddings and graph-spectral coordinates derived from a weighted chunk graph whose edges encode semantic similarity, document adjacency, section hierarchy, shared entities, citation links, and similarity between LLM-generated utility questions. At query time, SpectRAG builds a high-recall seed set using exact inner-product search and hybrid relevance signals, expands it over the chunk graph, re-scores candidates using graph proximity and spectral information, and assembles the final context under a fixed token budget. The goal is not merely to retrieve individually similar chunks, but to construct a context that is relevant, diverse, structurally connected, and sufficient for grounded reasoning. Experiments on MultiHopRAG, UltraDomain, and long-document question-answering settings show that SpectRAG improves broad evidence recovery and achieves competitive generation quality against dense, hybrid, HyDE, LightRAG, and NaiveRAG baselines. The results further indicate that graph-spectral structure is particularly useful for inference-oriented and cross-document retrieval, while highlighting trade-offs in early precision, computational cost, and domain-dependent performance.

[96] Guided-RLVR: Guided Exploration in Reinforcement Learning with Verifiable Rewards cs.AI | cs.CLPDF

Bo Yang, Zishuai Zhang, Weihua Li, Qinnan Zhang, Sijia Wen

TL;DR: 论文提出 Guided-RLVR，在带可验证奖励的强化学习（RLVR）中引入轻量级控制器，通过在采样分布中注入语义上有意义的引导 token，鼓励大语言模型从多种推理视角解题，从而扩大探索空间。实验表明该方法在数学推理和代码生成任务上稳定优于现有 RLVR 方法，并产生更多样的推理轨迹。

Details

Motivation: 现有 RLVR 虽能提升 LLM 在数学、编程等任务上的推理能力，但其 rollout 容易坍缩到少量固定推理路径，限制了对推理空间的探索。论文受关键步骤中的 thinking token 是推理路径重要结构元素这一发现启发，试图缓解探索不足问题。

Result: 在数学推理和代码生成基准上，Guided-RLVR 一致优于现有 RLVR 方法，并能生成更多样化的推理轨迹；摘要未给出具体数值或具体 benchmark 名称。

Insight: 创新点在于用轻量级控制器向采样分布注入语义引导 token，以引导式探索替代单纯随机探索，使模型从多推理视角接近问题。可借鉴之处是将关键 thinking token 作为结构化控制信号，用于提升 RLVR 的探索多样性和推理路径覆盖。

Abstract: Reinforcement learning with verifiable rewards (RLVR) has shown strong empirical gains in improving the reasoning ability of large language models (LLMs) on tasks such as mathematics and programming. However, recent studies have shown that RLVR rollouts often collapse onto a narrow set of reasoning trajectories, limiting the explored reasoning space. Motivated by findings that thinking tokens at critical steps serve as critical structural elements in LLM reasoning pathways, we propose Guided-RLVR, a novel approach that augments RLVR with a lightweight controller. The controller strategically injects semantically meaningful steering tokens into the sampling distribution, encouraging the model to approach problems from multiple reasoning perspectives and thereby expanding the exploration space. Experiments on mathematical reasoning and code generation benchmarks show that Guided-RLVR consistently outperforms existing RLVR methods while producing more diverse reasoning trajectories.

[97] Prospector: learning a compact graph traversal policy for scientific reasoning cs.AI | cond-mat.mtrl-sci | cs.CL | cs.LGPDF

Haiqian Yang, Markus J. Buehler

TL;DR: 论文提出 Prospector，一个在领域知识图上逐步遍历推理的紧凑模型。它通过一次调用一个工具来渐进构建上下文，并给出可逐边核验的证据链答案。在半导体制造文献知识图上，Prospector 接近同图同工具的最强现成模型，显著优于闭卷或预选子图方案，并能迁移到多个未见领域知识图。

Details

Motivation: 专业工程领域的科学推理依赖分散且常为专有的文献知识，前沿 LLM 往往给出泛泛答案；而完整知识图又过大，传统 LLM+KG 方法在推理前预选子图，可能过早决定相关性。论文希望让模型在图上按证据逐步遍历推理，并支持私有数据本地部署。

Result: 在半导体制造文献构建的知识图上，Prospector 接近给定相同图和工具的最强现成模型，并远超闭卷或从预选子图作答的现成模型；仅在半导体图训练后可迁移到生物启发材料、PFAS 和材料性质三个未见领域知识图。单 GPU 部署下，其回答更快且成本远低于托管服务，同时保留数据在本地； latent state 还能指示并调制策略行为模式。

Insight: 创新点在于将 LLM+KG 从“推理前预选子图”转为学习紧凑的图遍历策略：模型逐步调用工具检索证据、构建可核验证据链，并实现跨领域知识图迁移和本地低成本部署。客观看，其可借鉴之处包括小模型策略学习、可逐边验证的推理轨迹、潜在状态对行为模式的可解释与可控性。

Abstract: Scientific reasoning in specialized engineering fields depends on knowledge scattered across the literature, often proprietary and absent from frontier LLM training data. In such domains, frontier models tend to give generic answers. Knowledge graphs (KGs) can ground LLM reasoning. Because a graph is too large to provide in full, typical LLM+KG systems preselect a portion, deciding relevance before reasoning begins. Here we train a compact model, Prospector, to reason over a domain knowledge graph by traversing it step by step, seeking evidence in light of what it has already found. Prospector calls a small set of tools one at a time to progressively build its context, and answers with a chain of evidence that can be checked edge by edge against what its tools retrieved from the graph. On a knowledge graph built from the semiconductor-manufacturing literature, Prospector approaches the strongest off-the-shelf models given the same graph and tools, and it far exceeds off-the-shelf models answering closed-book or from a pre-selected portion of the graph. Although trained only on the semiconductor graph, Prospector transfers to unseen knowledge graphs from three other distinct fields: bio-inspired materials, per- and polyfluoroalkyl substances (PFAS) and materials properties. Served on a single GPU, Prospector answers faster and at a fraction of the cost of a hosted service, while keeping proprietary data on the owner’s hardware. We find that the policy’s latent state is indicative of its next move on both seen and unseen graphs, and modulating the latent state can steer the policy from one behavior mode to another. These results suggest a practical LLM + KG paradigm for scientific and engineering reasoning over private data, in which a compact model trained on a graph from published literature transfers its traversal to unseen graphs that stay with their owners.

cs.CR [Back]

[98] Watermark Basins: Black-Box Detection of Distilled LLMs cs.CR | cs.CL | cs.LGPDF

Hongyan Chang, Hady Elsahar, Tom Sander, Pierre Fernandez, Saeed Mahloujifar

TL;DR: 论文研究API蒸馏场景下专有LLM水印的黑盒检测问题，发现带水印教师轨迹微调后会在学生参数空间中留下与任务微调子空间正交的持久“水印盆地”信号，并提出WM-Basin检测器：通过API查询可疑模型，用固定干净探针模型将生成文本映射为梯度表示，投影掉普通任务方向后检验残差是否与密钥诱导的水印盆地对齐。

Details

Motivation: 现有黑盒水印检测依赖token级统计，当蒸馏轨迹采样温度较低时信号可能不可检测；论文旨在提供更可靠、无需访问模型权重或logits的蒸馏证据检测方法。

Result: 在数学推理蒸馏上跨多种教师模型和135M至8.2B学生架构评估，在0.1%严格假阳性率下，WM-Basin在n-gram=1和n-gram=3时分别检测79%和45%的蒸馏模型，显著优于token级基线29%和1.3%，并能经受量化、剪枝和强化学习后训练等修改。

Insight: 创新点在于发现参数空间中的水印盆地这一持久几何信号，并设计WM-Basin利用探针模型梯度表示、任务方向投影和盆地对齐检验实现黑盒检测；客观看，其将检测从token级统计转向参数空间几何信号，提升了低温和后处理场景下的鲁棒性。

Abstract: Proprietary language models and their reasoning traces are vulnerable to API-based distillation, where an adversary trains a student model on the teacher model’s outputs. To obtain evidence of distillation, model providers can watermark the teacher’s outputs, aiming to leave a radioactive signal in the student’s outputs. Existing black-box detectors search for such radioactivity using token-level statistics of the suspect model’s outputs. However, token-level signals are not always detectable, for instance when the distillation traces are sampled at a low temperature. In this work, we discover the watermark basin, a persistent signal that enables more reliable watermark detection: during fine-tuning, watermarked teacher traces embed a distinct, key-specific geometric signal in parameter space that lies orthogonal to the ordinary task-specific fine-tuning subspace. Building on this, we introduce Watermark Basin Alignment (WM-Basin), a black-box watermark detector that queries the suspect model through its API, without access to its weights or logits. WM-Basin maps generated text to gradient representations using a fixed clean probe model, projects out the ordinary fine-tuning task directions, and tests whether the residual gradient aligns with the watermark basin induced by the watermark key. We evaluate WM-Basin extensively across diverse teacher models and student architectures (from 135M to 8.2B parameters) distilled on mathematical reasoning. At a strict 0.1% false-positive rate, WM-Basin detects 79% of distilled models at n-gram=1 and 45% at n-gram=3, substantially improving over the state-of-the-art token-level baseline (29% and 1.3%, respectively). Because the watermark signal resides in a persistent basin, it survives post-distillation modifications including quantization, pruning, and reinforcement-learning post-training.

cs.IR [Back]

[99] QESChunker: A Single Objective Unifies Overlapping and Non-Overlapping Chunking for RAG cs.IR | cs.CL | math.OCPDF

Yifan Zhao, Qianyang Zhao, Xinyu Guan, kai wei, Yuming Deng

TL;DR: 论文提出QESChunker，首个利用问题-证据监督为RAG构建静态分块索引的方法。它将重叠与非重叠分块统一为问题-证据效用最大化问题：非重叠情形可用动态规划精确求解，重叠情形采用贪心近似；在多个数据集上完整证据召回率优于现有分块方法。

Details

Motivation: 现有RAG分块多依赖启发式规则，缺少面向证据检索的监督式索引构建，容易导致检索所需证据覆盖不足。论文希望无需测试查询，仅利用人工或LLM生成的问题-证据对来优化文档特定分块。

Result: 在2个自建和3个公开数据集上，在相同重叠设置和固定混合检索器下，两种变体在几乎所有设置中完整证据召回率超过现有分块方法。理论上，非重叠情形有多项式时间精确解，重叠情形有(1-1/√e)的贪心近似保证。

Insight: 创新点在于将分块重构为监督式组合优化和预算感知的索引构建，用问题-证据对定义统一效用，并分别给出精确动态规划与子模贪心近似；客观看，其价值在于把分块从启发式预处理提升为可优化、可理论分析的检索导向索引构建。

Abstract: Chunking is a fundamental preprocessing step in retrieval-augmented generation (RAG), yet most existing chunkers rely primarily on heuristics. We introduce the Question–Evidence-Supervised Chunker (QESChunker), to our knowledge the first method for constructing static chunk indices from question–evidence supervision. QESChunker constructs document-specific chunk indices by optimizing over human-annotated or LLM-generated question–evidence pairs, without access to held-out test queries. We formulate non-overlapping and overlapping index construction through a unified question–evidence utility-maximization objective, turning chunking into a supervised combinatorial optimization problem. Under the non-overlap constraint, the objective decomposes over chunks and can be optimized exactly in polynomial time by dynamic programming. When overlap is allowed, the objective becomes monotone submodular; exact optimization is NP-hard, but our greedy achieves a $(1-1/\sqrt{e})$-approximation. Across two self-constructed and three public datasets, both variants outperform existing chunking methods in complete-evidence recall in nearly all evaluated settings, under the same overlap regime and a fixed hybrid retriever. These results recast chunking as supervised, budget-aware index construction for evidence retrieval.

cs.LG [Back]

[100] How Much Does the Hint Matter? Scope Conditions for Hindsight Self-Distillation in Financial Reasoning cs.LG | cs.AI | cs.CLPDF

Johnny Yu, Cyril Gorlla

TL;DR: 论文提出 Fin-SD，一种面向多步金融推理的局部化事后自蒸馏方法：学生模型在失败问题上定位首个错误步骤并生成简短批评，截断到错误前形成正确前缀，用冻结的学生副本在“前缀+提示”条件下充当教师，通过反向KL将提示条件教师蒸馏回学生，无需外部模型提供梯度。作者在自建金融推理基准上以 GPT-OSS-20B 验证其有效性，并分析提示内容、提示来源模型强弱及 token 效率等作用条件。

Details

Motivation: 受数据驻留约束的金融机构只能本地部署小规模开源权重模型，这类模型在多步定量推理上落后于前沿系统；传统在线策略蒸馏依赖强外部教师，而基于完整参考解的自蒸馏仍需大量参考信号。论文希望回答：自蒸馏中究竟需要多少“提示/参考”信号，能否仅用学生自身对错误步骤的局部反思实现有效蒸馏。

Result: 在 FinanceReasoning Hard 子集上，Fin-SD 相比基座模型提升 3.03 个百分点；用通用反馈替换具体批评可取得相同增益，且同族约6倍参数模型写提示并不显著优于自写提示；其准确率与基于完整参考解的在线策略自蒸馏相当，同时生成 token 少 9.5%，并在 MMLU-Pro 和 FinTrust 合规审计上无明显退化。

Insight: 论文宣称的创新在于无需强外部教师或完整参考解的局部化事后自蒸馏：利用学生自我错误定位、正确前缀截断和反向KL，将监督集中在出错位置附近。客观看，其价值更在于揭示“监督施加位置/前缀条件”可能比提示内容或教师强度更关键，并为资源受限金融场景提供一种低token、低依赖的自蒸馏方案。

Abstract: Financial institutions operating under data-residency constraints are restricted to open-weight models small enough to locally host, which trail frontier systems on multi-step quantitative reasoning. On-policy distillation narrows such gaps by supervising a student on its own rollouts, but traditional OPD requires a strong external teacher, and on-policy self-distillation (OPSD) conditions on a full reference solution. We ask how much of that signal is necessary. We present Fin-SD, a localized hindsight self-distillation recipe for multi-step reasoning. Given a problem the student fails, the student itself identifies the first incorrect reasoning step and writes a short critique of that step. We truncate the student’s trajectory immediately before the error (aiming to obtain a correct prefix), condition a frozen copy of the student on the same prefix plus the hint, and distill that hint-conditioned teacher into the student over a short rollout under reverse KL. No external model contributes a gradient. We apply Fin-SD to GPT-OSS-20B on a custom multi-step financial reasoning benchmark that requires deriving formula inputs from realistic scenarios via accounting conventions. We find: (i) replacing the critique with a generic-feedback control matches Fin-SD’s entire 3.03 pp accuracy gain over the base model on the FinanceReasoning Hard subset, suggesting the gain comes from where supervision is applied, not hint content; (ii) the model writing the hint does not need to be stronger than the model receiving it: hints from same-family GPT-OSS-120B, at roughly 6x the parameters, yield no materially greater benefits than self-written hints; and (iii) Fin-SD matches on-policy self-distillation from full reference solutions on accuracy while generating 9.5% fewer tokens. With Fin-SD, general reasoning on MMLU-Pro and a fourteen-metric compliance audit on FinTrust show no material regression.

[101] PRISM: Pathology Reliability In Scarce-label Medicine. A Tile-level Benchmark of Pathology Foundation Models cs.LG | cs.CVPDF

Alicem Koyun

TL;DR: 论文提出 PRISM，一个面向病理基础模型的 tile 级可靠性基准，联合评估模型选择、标签比例、校准和跨机构分布外迁移。它在 8 个病理基础模型、6 个 tile 数据集、1% 到 100% 标签比例和 3 个随机种子上完成 864 次线性探针实验，并扩展到 Camelyon17 的五医院 20 个定向迁移对。研究发现低标签和迁移场景下校准行为显著退化，温度缩放并不总能修复，甚至可能有害，并发布可 pip 安装的基准包及全部 288 个实验单元参考结果。

Details

Motivation: 现有病理基础模型评测多在完整标签和分布内测试集上比较分类性能，较少同时考察低标签预算下的校准、跨机构分布偏移鲁棒性及二者交互。论文希望建立统一基准，揭示模型在稀缺标签和真实临床迁移中的可靠性风险。

Result: 实验覆盖 8 个模型和 6 个数据集，并包含 Camelyon17 的 20 个定向迁移对。结果显示：全监督下 AUROC 排名可较稳定预测校准排名（Spearman rho=0.52），但 1% 标签时关系变为数据集相关（rho=-0.24）；困难数据集在低标签比例下线性探针会坍缩为单一预测类；当迁移 AUROC 低于约 0.85 时，增加源域标签在 86–88% 组合中反而恶化目标域 ECE；温度缩放仍在 49/288 个单元中失效。

Insight: 创新点在于将模型选择、标签比例、校准和分布外迁移纳入同一 tile 级基准，系统性揭示低标签与迁移场景下的校准失效和线性探针坍缩现象。其可借鉴之处是强调不能仅用全监督 AUROC 评价病理基础模型，并提示温度缩放和增加源域标签在迁移失败时可能带来负收益。

Abstract: Pathology foundation models are typically benchmarked on classification performance at full label availability and on in-distribution test sets, while calibration, behavior at low label budgets, and robustness under cross-institutional shift are seldom evaluated jointly. We introduce PRISM, a tile-level benchmark that measures the reliability of pathology foundation models along four axes: model choice, label fraction, calibration, and out-of-distribution transfer. PRISM evaluates eight foundation models (CLIP, PLIP, CONCH, UNI, Virchow2, GigaPath, H-Optimus-0, Midnight) on six tile-level datasets at six label fractions (1% to 100%) with three seeds (864 linear probe runs), extended with a five-hospital Camelyon17 study of 20 directed transfer pairs. Four phenomena emerge. (i) Under full supervision, AUROC rank predicts post-temperature calibration rank consistently across datasets (Spearman rho = 0.52, I^2 = 0%), whereas at 1% labels the relationship becomes dataset-specific (rho = -0.24, I^2 = 67%). (ii) On harder datasets the linear probe collapses to a single predicted class below a dataset-specific label fraction. (iii) Whether additional source-domain labels help or harm target calibration tracks how well the probe transfers: below a transfer AUROC of roughly 0.85, more labels worsen target ECE in 86-88% of combinations. (iv) Temperature scaling fails to reduce ECE in 49 of 288 cells, miscalibration is predominantly under-confidence, and a source-fitted temperature is harmful under failing transfer. PRISM is released as a pip-installable package with reference results for all 288 cells. Code and data: https://github.com/alicemkyn/prism-benchmark

[102] Select-TTT: Query-Conditioned Selective Fast Weight Updates for In-Place Test-Time Training cs.LG | cs.AI | cs.CLPDF

Yue Wu, Yang Zhang, Weinan Gan, Chongming Gao, Xinsen Guo

TL;DR: 论文研究面向大语言模型的原地测试时训练（In-Place TTT），指出其统一写入所有上下文块会导致无关内容稀释查询相关信息，从而在真实网页长上下文问答中收益消失。为此提出 Select-TTT：根据每个上下文块与用户查询的相关性，调节其写入快速权重的强度。该方法无需监督和新增参数，并保持原有分块、上下文并行计算。

Details

Motivation: 现有 In-Place TTT 虽能把上下文即时写入快速权重，但对所有块采用相同写入强度，使无关内容污染快速权重记忆，导致在真实长上下文问答场景中效果有限。论文希望让快速权重更新更有选择性地关注与查询相关的内容。

Result: Select-TTT 在两个长上下文问答基准和两个骨干模型上均稳定提升长上下文问答表现，说明查询条件化的快速权重写入对实际测试时训练有效。摘要未给出具体数值。

Insight: 创新点在于用冻结输出头中查询 token 的概率质量作为无需监督的相关性信号，并通过温度缩放 softmax 转成学习率乘子，在不新增参数、不破坏分块并行计算的前提下实现查询条件化快速权重写入；该思路也可能推广到多用户查询、上下文演化的长程智能体任务。

Abstract: Test-time training (TTT) enables a deployed language model to internalize incoming context on-the-fly into a small set of fast-weight parameters, rather than merely treating it as transient context for attention. In-Place TTT recently made this idea practical for large language models by repurposing the down-projection of standard MLP blocks as fast weights. However, we find that its benefit largely disappears on realistic web long-context question answering. We attribute this to In-Place TTT’s uniform write schedule: every chunk is written with equal strength, so irrelevant content dilutes query-relevant information in fast-weight memory. Building on this insight, we propose Select-TTT, a model-agnostic, query-conditioned rule for In-Place TTT that scales each chunk’s write strength by its relevance to the user query. This relevance signal requires no supervision and no new parameters: it is read directly from the model’s frozen output head, as the probability mass that a chunk’s hidden states assign to the query tokens. A temperature-scaled softmax turns these scores into learning-rate multipliers, leaving the chunk-wise, context-parallel computation of In-Place TTT intact. Select-TTT consistently improves long-context question answering on two benchmarks and two backbones. These findings highlight the value of query-conditioned fast weight writes for practical test-time training on long-context question answering, which is also potentially promising for long-horizon agent tasks involving multiple user queries over evolving contexts.

[103] HiLo-RD: Cross-Budget Self-Distillation for Masked Diffusion Language Models cs.LG | cs.CLPDF

Xiang Li, Chengting Yu, Jungang Li, Yilun Zheng, Junzhuo Li

TL;DR: 论文提出HiLo-RD，一种面向掩码扩散语言模型（MDLM）的跨预算自蒸馏方法，用于在有限生成预算下改善解码。它将同一模型在大窗口下的教师状态与小窗口裁剪视图配对，并在已提交位置对齐隐藏表示，从而同时指导 token 预测和提交顺序。方法无需外部教师，在数学基准上提升低预算准确率并缩短回答，且可迁移到代码生成。

Details

Motivation: MDLM 需要同时决定生成什么内容以及在何处提交 token；在有限预算下容易出现“预算诱导的过早提交”，即在中间推理尚未完成时就固定窗口边界附近的答案片段，进而污染后续推理。现有后训练方法无法直接迁移同一模型在大窗口下更优的推理轨迹。

Result: 在数学基准上，HiLo-RD 提升了低预算准确率并缩短响应长度；在 GSM8K 上准确率提升 9.4 个百分点，平均可读长度下降约 25%。这些收益还可扩展到未进行代码域训练的代码生成任务。

Insight: 创新点在于利用同一模型的大窗口状态作为教师，对小窗口视图进行表示级自蒸馏，并在“已提交位置”对齐隐藏表示，以监督 token 预测与提交顺序。客观上，其价值在于无需外部教师即可把大预算推理轨迹迁移到小预算解码，并通过注意力诊断和固定预算顺序干预验证了机制。

Abstract: We propose HiLo-RD (High-to-Low Budget Representation Distillation) to improve masked diffusion language models (MDLMs) under a limited generation budget. Unlike autoregressive models, MDLMs choose both what to generate and where to commit it. This creates two coupled challenges: predicting useful content before reasoning is complete, and deciding when to fix answer tokens that shape later predictions. We identify budget-induced premature commitment: even when a correct solution fits, the model can fix answer fragments near the window boundary before resolving intermediate steps. These fragments become context for later predictions and redirect subsequent reasoning. Existing post-training methods do not directly transfer the stronger trajectories that the same model follows with a larger window. HiLo-RD addresses this gap by (1) pairing large-window teacher states with cropped small-window views and (2) matching their hidden representations at committed positions. This supervision guides both token prediction and commitment order without an external teacher. Across mathematics benchmarks, HiLo-RD improves low-budget accuracy and shortens responses; these gains extend to code generation without code-domain training. On GSM8K, accuracy rises by 9.4 percentage points as mean readable length falls by about 25%. Attention diagnostics and fixed-budget order interventions support the proposed mechanism. Together, these results show that large-window reasoning can improve small-window decoding.

[104] V-LUMEN: Visual Lookup Memory for Embedding Scaling in Vision-Language Models cs.LG | cs.AI | cs.CL | cs.CVPDF

Miso Choi, Daekeun Kim, Eunji Kim, Jungbeom Lee

TL;DR: 论文将语言模型中基于查表的嵌入级扩展（embedding-level scaling）思路推广到视觉语言模型（VLM），提出 V-LUMEN 视觉查表记忆。它用冻结码本离散化视觉特征，通过空间聚合哈希构建具有空间感知的记忆键，并用文本条件相关性信号进行调制；所有寻址均为非参数化，检索到的嵌入经门控残差通路融合进骨干网络。在五个推理密集型基准上显著提升冻结的 Qwen3-VL-2B 骨干，并可零样本迁移到三个图像不相交的基准。

Details

Motivation: 嵌入级扩展能通过查表记忆而非加深加宽模型来扩充容量，但视觉特征连续、具有空间结构且依赖查询，不像离散文本 token 天然支持 n-gram 查表，因此将该思路迁移到 VLM 存在挑战。

Result: 在五个推理密集型基准上显著提升冻结的 Qwen3-VL-2B 骨干，并零样本迁移到三个训练图像不相交的基准；在匹配参数预算下，将 LoRA 或 MoE 适配器的部分容量移入 V-LUMEN 后，在两个骨干上均保持精度。查表开销与表大小无关，且在 prefill 阶段仅执行一次。

Insight: 创新点在于冻结码本离散化、空间聚合哈希构建空间感知键、文本条件相关性调制以及门控残差融合；客观来看，其价值在于把外部视觉记忆作为一种独立于参数适配的容量扩展轴，且容量随存储而非每 token 计算增长。

Abstract: Embedding-level scaling expands the capacity of language models through lookup-based memory rather than by increasing model depth, width, or dense computation. Extending this idea to vision-language models (VLMs) is challenging because visual features are continuous, spatially structured, and query-dependent, unlike discrete text tokens that naturally support n-gram lookup. We introduce V-LUMEN, a visual lookup memory for VLMs. V-LUMEN discretizes visual features with a frozen codebook, builds spatially aware memory keys through spatial aggregation-based hashing, and modulates them with text-conditioned relevance signals; all addressing is non-parametric, and the retrieved embeddings are fused into the backbone through a gated residual pathway. On five reasoning-intensive benchmarks, V-LUMEN substantially improves a frozen Qwen3-VL-2B backbone and transfers zero-shot to three benchmarks whose images are disjoint from the training data. At a matched parameter budget, moving part of a LoRA or MoE adapter’s capacity into V-LUMEN preserves accuracy on two backbones, indicating that external visual memory is an independent capacity axis rather than a stronger adaptation method. Because lookup cost is constant in table size and incurred only once at prefill, this capacity grows in storage rather than in per-token compute. Code is available at https://github.com/miso-choi/V-Lumen.

[105] Exploring the Limits of the Delta Learning Hypothesis: Single-Pair Generator Biases in Direct Preference Optimization cs.LG | cs.AI | cs.CLPDF

Giuseppe Curci, Stefano Simonazzi, Andrea Molinari, Andrea Zugarini

TL;DR: 论文研究DPO中“delta学习假设”的局限，即是否可用单一模型对（将较大模型回答视为偏好）替代多模型池加外部LLM评判构造偏好数据。作者从理论和实验上表明，固定模型对会引入系统性的、依赖生成器的偏好偏差，使DPO学到与回答质量无关的虚假信号。因此单对生成器数据训练的DPO持续弱于多模型偏好数据，且增加生成器数量可在无外部评判时部分缓解该问题。

Details

Motivation: 验证delta学习假设能否以单模型对低成本构造高质量偏好数据，并探究其在跨模型家族和规模时的适用边界。

Result: 单对生成器偏好数据训练的DPO一致劣于多模型偏好数据，即使较大生成器回答质量更高；摘要未提及具体benchmark和数值，但指出该差距与对生成器偏差的reward hacking训练动态一致，增加生成器数量可部分缓解。

Insight: 创新点在于揭示固定模型对带来的生成器依赖偏差及其对DPO训练的影响，并强调生成器多样性是偏好数据构建的关键因素。

Abstract: Direct Preference Optimization (DPO) typically relies on preference pairs generated from diverse model pools and ranked by an external LLM judge. The recently proposed delta-learning hypothesis suggests that comparable supervision can instead be obtained from a single pair of models by treating the larger model’s response as preferred. We investigate the limits of this approach across model families and scales. We show, both theoretically and empirically, that fixed model pairs introduce systematic, generator-dependent biases that are predictive of the preference label independent of response quality. Consequently, DPO trained on single-pair data consistently underperforms multi-model preference data, despite the larger generator producing higher-quality responses, and we trace this gap to training dynamics consistent with reward hacking on these biases. We further show that simply increasing the number of generators, even without an external judge, is enough to partially mitigate this effect. Our results highlight generator diversity as a critical factor in preference-data construction.

[106] Self-Regularized Flow Language Models cs.LG | cs.CLPDF

Theodoros Kouzelis, Spyros Gidaris, Nikos Komodakis

TL;DR: 论文提出SERF，一种面向连续扩散/流式语言模型的自监督训练框架，利用去噪轨迹本身作为学习信号：让EMA教师接收同一插值路径上更干净的输入，对学生进行噪声到干净的自蒸馏。SERF有两种变体：SERF-L对齐中间表征，SERF-P对齐输出token分布，且不引入外部预训练模型、不改变推理流程。

Details

Motivation: 连续流式语言模型训练目标通常只依赖干净目标token的监督，去噪轨迹这一丰富学习信号未被充分利用，导致训练效率和生成质量受限。

Result: 在one-hot FLM、latent ELF、hyper-spherical S-FLM三种范式上均取得显著提升：OpenWebText上训练实际耗时加速最高7.7倍，MAUVE提升24.3%，GSM8K准确率最高提升28.8%，采样步数减少最多64倍时仍匹配基线性能。

Insight: 创新点在于将“噪声到干净”的自蒸馏引入连续扩散语言模型，利用EMA教师沿同一插值路径提供更干净监督，并以表征级（SERF-L）和分布级（SERF-P）两种对齐方式增强标准训练目标；其简洁、无需外部模型、不改变推理的特点使其易于推广到多种连续语言模型范式。

Abstract: We introduce SERF (Self-Regularized Flow Language Models), a simple self-supervised framework for improving continuous diffusion language models by exploiting the denoising trajectory itself. While continuous flow-based language models offer a promising alternative to autoregressive generation, their training objectives typically rely on supervision from clean target tokens, leaving the denoising trajectory underexploited as a learning signal. SERF addresses this limitation through noisy-to-clean self-distillation: an exponential moving average (EMA) teacher receives a cleaner version of the student’s noisy input along the same interpolation path and provides additional supervision. We investigate two variants: SERF-L, which aligns the student’s intermediate representations with deeper teacher representations, and SERF-P, which aligns their output token distributions, either at the student’s output or at an early student layer. All variants augment the standard training objective without requiring external pretrained models or modifying the inference procedure. Experiments across three continuous language-model paradigms (one-hot FLM, latent ELF, and hyper-spherical S-FLM) demonstrate substantial improvements in training efficiency, generation quality, and few-step generation. In particular, SERF achieves up to 7.7x wall-clock training acceleration on OpenWebText, improves MAUVE by 24.3% and GSM8K accuracy by up to 28.8%, and matches baseline performance with up to 64x fewer sampling steps. These results establish noisy-to-clean self-distillation as an effective approach to improving continuous diffusion language models.

[107] EmbeddingGemma 2: Unified and Efficient Multimodal Representations cs.LG | cs.CL | cs.CV | cs.SDPDF

Henrique Schechter Vera, Sahil Dua, Jovan Andonov, Sara Smoot, Sebastian Russo

TL;DR: 论文提出 EmbeddingGemma 2，一个开源多模态嵌入模型，可在统一空间中原生表示文本、图像、音频、视频及其组合。模型通过简化的逐层嵌入和多阶段课程学习（模态与目标逐步推进）提升单模态、跨模态和混合模态的嵌入质量，并兼顾资源受限场景下的部署灵活性。

Details

Motivation: 解决多模态嵌入中不同模态难以统一表示、跨模态与混合模态检索/匹配质量不足，以及大模型在资源受限环境中部署成本高的问题。

Result: 在覆盖所有模态的综合嵌入基准上达到新的 Pareto 前沿，优于专用模型以及参数量大 2 到 12 倍的模型；同时支持选择性加载模态编码器、可调视觉 token 预算、可截断嵌入维度和量化权重。

Insight: 创新点在于统一多模态嵌入空间、简化的逐层嵌入设计，以及按模态和目标推进的多阶段课程训练；客观上看，其价值还包括在保持性能的同时提供多种推理效率调节手段，增强实际部署可用性。

Abstract: We present EmbeddingGemma 2, an open multimodal embedding model which natively represents text, images, audio, video, and combinations thereof in a unified space. Our approach introduces improvements such as simplified per-layer embeddings and a multi-stage curriculum with modality and objective progression which ensure strong quality across unimodal, cross-modal, and mixed-modal evaluations. Notably, in comprehensive embedding benchmarks across all modalities, EmbeddingGemma 2 establishes a new Pareto frontier, outperforming both specialist models and models two to twelve times larger. Furthermore, our design offers flexibility for resource-constrained environments, supporting selective modality encoder loading, adjustable vision token budgets, truncatable embedding dimensions, and quantized weights. We present ablations analyzing key decisions and release EmbeddingGemma 2 to promote further research.

[108] LinMix: Data Mixing as Linear Domain Attribution cs.LG | cs.CL | stat.MLPDF

Hubert Baniecki, Przemyslaw Biecek, Martin Pawelczyk

TL;DR: 论文把预训练数据的领域配比（data mixing）问题重新表述为领域归因问题，并提出 LinMix：在带上限的单纯形（capped simplex）的顶点处采样代理数据混合，也就是让每个领域要么取到上限，要么完全不出现；基于这些采样，拟合一个可解释的线性模型，其系数就是各领域的归因值，并据此直接解析地给出最优配比。作者在 1B 参数的 OLMo2 和 4B 参数的 Qwen3.5 上做了预训练实验，结果显示该方法在推理、代码和数学基准上均优于基于 Dirichlet 采样的数据混合方法。

Details

Motivation: 现有数据混合方法通常先按 Dirichlet 分布随机采样出若干混合方案，用它训练大量小代理模型，再用回归模型从混合方案预测模型性能。作者指出，这本质上是一个领域归因问题；而 Dirichlet 采样作为回归设计并不理想：它需要调参，并且在处理数百个领域时无法支撑线性拟合。

Result: 在 1B 参数的 OLMo2 和 4B 参数的 Qwen3.5 预训练实验中，LinMix 在推理、代码和数学基准上全面超过基于 Dirichlet 的方法，其中 MBPP 相对提升 46%，GSM8K 提升 14%；当领域数为 288 个细粒度领域时，只有顶点采样能支持线性拟合，并且 LinMix 的平均基准准确率比最好的非线性基线高 11%。

Insight: 论文宣称的创新点包括：把数据混合建模为线性领域归因；用 capped simplex 上的顶点采样来解耦领域效应，并且无需超参数；理论上证明顶点采样最优，而 Dirichlet 采样在混合集中在先验配比附近时会放大大归因误差；同时，最优配比只保留正向归因的领域，并且无需搜索候选混合方案。从客观角度看，论文借鉴经典最优实验设计思想，把“训练代理模型+回归”的流程简化为可解释的线性模型，并给出理论保证，思路简洁而扎实。

Abstract: Language model performance depends strongly on the mixture of domains in the pretraining data corpus. Popular data mixing methods select these domain weights by training many small proxy models on random mixes and then fitting a regression model to predict performance from the mix. We recognize this practice as a domain attribution problem, where the regression measures each domain’s contribution to the target metric. Through this lens, the standard Dirichlet sampling of proxy data mixes is a poor regression design that requires tuning and cannot support a linear fit for hundreds of domains. Building on classical optimal experimental design, we introduce LinMix, which samples mixes at the vertices of the capped simplex so that each domain is at its cap or absent. This hyperparameter-free vertex sampling decorrelates domain effects, enabling an interpretable linear model whose coefficients are the domain attributions. The optimal mix given by LinMix provably keeps only domains with positive attribution values and requires no search over candidate mixes. We further establish the optimality of vertex sampling, and prove that Dirichlet sampling inflates the attribution error when its mixes concentrate around the prior data mix. Pretraining experiments with 1B-parameter OLMo2 and 4B Qwen3.5 language models show that LinMix outperforms Dirichlet-based data mixing across reasoning, code, and math benchmarks, with relative gains of 46% on MBPP and 14% on GSM8K. At 288 fine-grained domains, only vertex sampling supports a linear fit, and LinMix improves benchmark accuracy by 11% on average over the best nonlinear baseline.

[109] When Visual Evidence Is Insufficient: How Instructions, Post-Training, and Internal Activations Shape Refusal in Multimodal Language Models cs.LG | cs.CVPDF

Ananth K Suresh, Aveen Dayal, Gouthaman KV, Aditya Pola, Vineeth N Balasubramanian

TL;DR: 论文研究多模态大模型在图像证据不足时是否会拒答，并系统分析推理时指令、后训练和内部激活干预三类控制来源对拒答行为的影响。作者在三个MLLM和MoHoBench基准上发现，拒答对指令语义与位置高度敏感，后训练会显著提高拒答但也可能混合拒答与作答，激活干预影响较小但可迁移。

Details

Motivation: 可靠部署要求MLLM识别图像证据不足并拒绝无依据回答，但模型拒答不一定代表其真正识别到证据缺失，也可能受安全约束等无关因素影响。论文希望厘清外部指令、后训练和内部表征如何共同塑造拒答行为。

Result: 在三个MLLM和MoHoBench上，拒答对指令含义和放置位置高度敏感，且样本级变化会被总体拒答率掩盖；证据感知和通用拒答指令比不匹配的安全指令更有效。后训练带来更大的拒答提升，但可能产生拒答语言与尝试作答混合的响应；激活干预变化较小但跨提示条件可迁移，且在可视觉回答样本上未出现明显无差别拒答。

Insight: 创新点在于从推理时指令、后训练和内部激活三个层面系统操控拒答行为，并利用自然作答/拒答样本构造拒答相关激活方向进行干预。客观来看，论文强调需区分“证据识别”与模型学到的“拒答倾向”，为多模态模型拒答机制分析提供了可借鉴的实验框架。

Abstract: For reliable deployment, multimodal large language models (MLLMs) should recognize when an image provides insufficient evidence to answer a question and withhold unsupported answers. Yet refusal alone is not necessarily evidence of such recognition: models may refuse because of unrelated safety constraints or without identifying the missing evidence. We study how refusal on visually unanswerable questions is shaped by three sources of control: inference-time instructions, post-training, and internal activation interventions. Across three MLLMs on the MoHoBench benchmark, we systematically manipulate the meaning and placement of refusal instructions, compare supervised and preference-based post-training, and intervene on a refusal-related activation direction derived from naturally answered and refused examples. We find that refusal is highly sensitive to both the meaning and placement of instructions, with model-specific changes at the sample level that are obscured by aggregate refusal rates. Evidence-aware and general refusal instructions have stronger effects than an intentionally mismatched safety instruction. Post-training produces larger increases in refusal, but can also induce responses that combine refusal language with attempted answers. In contrast, activation intervention produces smaller yet transferable changes across prompt conditions, without evidence of indiscriminate refusal on tested visually answerable samples. These findings show that refusal is jointly shaped by external instructions, post-training, and internal representations, highlighting the need to disentangle evidence recognition from a model’s learned propensity to refuse.

[110] EnvSched: Environment-Aware Scheduling for Rollout-Efficient Agent Reinforcement Learning cs.LG | cs.CLPDF

Sihan Yang, Weichen Bai, Ruichuan An, Yiming Cheng, Zhiyuan Feng

TL;DR: 论文提出EnvSched，一种环境感知调度器，旨在提高语言智能体强化学习中的rollout效率。它通过双关系图建模环境-任务对，利用历史rollout统计预测奖励方差，并调度批次以优先选择高信息量和观察不足的环境与任务，从而在有限rollout预算下加速RL训练。

Details

Motivation: 随着语言智能体可执行环境规模扩大，生成和验证rollout成本高昂。论文关注在有限rollout预算下，如何通过环境级别的数据管理（组织、评估、调度任务实例及其环境）来提升agentic RL的效率。

Result: 在Qwen3-8B的RL训练中，EnvSched在BFCL-v3、tau-Bench和ACEBench-Agent上40步后平均得分48.81，超过随机任务选择120步的最高分；相同训练预算下还提升了不同模型规模的下游性能。

Insight: 论文宣称的创新包括：在线调度器利用双关系图传播历史rollout统计以预测奖励方差，结合利用与覆盖策略，并通过环境配额和相似性惩罚保证批次多样性与信息量，无需额外试点rollout。客观分析，其创新在于将环境结构信息（工具、状态、验证器）纳入调度，通过图建模跨环境传播知识，实现更高效的任务选择。

Abstract: As executable environments for language agents scale, generating and verifying their rollouts becomes increasingly expensive. Under finite rollout budgets, this raises an important question: can data management at the environment level make agentic reinforcement learning (RL) more efficient by organizing, evaluating, and scheduling task instances together with the environments that determine their tools, states, transitions, and verification? To address this question, we introduce EnvSched (Environment-aware Scheduling), an online scheduler that learns from rollout history. EnvSched assigns scheduling value through two complementary modeling choices: exploitation prioritizes tasks likely to provide informative reward differences, while coverage spreads each batch across underobserved environments and interaction patterns. EnvSched organizes environment–task pairs as nodes in a dual-relation graph, with within-environment edges capturing task semantics and shared tool, state, and verifier structure, and cross-environment edges capturing abstract execution and reward topology. By propagating historical rollout statistics along both edge types, it predicts within-task reward variance. These predictions guide a scheduler that allocates each batch to tasks with high predicted reward variance and underobserved environments and tasks. Environment quotas and a similarity penalty further limit concentration and redundancy, yielding informative and diverse batches without additional pilot rollouts. In RL on Qwen3-8B, EnvSched achieves an average score of 48.81 across BFCL-v3, tau-Bench, and ACEBench-Agent after 40 training steps, exceeding the highest score attained by random task selection after 120 steps. It also improves downstream performance across model sizes at the same training budget.

cs.NI [Back]

[111] Real-Time Smart Retail Analytics through Fog Computing using the RetailLightNet Lightweight Model cs.NI | cs.CVPDF

Mayank Atreya, Navin Chhibber, Shilpa Ajay, R Loshini, K Anitha

TL;DR: 论文提出面向智能零售实时视频分析的雾计算框架FRLN，并设计超轻量模型RetailLightNet以适配NVIDIA Jetson Nano等资源受限边缘设备。方法利用SKU110K和RPC零售数据集训练，通过归一化预处理和运动门控帧选择减少冗余帧，并采用深度可分离卷积提取特征，配合存在分类与密度回归两个预测头支持货架监控和队列评估。

Details

Motivation: 云端视频分析存在通信高延迟、原始视频传输隐私风险以及重型视觉模型对有限网络和硬件资源要求高的问题，难以满足零售场景中及时决策的需求。论文旨在面向边缘/雾计算环境构建轻量、实时的智能零售分析方案。

Result: 所提FRLN框架在零售相关数据上取得精确率0.988、召回率0.976、F1分数0.976，并宣称相比现有卷积神经网络模型表现更好。

Insight: 创新点在于将雾计算部署、超轻量RetailLightNet、运动门控帧选择和双预测头结合，形成面向Jetson Nano等边缘硬件的系统化实时零售分析方案；客观看，其可借鉴之处是同时从模型轻量化、输入帧筛选和多任务输出三个层面优化边缘视频分析效率。

Abstract: Smart retail systems now depend on the application of real-time video analytics, which can track customer behavior, manage shelf inventory, and enhance in-store processes. Currently cloud-based applications of video analytics face multiple challenges such as high latency of the communication itself, privacy concerns surrounding transmitting raw video, and the computational demands of the heavy vision models used on the often limited network infrastructure. These challenges lead to delays in decision making where prompt action is crucial. To address these short comings, this research proposes Fog-based analytics withRetailLightNet (F RLN), an ultra lightweight deep learning model which is specifically suited for the edge, resource constrained hardware capabilities of devices like the NVIDIA Jetson Nano. The approach utilizes publicly available retail datasets, SKU110K, and RPC, which contains an assortment of product, shelf, and customer interaction scenarios for training.The processes include preprocessing images through normalization and employing motion-gated frame selection to improve efficiency by omitting redundant frames to ease the workload on the fog processing. Internal within Retail LightNet, image feature extraction is realized through depth wise separable convolutions which help support spatial and channel-wise learning without heavy computational costs. The model employs two prediction heads presence classification, and density regression, supporting shelf monitoring and queue assessment. The proposed F RLN framework attained better results in terms of precision (0.988), recall (0.976), and F1-score(0.976) when compared with existing convolutional neural network. Keywords: fog computing, motion-gated frame selection,NVIDIA jetson, retail lightnet, smart retail.

cs.RO [Back]

[112] Closed-Loop Aware Planning via Imitation-Free Speed Scoring for End-to-End Autonomous Driving cs.RO | cs.CV | cs.LGPDF

Emad Aghajanzadeh, Thomas Schwarzfischer, Marc Kaufeld, Moritz Thoma, Ryan Pégoud

TL;DR: 论文提出 CLAP-AD，一种可插入现有端到端自动驾驶模型的规划头：路径通过模仿学习获得，候选速度则由受物理安全约束监督的学习函数进行多准则评分，从而避免对速度进行开环模仿。该方法无需速度条件训练即可在测试时显式控制速度选择，并在多个闭环与真实数据基准上提升速度遵循性和驾驶性能。

Details

Motivation: 现有端到端自动驾驶主要依赖开环模仿学习，专家演示每个场景仅有一种行为，模型难以学习其他有效行为，也无法在测试时适应用户偏好；同时演示中的噪声或歧义会带来不一致监督。驾驶本质上是多模态的，且多样性主要来自速度而非路径。

Result: 在 Bench2Drive-Speed 上取得 82.53 的 Speed-Adherence Score，较最强基线 TFv6 提升 12.9 分并达到 SOTA；在 Bench2Drive 上较 UniAD 和 VAD 分别提升驾驶分数 +28.5 和 +32.7；在真实数据构建的 NAVSIM v2 上较 backbone 提升 EPDMS +6.1。

Insight: 创新点在于将路径学习与速度选择解耦：路径仍可模仿学习，速度则通过多准则评分函数选择，并由物理安全约束监督，避免速度维度的开环模仿缺陷。该方法作为插件式规划头可集成到现有端到端模型中，并首次在无速度条件训练下支持测试时显式速度偏好控制。

Abstract: End-to-end autonomous driving has predominantly relied on open-loop imitation learning for training, which presents two limitations. First, expert demonstrations show only one behavior per scenario, so the model cannot learn alternative valid behaviors and, in turn, cannot adapt to user-specified preferences at test time. Second, noise or ambiguity in these demonstrations can produce inconsistent, unreliable supervision for the end-to-end models. Driving is inherently multi-modal: multiple different behaviors can be valid responses to the same scene, and this diversity arises largely from speed rather than path. We introduce a planning paradigm that learns the path through imitation while scoring a set of candidate speeds across multiple criteria with a learned function supervised by physically grounded safety constraints. We realize this paradigm as Closed-Loop Aware Planning for Autonomous Driving (CLAP-AD), a plug-in planning head that integrates with existing end-to-end driving models and, to our knowledge, is the first to enable explicit test-time control over their speed selection without speed-conditioned training. On Bench2Drive-Speed, CLAP-AD sets a new state of the art Speed-Adherence Score of 82.53, topping TFv6, the strongest baseline on this metric, by 12.9 points while retaining strong closed-loop driving performance. On Bench2Drive, it improves driving score by +28.5 and +32.7 over UniAD and VAD, and on NAVSIM v2, built on real-world driving data, EPDMS by +6.1 over its backbone. Project page: https://clap-ad.github.io

[113] Camera-to-Quadruped Robot Pose Estimation from Sequential Images cs.RO | cs.CVPDF

Seobin Hwang, Ji-Seong Beom, Yeong-Jun Cho

TL;DR: 论文提出QuadPose，从单目视频序列中估计四足机器人全身18-DoF位姿，包括6-DoF基座位姿和12-DoF关节角度，并构建首个相机到四足机器人位姿估计benchmark QuadSyn。方法面向关节时变、腿部对称和自遮挡导致关键点对应模糊的问题，在自建benchmark上取得较高精度并达到约33 FPS。

Details

Motivation: 四足机器人常用于灾害响应、设施巡检等危险环境，外部相机监控对安全运行很重要。但四足机器人是多关节系统，存在腿部关节时变、左右腿对称和频繁自遮挡，导致关键点对应歧义，且此前缺少公开benchmark，难以定量比较。

Result: 在QuadSyn上，QuadPose达到92.18% PCK@2.5px、95.54% PCK-AUC、85.94% ADD@20cm、66.98% ADD-AUC，平均关节角误差为3.65度，运行速度约33 FPS。

Insight: 创新点在于从单目序列联合估计基座位姿与关节角度，实现四足机器人全身18-DoF位姿估计，并发布首个相机到四足机器人位姿benchmark QuadSyn。客观来看，利用视频时序信息缓解对称腿和自遮挡带来的关键点歧义，同时填补该任务缺乏公开评测基准的空白，具有较强借鉴意义。

Abstract: Quadruped robots are increasingly deployed in hazardous environments such as disaster response and facility inspection, where external-camera monitoring is essential for safe operation. This setting is challenging because quadrupeds are articulated systems with time-varying leg joints, leg symmetry, and frequent self-occlusions, which make keypoint correspondence ambiguous. Moreover, no public benchmark exists, making quantitative comparison difficult. To address these challenges, we propose QuadPose, which estimates the full-body 18-DoF pose (6-DoF base pose + 12-DoF joint angles) from monocular video sequences, and introduce QuadSyn, the first camera-to-quadruped robot pose benchmark. On QuadSyn, QuadPose achieves 92.18% PCK@2.5px, 95.54% PCK-AUC, 85.94% ADD@20cm, 66.98% ADD-AUC, and 3.65 degree mean joint-angle error at approximately 33 FPS.

[114] TAPESTRY: A Single Backbone for Video Generation and Robot Control cs.RO | cs.AI | cs.CV | cs.LGPDF

Jeremy A. Collins, Seungjae Lee, Rachanon Wachakorn, Krishnan Srinivasan, Animesh Garg

TL;DR: TAPESTRY 将预训练视频扩散 Transformer 改造成机器人策略，把机器人动作表示为视频 latent 空间中的单 patch token，并与视频 token 共享 patchify 层和 Transformer 骨干。模型通过视频扩散、联合视频+动作扩散和仅动作扩散三种目标训练，推理时无需生成视频，仅用单步 DDIM 从纯噪声生成动作块。

Details

Motivation: 预训练视频生成模型包含丰富的物理交互先验，但用于机器人控制时通常需要在测试时进行昂贵的视频生成，或从零训练动作头而无法充分利用预训练骨干。论文希望低成本、高效地将视频生成先验迁移到机器人策略中。

Result: 在多个仿真 benchmark 和真实机器人实验中取得较强性能，推理成本显著低于可比的联合视频-动作策略；在消费级硬件上每个动作块约 84 ms，单步 DDIM 即可生成平滑动作块。

Insight: 核心创新是将动作 token 化为视频模型 latent 空间中的单 patch token，使动作和视频共享同一 Transformer 骨干，并通过三种扩散目标建立动态先验、联合建模和动作去噪。客观上看，其可借鉴之处在于统一动作与视频 patch 表示、避免推理时视频生成，以及用单步动作生成降低部署成本。

Abstract: Pretrained video generation models encode rich priors about physical interactions, but leveraging them for robot control typically requires either resource-intensive video generation at test time or action heads trained from scratch that do not fully utilize the pretrained backbone. We introduce Tokenizing Actions as Video Patches in a Shared Transformer Policy (TAPESTRY), a world action model that repurposes a pretrained video diffusion transformer into a robotic policy without requiring video generation at inference. Our key idea is to treat robot actions as single-patch tokens in the video model’s latent space. A lightweight linear projection maps each action timestep to the patch dimensions of the pretrained VAE, and the resulting tokens are processed by the same patchifying layers and transformer backbone used for video. The shared backbone is optimized through three diffusion objectives sampled per minibatch element: Video diffusion uses VAE latents from robot and human videos to establish dynamics priors. Joint video+action diffusion trains on robot demonstrations using a coupled diffusion timestep and an asymmetric attention mask. Action-only diffusion denoises actions conditioned on clean image latents. At test time, the model generates a compact action token sequence in a single DDIM step from pure noise, producing smooth action chunks at 84 ms per action chunk on consumer hardware and avoiding the cost of full video generation. Across simulation benchmarks and real-robot experiments, TAPESTRY achieves strong performance with substantially lower inference cost than comparable joint video–action policies. More information can be found at https://tapestry-wam.github.io.

[115] PDRVLA: A Vision-Language-Action Model with Physical Dynamics Representations cs.RO | cs.CV | cs.LGPDF

Hanqing Wang, Zhaotian Cai, Ruichuan An, Bohan Zeng, Tao Zhang

TL;DR: 论文提出 PDRVLA，一种带有物理动态表示的视觉-语言-动作模型。该方法通过轻量级物理动态模块从视觉-语言特征中提取物理 token 来条件化动作生成，避免在线生成未来视频带来的高延迟。模型通过预测未来 RGB、深度和光流特征进行学习，并利用冻结的预训练视频世界模型提供辅助监督。

Details

Motivation: 现有 VLA 模型对物理动态的理解有限，而视频世界模型虽能捕捉场景动态，但生成未来视频会显著增加控制回路延迟。论文希望在不进行在线视频生成的条件下增强 VLA 的物理动态理解能力。

Result: 在四个仿真基准上取得较高成功率：LIBERO 99.3%、LIBERO-Plus 89.9%、RoboTwin 2.0 93.6%、SimplerEnv 81.5%。消融实验表明世界模型监督可改善未来预测，并分析了子任务条件、未来物体位姿监督和 ECoT 训练对动作成功率的影响。

Insight: 创新点在于将物理动态表示学习与高成本视频生成解耦，用轻量模块提取物理 token 并条件化动作生成；通过未来 RGB/深度/光流特征预测和冻结视频世界模型辅助监督学习动态表示，同时结合人到机器人数据合成、子任务标注和具身思维链标注扩展训练数据。

Abstract: Vision–Language–Action (VLA) models generate robot actions from visual observations and language instructions, but their understanding of physical dynamics remains limited. Video world models can capture scene dynamics, yet generating future videos introduces high latency into the control loop. We introduce PDRVLA, whose lightweight Physical Dynamics Module extracts physical tokens from vision–language features to condition action generation without online video generation. The module learns through direct prediction of future RGB, depth, and optical-flow features, with auxiliary supervision through a frozen pretrained video world model. Training relies entirely on publicly available data, expanded through human-to-robot synthesis, subtask annotation, and embodied chain-of-thought (ECoT) annotation. Across four simulation benchmarks, PDRVLA achieves success rates of 99.3% on LIBERO, 89.9% on LIBERO-Plus, 93.6% on RoboTwin~2.0, and 81.5% on SimplerEnv. Ablations show improved future prediction with world-model supervision and evaluate how subtask conditioning, future object-pose supervision, and ECoT training affect action success.

[116] ThinkWorld: World Imagination Reinforces Chain-of-Dynamics Reasoning cs.RO | cs.AI | cs.CVPDF

Chaehyun Kim, Chi-Pin Huang, Jisoo Kim, Min-Hung Chen, Yu-Chiang Frank Wang

TL;DR: 论文提出 ThinkWorld，一个面向视觉-语言-动作（VLA）模型的学习框架。它将动作推理表示为 Chain-of-Dynamics（CoD），即连接当前世界与未来状态的离散运动 token 序列，并作为跨具身共享的动作表示。方法先用带动作标签的示范初始化 CoD，再通过世界想象预测每个 CoD 诱导的未来，并强化预测与真实观测更匹配的 CoD，从而利用无动作标注视频进行监督。

Details

Motivation: VLA 模型在机器人操作中虽有进展，但泛化受限于带动作标签机器人示范的多样性和覆盖范围。人类视频提供更广泛的经验，但从人类视频学习面临可扩展数据与显式动作 grounding 之间的根本权衡。

Result: 仿真和真实机器人实验表明，该方法能有效实现人类到机器人的迁移，并对未见任务和环境具有较强泛化能力；摘要未给出具体定量指标或 SOTA 对比。

Insight: 创新点在于用离散运动 token 序列 CoD 作为跨具身共同动作表示，并以世界动态预测结果作为强化信号，使无动作标注视频可通过观测状态转移提供监督。客观来看，该工作将动作推理、世界模型式未来预测和强化学习结合，把人类与机器人经验统一到同一推理空间。

Abstract: Vision-Language-Action (VLA) models have made substantial progress in robotic manipulation, but their generalization remains constrained by the diversity and coverage of action-labeled robot demonstrations. Human videos provide much broader behavioral and environmental experience, while learning from them faces a fundamental trade-off between scalable data and explicit action grounding. We introduce ThinkWorld, a VLA learning framework that grounds action reasoning in world dynamics and strengthens it through reinforcement learning with world imagination. ThinkWorld represents each action thought as a Chain-of-Dynamics (CoD), a sequence of discrete motion tokens that captures the intermediate dynamics connecting the current world to a future state and serves as a common action representation across embodiments. We first initialize CoD with action-labeled demonstrations, then predict the future induced by each CoD and reinforce those whose predictions better match the observed outcome. This enables videos without action annotations to provide direct supervision from observed state transitions, allowing human and robot experience to refine the same reasoning space. Experiments in simulation and on real robots demonstrate effective human-to-robot transfer and strong generalization to unseen tasks and environments.

eess.IV [Back]

[117] When Slides Met Scans: Cross-Attention for Renal Cancer Prognosis eess.IV | cs.CVPDF

Anusha Krishnamoorthy, Basudha Pal

TL;DR: 论文提出一种基于双向交叉注意力的多模态融合方法，将组织病理学原型与肾脏中心CT token连接，用于肾癌预后预测。方法在TCGA-KIRC队列上取得最高平均C-index和3年AUROC，并在CPTAC-CCRCC外部队列上进行了验证，同时通过归因分析揭示预测相关图像区域及采集伪影。

Details

Motivation: 组织病理和CT从不同尺度提供肾癌互补信息，论文旨在有效融合这两种模态以提升肾癌预后预测，并关注跨尺度多模态融合与外部验证问题。

Result: 在210例TCGA-KIRC患者、46个DSS事件上，5折评估C-index为0.760±0.081，3年AUROC为0.773（95% CI 0.678–0.860），在评估的单模态与融合基线中平均表现最高；在180例CPTAC-CCRCC外部测试中，不使用CT时DSS C-index为0.629。

Insight: 创新点在于用双向交叉注意力连接组织学原型与冻结基础模型提取的CT token，实现跨尺度多模态融合；同时结合归因分析定位预测相关图像区域并暴露采集伪影，强调独立多模态验证的重要性。

Abstract: Histopathology and computed tomography (CT) provide complementary views of renal cancer. We connect these modalities through bidirectional cross-attention between 16 histology prototypes and up to 32 kidney-centered CT tokens from frozen foundation models. In 210 TCGA-KIRC patients with 46 disease-specific survival (DSS) events, 5-fold evaluation yielded a C-index of $0.760\pm0.081$ and 3-year AUROC of 0.773 (95% CI 0.678–0.860), the highest averages among evaluated unimodal and fusion baselines. External testing in 180 CPTAC-CCRCC patients yielded DSS C-index of 0.629 with CT withheld. Attribution analysis linked predictions to image regions while exposing acquisition artifacts. These findings support further study of cross-scale fusion and independent multimodal validation.

[118] Quantized Edge Inference of a Self-Supervised SAR-Optical Encoder on Commodity x86 Hardware eess.IV | cs.CV | cs.LGPDF

Christian Sanchez-Leon, Dongming Chen

TL;DR: 论文将预训练的SAR-光学掩码自编码器视觉Transformer编码器部署到三类商用x86边缘节点上，系统评估FP32、FP16和INT8在CPU及集成GPU上的推理延迟、能耗和线性探针精度。研究发现INT8量化几乎不损失精度，并能降低CPU延迟和能耗；最优设备会随CPU能力提升从集成GPU转向CPU。

Details

Motivation: 自监督视觉Transformer越来越多用于对地观测可复用编码器，但其在野外可用硬件上的推理成本缺乏系统刻画，论文旨在为边缘部署中的精度、设备和量化选择提供依据。

Result: INT8使micro-mAP仅变化+0.0006，处于随机种子波动范围内；INT8在所有节点上降低CPU延迟，无AVX2时约1.45倍，VNNI参考节点约1.53倍，并在测量范围内降低CPU能耗；最弱节点上集成GPU比CPU快4.38倍，但随着CPU等级提升Pareto最优设备转向CPU。

Insight: 创新点在于面向SAR-光学自监督编码器的商用x86边缘推理基准与帕累托分析，表明INT8量化在该任务上精度中性，使数值精度选择主要成为成本决策，并为CPU/集成GPU的部署选择提供实证指导。

Abstract: Self-supervised vision transformers are increasingly used as reusable encoders for Earth observation, but their inference cost on the hardware available in the field is rarely characterized. We deploy a pretrained SAR-optical masked-autoencoder encoder on three commodity x86 nodes, from a low-power CPU without AVX2 to an AVX2+VNNI reference, at FP32, FP16 and INT8 on the CPU and, where present, the integrated GPU, and report latency, energy (RAPL and wall plug) and linear-probe accuracy. Quantization is accuracy-neutral: INT8 changes micro-mAP by only +0.0006, within the seed spread, so precision becomes a pure cost decision. INT8 lowers CPU latency on every node, by 1.45x even on the CPU without AVX2, and CPU energy wherever it is measured, and the VNNI reference (1.53x) suggests a further gain beyond AVX2. The Pareto-optimal device shifts from the integrated GPU, which is 4.38x faster than the CPU on the weakest node, to the CPU as the CPU class improves.

Table of Contents

cs.CL [Back]

[1] OmniBrain-0: A Unified Agentic Model for Embodied Intelligence cs.CL | cs.CVPDF

[2] Lipi-Ghor: A Large-Scale, Diarized, Multi-Domain Bengali Speech Corpus and a Scalable Pipeline for Building It cs.CLPDF

[3] CONCUR: Concordant Auxiliary Rewards for Multi-Task, Multi-Reward Reinforcement Learning cs.CLPDF

[4] Structured Maps for Landmark-Based Route Instructions with LLMs cs.CLPDF

[5] Is Token Confidence Just Chain Length? A Signal-Processing Audit of Reasoning Traces cs.CLPDF

[6] CAMR: Compatibility-Aware Memory Replay for Continual Learning in Large Language Models cs.CLPDF

[7] DocSkill: Skill-Integrated Agentic Reasoning for Multilingual Multi-Hop Document Question Answering cs.CL | cs.CVPDF

[8] Developing a Speech Language Model for Basque cs.CL | cs.AI | eess.ASPDF

[9] DocXEval: A Systematic Evaluation of Explainability in Multimodal Document Understanding cs.CL | cs.CVPDF

[10] Hierarchical Low-Rank Reparameterization for Parameter-Efficient cs.CLPDF

[11] Text Generation in Low-Resource Languages Using Human-Curated Vocabularies cs.CLPDF

[12] UCVE: Graph Structured Retrieval for Compliance Validation cs.CLPDF

[13] Same Verdict, Different Readers: Personalizing Fact-Checking Explanations from Relative Preferences cs.CLPDF

[14] Teach reality, not ethics: a drop-in continued-pretraining corpus installs coupled human-biosphere-AI welfare dispositions across model families cs.CLPDF

cs.CV [Back]

[15] Task-Utility IQA: Rethinking Image Quality Assessment in Embodied Tasks cs.CV | cs.AIPDF

[16] Physiognomy Revisited: How Vision-Language Models Encode Sensitive Attributes Through Image Transformation cs.CV | cs.AI | cs.CL | cs.CYPDF

[17] Learning and Reasoning with Geometry: How Generation Helps Spatial Understanding cs.CVPDF

[18] Any3R: Feedforward 3D Reconstruction from Any Camera Type cs.CVPDF

[19] Learning to Read Braille: Benchmarking and Improving Braille Recognition Capabilities of VLMs cs.CV | cs.CLPDF

[20] UT-HMR: Uncertainty-Guided Temporal Modeling for Robust 3D Human Mesh Recovery cs.CVPDF

[21] Diagnosing and Correcting Bias in MLLM for Long Video Understanding cs.CVPDF

[22] Hierarchical Multimodal Interaction Reasoning for Dyadic Relationship Recognition cs.CV | cs.MMPDF

[23] Cross-modal Steering Vectors Exist Across Both Image and Text Generation cs.CV | cs.CL | cs.LGPDF

[24] Training-Free Inference Acceleration for Video Large Language Models: Methods, Trade-offs, and Evaluation cs.CVPDF

[25] CRLoMTL: Conflict-Resolution through Low-rank matrices for Multi-Task Learning cs.CVPDF

[26] BoxWorld: Let World Engines Control and Video Generators Render cs.CVPDF

[27] Evaluating Federated Vision-Language and Vision-Only Models for Privacy-Aware Breast Cancer Imaging cs.CV | cs.CR | cs.LGPDF

[28] SOMA: A Benchmark for Fine-Grained Egocentric Behavior and Body-Motion Understanding cs.CV | cs.AI | cs.CL | q-bio.NCPDF

[29] LociArt: A Unified Paradigm for Free-Style Local Video Stylization cs.CV | cs.AI | cs.MMPDF

[30] Nemotron-Labs-Diffusion-Uni: Elastic Self-Refining Unified Diffusion LLM with Cross-Paradigm On-Policy Distillation cs.CVPDF

[31] Confident First, Reflective Second: Reinforcement Learning of Confidence-Preserving Reflection for Video Understanding cs.CV | cs.MMPDF

[32] ARISE: Authentic Reference-Informed Stability Enhancement for Continual Multimodal Forgery Detection cs.CVPDF

[33] Identity or Evidence? Auditing Evidence Dependence in Agentic Life-Science Benchmarks cs.CVPDF

[34] FrameForge: Harnessing Priors of Video Foundation Models for Unified Image Generation cs.CV | cs.AIPDF

[35] PRISM-BENCH: Diagnosing Grounded and Faithful Multimodal Reasoning Beyond Final-Answer Accuracy cs.CV | cs.CLPDF

[36] Decoupling Consistency and Conflict across MLLM Layers for Fake News Detection cs.CV | cs.CLPDF

[37] LabWorld: Learning Interactive 4D Worlds for Scientific Experimentation cs.CVPDF

[38] WithinVAE: Compact Visual Autoencoding with Self-Distilled Semantics cs.CVPDF

[39] Adapting Video Models for Robot Control via Motion-guided Asymmetric Prediction cs.CV | cs.AIPDF

[40] Look Only When Needed: Certificate-Guided Visual Routing for Efficient Diffusion VLMs cs.CV | cs.AI | cs.LGPDF

[41] SAGE: Saliency-Annealed Gated Encoder for Universal Vision Recognition cs.CVPDF

[42] Hy-Nomas: Towards Free-form Interactive World Generation cs.CVPDF

[43] Efficient Remote Sensing Image-Text Retrieval via Interaction-Free Recall and Text-Guided Reranking cs.CVPDF

[44] Evaluation-Guided Demonstration Refinement for Vision-Language-Action Policies: The Winning Entry in the MetaFine Open Competition cs.CVPDF

[45] CVBL-CCTV-Face Dataset for Real-World Face Enhancement and Recognition in Surveillance cs.CVPDF

[46] Latent Visual Contrastive Reasoning with Multi-Branch Exploration cs.CV | cs.CLPDF

[47] Drop the Details, Get the Gist: Dynamic Conceptual Representation for Video Recognition cs.CVPDF

[48] MVRCap: Towards Comprehensive Video Captioning via Hierarchical Multi-Dimensional Verifiable Rewards cs.CVPDF

[49] LingGS: Decoupling Feed-Forward Geometry Prediction from Camera Tracking for High-Fidelity Gaussian SLAM cs.CV | cs.GRPDF

[50] AdaGround: Token-Normalized Adaptive Zooming with Verifier-in-the-Loop Refinement for GUI Grounding cs.CV | cs.AI | cs.CLPDF

[51] PREEN: Exemplar-Based Color Style Transfer with Smartphone Editor Presets cs.CV | cs.LGPDF

[52] Mining the Nineteenth Century for the Thirteenth: Register-Scale Forced Alignment and HTR Bootstrapping via Legacy Regesta and Antiquarian Editions cs.CV | cs.CLPDF

[53] Explicit Regional Representations For Video Quality Assessment cs.CVPDF

[54] LearnPrunerV2: Learnable Progressive Token Pruning for Vision-Language Models cs.CV | cs.AIPDF

[55] RePAIR: Inverting Backdoor Attacks in Semantic Segmentation cs.CV | cs.CR | math.PRPDF

[56] Ground-to-Satellite Knowledge Distillation for Crop Classification under Temporal and Geographic Shift cs.CVPDF

[57] Zooming without Zooming v2: From Seeing Details to Connecting Evidence for Generalizable Multimodal Perception cs.CV | cs.AI | cs.CL | cs.LGPDF

[58] Learning to Code the World: Harness Evolution for Compositional 3D Reconstruction cs.CVPDF

[59] Learning to Look Beyond Screens: Transferable Visual Observation from GUI Grounding cs.CV | cs.AIPDF

[60] TriDifNet: A Unified Three-Task Diffusion Network for End-to-End RGB-T Perception cs.CVPDF

[61] SplatScout: Query-Driven Part-Level Referring Segmentation in Gaussian Scenes cs.CVPDF

[62] A First-Step Bias Restores Lost Diversity in Diffusion Model Distillation cs.CV | cs.AIPDF

[63] Multibranch Ensemble Head in Vision Foundation Models for Efficient Cucumber Phenological Stage Classification cs.CVPDF

[64] FastBEVFusion: Simple Multi-Modal Fusion for 3D Object Detection at 50 FPS cs.CV | cs.ROPDF

[65] VGIQA: Fine-Grained Image Quality Assessment with Visual Grounding Rewards cs.CVPDF

[66] 3D-SceneQ: Empowering 3D LLM With Query-Guided Adaptive Pruning and Multi-Modal Feature Enhancement cs.CVPDF

[67] Uncovering Shortcut Amplification in Video RLVR through the Lens of Perceptual Uncertainty Collapse cs.CV | cs.CLPDF

[68] Zhengrong-Spatial-Aware 1.0: Advancing 3D Multimodal Intelligence cs.CVPDF

[69] Vision-Language-Action or World-Action? Just Action Models cs.CV | cs.ROPDF

[70] HDVS: semi-supervised semantic segmentation via heterogeneous dual-branch voting supervision cs.CVPDF

[71] GameVerse: A Minute-Scale Gameplay Dataset for Long-Horizon Interactive World Modeling cs.CVPDF

[72] Taming Noise Initialization for Zero-shot Video Restoration with Image Diffusion Priors cs.CVPDF

[73] TRUCO: A Scalable Lock-Free Algorithm for Parallel Contour Extraction cs.CVPDF

[74] CodeCap-Bench: Can Code Serve as a Good Video Caption? cs.CVPDF

[75] TimeLineXR: An Interactive Exploration of Time in Vision-Language Models cs.CVPDF

[76] Representing the Unseen: Rethinking Visual Imagination for Spatial Reasoning cs.CV | cs.AI | cs.CLPDF

[77] PerceptRL: Making Visual Perception Rewardable in MLLMs from Answers to Interactions cs.CV | cs.AIPDF