cs.CL [Total: 19]
cs.CV [Total: 72]
cs.AI [Total: 6]
cs.CR [Total: 1]
cs.DC [Total: 1]
cs.LG [Total: 6]
cs.RO [Total: 5]
eess.IV [Total: 2]

cs.CL [Back]

[1] Why Pay for Long Reasoning When Rewrite Do Trick? Causal Effects of Reasoning History Compaction on Future Generation and Agent Cost cs.CL | cs.AIPDF

Anton Hristov

TL;DR: 论文研究在不重新训练模型的情况下，通过将模型先前的推理历史改写为紧凑的状态记录，来减少后续推理的token消耗。在多个测试模型上，压缩后的历史通常能缩短未来推理；在软件代理部署中可降低总成本，但任务质量和包含改写的时间取决于目标与压缩器。

Details

Motivation: 长推理链虽然提升语言模型解决难题的能力，但导致推理成本高昂。由于专有模型的隐藏草稿无法直接研究，作者探索开放权重模型能否通过改变自身早期思考的形式来生成更简洁的推理，从而降低推理token使用和代理成本。

Result: 在测试的模型上，被接受的紧凑历史通常能缩短后续推理；编码案例显示共享候选解后重复验证减少。在软件代理部署中，压缩可降低总花费，但任务质量和包含改写的时间依赖于目标与压缩器。

Insight: 创新点在于将推理历史作为可干预对象，通过改写为紧凑状态记录来因果地影响未来推理长度和代理成本，且无需重训练。客观可借鉴之处是提出模型特定的压缩策略，并同时评估效率与能力保留。

Abstract: Reasoning substantially improves language models’ ability to solve difficult problems, but long reasoning traces make these gains expensive. Efficient reasoning should express useful analysis in fewer tokens by reducing repetition and text that adds little value to solving the problem. Progress in proprietary models motivates this goal, yet their hidden scratchpads limit direct study of how it is achieved. We investigate whether open-weight models can generate less verbose reasoning by changing the form of their own earlier thoughts, without retraining. Our intervention rewrites prior reasoning into compact state records before pending tool results are revealed, then compares continuations from original and compact histories. Across the tested models, accepted compact histories usually shorten future reasoning; a coding case study also shows less repeated verification after a shared candidate solution. In software-agent deployments, compaction can reduce total spending, although task quality and rewrite-inclusive time depend on the target and compressor. These findings identify reasoning history as one way to reduce future reasoning-token use and motivate model-specific compaction policies evaluated for both efficiency and retained capability.

[2] Balancing Chain-of-Thought and Answer Supervision for Small Language Models on Medical Question Answering cs.CL | cs.AIPDF

Le Zhou, Xiao Luo, Chinmay Chandra, Zhan Zhang

TL;DR: 论文研究小型语言模型在医学问答任务上的思维链（CoT）监督微调策略，比较对整个CoT与答案序列等权加权的UaS，以及分别加权CoT损失和答案损失的SwS。实验覆盖多个通用与医学领域模型、选择题QA和自由文本QA，发现不存在通用最优策略：选择题任务更偏向SwS，自由文本任务更偏向UaS，且过长CoT可能损害部分模型性能。

Details

Motivation: 小型语言模型适合本地医学问答以降低隐私和延迟风险，但有效微调仍困难；论文旨在探究CoT监督中损失加权应如何匹配任务格式和基座模型已有医学知识。

Result: 在MedQA和EMS-MCQA选择题QA上，SwS通常更优，通用模型受益最大，Qwen3.5-4B在MedQA上较零样本基线最高提升44.8%，而医学领域模型提升较小甚至为负；在Med-O1自由文本QA上，UaS对几乎所有模型最佳，3B Llama-3.2接近更大7B医学领域模型。

Insight: 主要创新在于系统比较UaS与SwS两种CoT监督损失加权，并指出不存在适用于所有临床QA的统一微调策略，需按任务格式和基座模型知识选择加权方式；此外，过长CoT与性能下降相关，提示训练时限制CoT长度可能是有益的实用策略。

Abstract: Small language models (<=9B parameters) can support local medical question answering (QA) while reducing privacy and latency concerns, but effective fine-tuning remains challenging. This paper compares two chain-of-thought (CoT) supervised fine-tuning methods: Uniform-across-Sequence (UaS), which applies equal loss weighting to the full CoT-and-answer sequence, and Segment-weighted (SwS), which separately weights CoT and answer losses. We evaluate six general- and medical-domain models on multiple-choice QA (MedQA and EMS-MCQA) and free-text QA (Med-O1). Results show that neither method is universally best: on multiple-choice QA (MedQA and EMS-MCQA), SwS generally performs better, with general-purpose models benefiting most; notably, Qwen3.5-4B improves by up to 44.8% over its zero-shot baseline on MedQA. In contrast, medical-domain models show smaller or even negative gains on MCQs. On free-text QA (Med-O1), UaS performs best for nearly all models and provides consistent improvements across both model types, with the 3B Llama-3.2 model approaching the performance of much larger 7B medical-domain models. We also find that excessively long CoT generation is associated with lower performance in some models on either multiple-choice or free-text QA tasks, suggesting that limiting the CoT length during training may benefit selected models. Overall, the results show that CoT loss weighting should be matched to the task format and the knowledge already encoded in the base model, rather than using a single fine-tuning strategy for all clinical QA tasks.

[3] Dory1.0-4B: A Looped Normalized-GPT cs.CLPDF

Maxim Fishman, Simeng Sun, Shantanu Acharya, Faisal Ladhak, Brian Chmiel

TL;DR: 论文提出Dory1.0-4B，一个40亿参数的循环式（潜在递归）语言模型，具有弹性深度：同一checkpoint可通过在推理时改变共享循环块的执行次数来适配不同计算预算。模型基于归一化Transformer（nGPT）骨干，交错滑动窗口与全注意力层，并使用6万亿token预训练、256K长上下文扩展以及SFT和RLVR后训练。实验表明增加循环次数可提升准确率，尤其在编码、智能体和长上下文任务上，同时通过仅缓存最后循环的KV降低循环推理内存。

Details

Motivation: 论文关注如何在不大幅增加参数和checkpoint数量的情况下，让语言模型在推理时灵活调整计算量，并提升长上下文、编码和智能体任务的性能与token效率。其出发点是利用潜在循环深度提供弹性推理计算，并缓解循环带来的显存开销。

Result: 在12个基准上，Dory1.0-4B使用两个或更多循环时平均超过Qwen3.5-9B、Gemma 4 12B和Gemma 4 26B-A4B等更大模型；循环次数增加带来准确率提升，编码、智能体和长上下文基准收益最大。智能体基准上，四循环比单循环少生成1.3-2.5倍token；仅缓存最后循环的KV可在任意循环数下保持与非循环模型相同的KV缓存占用，并保留大部分准确率收益。

Insight: 摘要宣称的创新包括：归一化Transformer骨干上的弹性深度循环结构、难度感知的循环课程用于后训练、面向智能体任务的潜在计算换token效率，以及仅缓存最后循环KV的内存优化。客观来看，其创新点主要在于将循环深度、训练/后训练课程和KV缓存策略系统结合，使单一4B checkpoint可覆盖不同推理预算并兼顾性能与效率。

Abstract: We introduce Dory1.0-4B, a 4-billion-parameter looped (latent-recurrent) language model with elastic depth: a single checkpoint serves different inference compute budgets. Dory1.0-4B is built on a normalized Transformer (nGPT) backbone with interleaved sliding-window and full-attention layers, organized into an input block, a shared recurrent block, and an output block. At inference, the recurrent block can be applied a variable number of times. We pre-trained Dory1.0-4B on 6 trillion tokens. The first 4.5 trillion tokens were trained in NVFP4, and additional pre-training then extended the context length to 256K tokens. Next, we post-trained the model using Supervised Fine-Tuning (SFT) and Reinforcement Learning with Verifiable Rewards (RLVR), with a difficulty-aware loop curriculum that routes harder prompts to deeper loops. Increasing the loop count improves accuracy, with the largest gains on coding, agentic, and long-context benchmarks. With two or more loops, Dory1.0-4B outperforms larger models such as Qwen3.5-9B, Gemma 4 12B, and Gemma 4 26B-A4B on average across 12 benchmarks. Looping also makes Dory1.0-4B more token-efficient on agentic benchmarks: with four loops, it generates 1.3-2.5x fewer tokens than with one loop, as the extra computation is spent in latent space rather than on longer outputs. To reduce memory during looping, Dory1.0-4B caches keys and values only from the last loop, keeping its KV-cache footprint equal to that of a non-looped model at any loop count while retaining most of the accuracy gains. We open-source the BF16, MXFP8, NVFP4, and GGUF quantized checkpoints.

[4] Language Model Augmented Decision Transformer for Multi-Task Offline Reinforcement Learning cs.CL | cs.LGPDF

Xinbo Zhao, Yuhang Liu, Xin Zhang, Lei Le, Zhaohan Xi

TL;DR: 论文提出LaMA-DT，一个统一序列建模框架，利用同一个大语言模型同时完成离线多任务强化学习中的轨迹增强和多任务策略学习。该方法用自然语言任务描述引导任务对齐的轨迹生成，并通过评分过滤保留高质量增强轨迹，再将同一LLM复用为语言条件Decision Transformer，在真实与增强混合数据上微调。

Details

Motivation: 现有离线多任务强化学习方法面临严重的单任务数据稀缺和跨任务泛化能力有限的问题，论文希望借助语言模型和语言引导的数据增强来缓解这两个瓶颈。

Result: 在Meta-World和MuJoCo多任务基准上，LaMA-DT一致优于竞争基线，尤其在低数据情形下表现更明显，表明语言引导的数据增强能显著提升跨任务泛化。

Insight: 创新点在于用单一LLM同时承担轨迹增强与多任务策略学习，并通过自然语言任务描述、监督生成与对齐目标以及评分过滤机制实现任务对齐的高质量增强；客观上看，复用LLM为语言条件Decision Transformer并将真实数据与增强数据混合微调，是较有借鉴价值的一体化设计。

Abstract: Offline multi-task reinforcement learning (MTRL) aims to learn a unified policy from heterogeneous offline datasets that generalizes across tasks. However, existing offline MTRL methods often struggle with two fundamental issues: severe per-task data scarcity and limited cross-task generalization. In this paper, we propose Language-Model Augmented Decision Transformer (LaMA-DT), a unified sequence-modeling framework that leverages a single large language model (LLM) for both trajectory augmentation and multi-task policy learning. LaMA-DT uses natural-language task descriptions to guide task-aligned trajectory augmentation through a set of supervised generation and alignment objectives, and incorporates a scoring-and-filtering mechanism to retain high-quality augmented trajectories. The same LLM is then repurposed as a language-conditioned Decision Transformer and fine-tuned on a mixture of real and augmented data. Experiments on Meta-World and MuJoCo multi-task benchmarks demonstrate that LaMA-DT consistently outperforms competitive baselines, particularly in low-data regimes, showing that language-guided data augmentation substantially improves cross-task generalization.

[5] Evaluation Contracts and Evaluation Leverage: Sliced-Gap Readings in Structured Prediction Can Move Without Model Change cs.CLPDF

Hao Lu, Weixing Zhu, Ruiyang Li

TL;DR: 论文指出，NLP评估中常见的切片性能差异（如文档级关系抽取的intra-inter gap）被当作模型能力证据，但实际上是模型与”评估契约”（打分函数、gold版本、阈值、输出格式、prompt语言）的联合属性：在预测冻结的情况下，仅更换gold标准就能使方法排序反转。作者将这一现象形式化为以rank-flip条件为核心的”评估杠杆”理论，并提出一种兼顾契约、声明平局、固定随机种子并经过门控过滤的审计式评估流程AER。

Details

Motivation: 当前NLP评估报告惯例性地将切片性能差异（尤其是DocRE中的intra-inter gap）解读为模型能力差异，却忽略了这些差异同时依赖于评估契约的选择。这可能导致在模型完全不变的情况下得出误导性的能力结论和方法排序。

Result: 在DocRE、HacRED、UD探针及WMT19网格上，gold标准交换可使方法排序反转（+0.0363，置信区间不含零）；重命名属于gauge变换（Delta=0），而重标注则带来+0.862的变化。AER的平局声明覆盖了WMT19网格上所有由契约引发的翻转（14/14；留出集73% vs 平局并集89%），在epsilon=0.10等价性下与最佳单一指标契约持平；在LLM-as-judge网格上，审计集成将13–15个不稳定配对全部稳定为零。

Insight: 创新点在于将评估结果重新定义为模型与评估契约的联合产物，提出”评估杠杆”框架与rank-flip条件来刻画切片差距的可移动性，并据此设计契约边际化、平局声明、种子规定与违规门控的AER流程，同时发布可复现的EvalLever审计工作流。其客观价值在于提醒社区：切片差距不能直接当作能力证据，评估需显式声明并审计契约选择。

Abstract: NLP evaluation reports routinely cite sliced performance differences, most prominently the intra-inter gap in document-level relation extraction, as capability evidence. We show that such gap readings are joint properties of models and evaluation contracts: scoring function, gold version, threshold, output format, and prompt language. With predictions frozen, a method ordering reverses under a gold-standard exchange (+0.0363, CI excluding zero; tie-level after), movement localizing where gold re-partitions: renames are gauge transformations (Delta=0; re-annotation +0.862). We formalize this as an evaluation leverage account centered on the rank-flip condition. Constructively, the audited evaluation reading (AER) is a contract-marginalized, tie-declared, seed-prescribed, gate-filtered procedure whose tie declarations cover every contract-induced flip on the declared WMT19 grids (14/14; held-out 73% vs. union-of-ties 89%), matching the best single metric contract at epsilon=0.10 equivalence. Evidence spans DocRE, HacRED, a UD probe, and released gold revisions; an alternative-reference-set replication extends it beyond our testbeds. Failed predictions retained as boundaries; a 5% violation gate flags unverifiable settings and bounds survivor-selection bias. We release EvalLever, a preregistered workflow with an amendment trail; on LLM-as-judge grids, the audited ensemble stabilizes every unstable pair (13-15 to zero; generic ensemble shrinkage: the preregistered correlation criterion failed, the companion stabilization criterion passed).

[6] Where Do Distilled Students Store Thinking? cs.CL | cs.LGPDF

Taeyeong Kim, Jaemin Jung, Ahhyun Kim, Unggi Lee

TL;DR: 论文研究蒸馏学生模型在教师模型停止“思考”（长推理）后，权重中保留了什么。作者比较三组教师-学生对的LoRA适配器在不同插值比例λ下的更新子空间，定位到一个与“类思考”行为（反思标记、独白格式、长解答）相关的共享适配器子空间，并验证移除或移植该子空间对行为的影响。

Details

Motivation: 大推理模型的长解答难以蒸馏到小模型，已有方法通过将LRM与短答模型按比例插值来缩短解答；论文想弄清教师停止思考时学生权重究竟保留或丢失了什么。

Result: 更新子空间在教师停止思考且解答明显缩短处变化最大；识别出的共享子空间在思考停止后大幅清空，移除它基本能关闭各教师对、学生和benchmark上的类思考行为。但在部分设置下与移除更新主成分相比优势有限，同能量随机方向也常接近；移植到短答学生可激活早期标记但学生很少完成解答。

Insight: 创新在于从权重子空间层面定位蒸馏后的“类思考”行为，而不仅是输出指标；通过LoRA适配器跨λ比较和移除/移植干预，将行为变化与共享子空间联系起来，为理解推理蒸馏和模型融合提供可解释视角。

Abstract: Small students degrade on large reasoning models’ (LRMs’) long solutions, so recent work shortens them by interpolating an LRM with a short-answer model at ratio $λ$. We ask what student weights retain when this teacher stops thinking, comparing LoRA adapters of three students from three teacher pairs at each $λ$. In each pair, the adapters’ update subspace shifts most where the teacher stops thinking and solutions shorten most, locating the change without attributing it. We identify a shared adapter subspace tied to thinking-like behavior (reflective markers, monologue format and long solutions). Shared by adapters before that point, it keeps about half (post hoc) while one pair’s teacher halves its solutions but keeps its think tag, and largely empties after thinking stops. Removing it largely turns this behavior off for every pair, student and benchmark. Only in that pair and one other student does it leave clearly fewer markers than removing the update’s top components; elsewhere, same-energy random directions often nearly match it. Grafted into a short-answer student at twice its weight, it turns early markers (a post hoc count) on, like the whole update, but the student rarely finishes. Our work identifies a weight subspace tied to distilled thinking-like behavior.

[7] Constructing a Regional Slang Lexicon for Electricity Demands via Soft-Label Bidirectional Anchoring and Active Learning cs.CL | cs.AIPDF

Rong Chang, Lijing Shi, Zhaoxiang Zang

TL;DR: 论文提出一种弱监督框架，从云南玉溪微信客服消息中构建面向电力需求的地区俚语词表。方法结合PMI与软标签双向锚定生成候选，利用Sentence-BERT语义相似度和jieba成词度打分，并通过不确定性感知主动学习及规则/检索增强LLM标注进行筛选，最后用经人工校准的LLM评估器评测。

Details

Motivation: 电网客服中用户常用方言俚语描述停电、账单和设备故障，存在非规范形态、语义漂移和未登录词，导致人工构建词表成本高且常规NLP工具效果下降。

Result: 在11,177条玉溪真实微信消息上，方言精确率DP=1.0且零假阳性；软标签分层在同等精度下比严格电力门控多保留751%候选，候选池由41扩至349；人工校准n=46得κ_dialect=1.0、κ_power=0.85，最终得到25–33词、其中4–5个电力相关。

Insight: 创新点在于软标签双向锚定将方言归属与电力相关性解耦并作为元数据，结合主动学习和检索增强LLM oracle降低标注成本；可借鉴处是弱监督候选生成、不确定性主动学习与LLM评估校准相结合的流程，适合低资源领域俚语资源构建。

Abstract: Power-grid customer service increasingly relies on community social media, where customers describe outages, billing, and equipment faults in informal regional dialect. Such messages exhibit non-standard morphology, semantic drift, and out-of-vocabulary expressions that make manual lexicon construction costly and degrade conventional NLP tools. We propose a weak-supervision framework constructing a regional slang lexicon from curated dialect anchors and power-domain seed terms. Candidates are generated via Pointwise Mutual Information with soft-label bidirectional anchoring, retaining dialect-anchored terms while recording power-domain relevance as separate metadata, and are scored by combining Sentence-BERT semantic similarity with a jieba-based wordhood measure. Uncertainty-aware active learning with rule-based and retrieval-augmented LLM oracles is applied under three query strategies, and the resulting lexicon is evaluated by a higher-capability LLM calibrated against human expert annotations via Cohen’s $κ$. On 11,177 real WeChat messages from Yuxi (Yunnan, China), the framework achieves an evaluator-estimated dialect precision of DP~~$=$~~1.0 with zero false positives. An ablation shows that soft-label layering retains 751% more candidates than strict power-gating at identical precision, expanding the candidate pool from 41 to 349 terms. Human calibration ($n=46$) yields $κ_{\text{dialect}}=1.0$ and $κ_{\text{power}}=0.85$. The resulting 25–33-term lexicon, 4–5 of which are power-relevant, provides a preliminary resource for dialect-aware power-grid service NLP.

[8] Do Multi-Turn Conversational Capabilities Carry Over from Text to Speech? A Parallel Multilingual Benchmark cs.CLPDF

Jiaming Zhou, Pei Zhang, Shiwan Zhao, Zhijun Wang, Baosong Yang

TL;DR: 论文构建了 Omnilingua-AudioMC，一个覆盖十种语言、文本与语音条件严格对齐的多轮对话多语言基准，用于检验文本多轮对话能力是否能迁移到语音交互。作者在六种语音语言模型上比较参考转录文本输入与直接语音输入的表现，并用配对评分项分析和替换模型生成转录文本的实验，定位语音多轮对话中的性能退化来源。

Details

Motivation: 现有多轮对话能力多基于转录文本评估，尚不清楚同一对话在语音输入下是否仍能保持历史记忆、持续指令遵循和对话状态修订等能力，也不清楚这种迁移是否跨语言一致。

Result: 在包含354段对话、1,404个评分项和十种语言的基准上，六种语音语言模型的文本到语音性能变化高度依赖模型和语言；转录表现相近的模型在语音下可能显著分化，跨语言相对优势也可能改变。对两个文本-语音差距最大的模型，替换前序用户轮为模型生成转录在十种语言中均比替换当前轮带来更大宏观提升，而最强直接语音模型收益很小。

Insight: 创新点在于严格平行对齐文本与语音条件的多语言多轮对话基准，以及配对评分项和轮次替换分析。客观看，其价值在于揭示转录文本表现不能可靠代理多语言语音对话能力，并将前序语音上下文的表示识别为可恢复退化的重要来源。

Abstract: Successful multi-turn voice interaction requires models to retain information from earlier turns, follow persistent instructions, and track revisions to conversational state. These abilities are usually evaluated from transcripts, leaving open whether successes demonstrated from text carry over when the same turns are spoken, and whether this transfer is consistent across languages. We introduce Omnilingua-AudioMC, a strictly parallel multilingual extension of Audio MultiChallenge, to study this question. The benchmark contains 354 conversations and 1,404 rubric items across ten languages, with content, turn structure, cross-turn dependencies, and evaluation criteria aligned across transcript and speech conditions. Across six speech-language models, the shift from reference transcripts to direct speech input is strongly model- and language-dependent: models with similar transcript performance can diverge substantially under speech, and their relative strengths across languages can change. Paired rubric-level analysis shows that aggregate transcript-speech gaps can obscure which conversational requirements are preserved, because losses of transcript-demonstrated successes can be offset by gains on other items. Replacing preceding user turns, the current turn, or both with model-generated transcripts localizes recoverable degradation: for the two models with the largest gaps, history substitution yields a larger macro-level improvement than current-turn substitution in all ten languages, while the strongest direct-speech model gains little from either. These results show that transcript performance is an unreliable proxy for multilingual spoken conversational ability and identify the representation of preceding spoken context as an important source of recoverable degradation. Our code is publicly available at https://github.com/QwenLM/Omnilingua-Bench/tree/main/Omnilingua-AudioMC.

[9] Typological Feature Induction as a Benchmark for Inductive Metalinguistic Reasoning cs.CLPDF

Chen Zhang, Seth Aycock, Yusuke Miyao

TL;DR: 论文提出类型学特征归纳（TFI）基准，将元语言推理拆分为“规则推断”步骤：给定未命名语言的句子及已知枢纽语言翻译，要求模型预测该语言的类型学特征。作者基于Grambank和多平行语料自动构建覆盖195种语言的评测问题，并发现LLM主要依赖对语言的先验识别而非真正从文本归纳。

Details

Motivation: 现有元语言推理基准同时要求推断规则和应用规则，导致失败难以诊断；论文希望单独评估模型从原始文本中归纳语言类型学特征的能力。

Result: LLM仅凭16个句对即可识别59%—80%的语言，翻译错位几乎不影响准确率；加密文本可提升语料效用，但主要限于目标与枢纽语言特征一致的情况。对于模型无法识别的语言，开放模型均未超过多数值基线。

Insight: 创新点在于将元语言推理中的规则推断独立成可扩展基准，并自动构建跨195种语言的评测集；客观分析表明，该工作揭示了LLM在TFI任务上依赖先验语言知识而非真正归纳，且当前模型尚不适合田野语言学式的原始文本归纳角色。

Abstract: Existing metalinguistic reasoning benchmarks require both inference of rules and their application, making failures hard to diagnose. We introduce typological feature induction (TFI) which targets the former step. Given sentences in an unnamed language paired with translations in a known pivot language, models must predict a typological feature of the unnamed language. From Grambank and multi-parallel corpora, we automatically build evaluation questions over 195 languages. We find that LLMs identify 59–80% of these languages from only 16 sentence pairs; and misaligning the translations barely affects their accuracy, suggesting reliance on prior knowledge of the language. Enciphering text to prevent recognition increases the corpus’ utility, but mainly for features where target and pivot languages agree. On languages which models cannot identify, neither open model outperforms a majority-value baseline. Since induction from raw text aligns with field linguists’ use-case, this work indicates the models tested are not yet ready for this role. TFI offers a scalable path to track progress in this direction.

[10] When Ground Truth Is Not Ground Truth: Automatic Labeling for Hallucination Detection cs.CL | cs.AIPDF

Rounak Sen, Divya Kumari, Deeksha Varshney

TL;DR: 论文系统研究了LLM幻觉检测中自动构造ground-truth标签的可靠性问题。作者指出，现有方法常用BLEURT等相似度指标并配合阈值来生成标签，但这会把相似度误当作事实正确性。通过在15种自动监督判别器、7种幻觉检测基线、2个生成模型、2个QA数据集和5种语言上的实验，论文发现检测器性能高度依赖监督判别器的选择；人工验证也显示不同判别器的标注错误率差异很大。

Details

Motivation: 幻觉检测需要可靠的ground-truth标签，但人工标注成本高；现有工作常以BLEURT等相似度分数自动生成标签，隐含假设相似度可代表幻觉/事实性。论文旨在揭示这种自动标注方式可能不可靠，并分析其对幻觉检测评估的影响。

Result: 实验覆盖15种自动监督判别器、7种检测基线、2个生成模型、2个QA数据集和5种语言，表明检测器性能会随监督判别器显著变化。人工验证中，QGen-9B总体误分类率最低为12.86%，而BLEURT和BLEURT-n分别达到47.23%和47.32%，说明基于相似度的自动标签可靠性较差。

Insight: 论文的创新点在于系统性地将“自动标签质量”本身作为幻觉检测评估的关键变量，并通过多语言、多模型、多判别器实验证明相似度监督不等于事实性监督。其可借鉴之处是：构建幻觉检测基准时应验证自动标注的可靠性，并优先采用更能捕捉语义与事实正确性的判别方法。

Abstract: Hallucinations in Large Language Models (LLMs) remain a major challenge, as they can produce misleading, factually incorrect, or unverifiable information, thereby undermining their trustworthiness and reliability. However, reliable ground-truth labels remain a key challenge in evaluating hallucination detectors. Recent studies often construct labels automatically by comparing model generations with reference answers using similarity metrics such as BLEURT. This implicitly treats BLEURT similarity as a proxy for hallucination, despite its English-centric supervision, unbounded score range, and design as a similarity metric rather than a factuality measure, making threshold-based ground-truth construction potentially unreliable. Through comprehensive experiments using 15 automatic supervision judges, seven hallucination detection baselines, two generator models, two QA datasets, and five languages, we demonstrate that detector performance varies substantially with the choice of supervision judge. Human validation reveals substantial differences in labeling reliability, with QGen-9B achieving the lowest overall misclassification rate (12.86%), while BLEURT and BLEURT-n exhibit rates of 47.23% and 47.32%, respectively. Our findings highlight the limitations of automatic similarity-based labeling and the need for more reliable judging methods that better capture semantic and factual correctness.

[11] ReSI-Math: Exploring Recoverable Capability Frontiers for Efficient Reasoning Improvement cs.CL | cs.AIPDF

Yao Cui, Zhen Yang, Yuandong Wang, Yangfu Zhu, Tianyu Wo

TL;DR: 论文研究RLVR中模型直接答错是否意味着缺乏能力，发现部分失败可通过探索模型自身轨迹恢复，并提出ReSI-Math框架：用自评判引导的进化搜索识别可恢复失败，并随策略演化自适应选择前沿问题用于序列化RLVR监督，从而将可恢复能力转化为数据高效的训练信号。

Details

Motivation: RLVR虽能提升大模型数学推理，但直接解码失败不一定代表模型真正缺乏该能力；论文希望挖掘模型自身轨迹中可被诱导出的潜在能力，并高效用于推理改进。

Result: 在Qwen3-VL-8B上，正确-错误轨迹交叉从621次尝试中得到272个验证器确认的正确后代（43.8%）；经过三个序列化one-shot RLVR阶段，MATH500和WE-MATH平均分从78.71提升到80.16；匹配单题RLVR下，前沿引导选择为54.16，高于随机选择的53.31。

Insight: 创新点在于提出策略依赖的“可恢复能力前沿”，将直接失败但轨迹搜索可恢复的样本转化为RLVR监督，并结合自评判进化搜索与随策略演化的前沿问题选择。客观看，其价值在于把失败样本从简单过滤对象变为可回收训练资源，提升RLVR的数据效率和能力边界利用。

Abstract: Reinforcement learning with verifiable rewards (RLVR) has become an effective approach for improving mathematical reasoning in large language models. However, a fundamental question remains: when a model fails to solve a problem, does it truly lack the underlying capability? In this work, we reveal that direct failure does not necessarily imply inability. Some reasoning failures can be recovered by exploring the model’s own trajectories, exposing a recoverable capability frontier: a policy-dependent region where solutions are not reliably generated through direct decoding but can be elicited through trajectory-level search. Based on this observation, we introduce ReSI-Math, a framework that explores recoverable capability frontiers and converts recovered capabilities into data-efficient RLVR supervision. ReSI-Math uses Self-Judge-guided evolutionary search to identify recoverable failures under a fixed policy and adaptively selects frontier problems for sequential RLVR as the policy evolves. On Qwen3-VL-8B, correct-incorrect trajectory crossover produces 272 verifier-confirmed correct offspring from 621 attempts (43.8%). After three sequential one-shot RLVR stages, the MATH500 and WE-MATH average improves from 78.71 to 80.16. Under matched single-problem RLVR, random selection reaches only 53.31 versus 54.16 for frontier-guided Q1.

[12] One Assistant, Many Memories: Benchmarking Real-World Group Assistance When Recall Is Not Enough cs.CL | cs.AIPDF

Zhanghao Hu, Linhai Zhang, Qian Zhao, Qi Zhu, Yuan Hua

TL;DR: 论文提出GroupAssistBench，首个基于真实群聊记录（50个群、80,299条消息）的群组助手记忆基准，用于评估记忆召回（L1，有参考回答）和记忆使用（L2，无固定答案的服务请求，覆盖单点、分布式和冲突记忆）。实验表明现有前沿模型和记忆系统在召回上可接近全上下文，但在记忆使用上显著落后，尤其面对分布式和冲突记忆；通过关联检索记忆和将记忆事实绑定为响应约束可带来稳定提升。

Details

Motivation: 现有记忆基准多针对个人化或智能体任务，且依赖合成对话和固定答案评估，缺乏对真实多用户群聊中无固定答案的辅助能力、跨成员信息整合、修订与冲突跟踪的评测。论文旨在填补这一空白，强调记忆的价值不仅在于召回，更在于使用记忆提供合适的群组辅助。

Result: 在GroupAssistBench上评估前沿模型、检索方法和记忆系统，最强记忆系统在召回（L1）上接近全上下文，但在使用（L2）上远落后，尤其在分布式和冲突记忆场景；加入“关联检索记忆”和“将记忆事实绑定为响应约束”两项改进后，性能获得一致提升。

Insight: 创新点在于构建首个基于真实群聊且部署过群组助手的记忆基准，将评估从固定答案的召回扩展到无固定答案的记忆使用，并设计单点、分布式、冲突三类记忆场景及基于真实用户反馈的惩罚性评分规则；客观看，其强调“召回不等于使用”的评测视角和记忆绑定约束的改进思路对群组记忆系统设计有借鉴意义。

Abstract: The value of memory lies not only in recall but in using it to provide appropriate assistance. Recent benchmarks test memory use in personalisation and agentic tasks, but seldom evaluate assistance without fixed answers or integration across multiple participants. Group chats make both challenges more acute: an assistant must integrate information from multiple members, track revisions and conflicts over time, and produce one response for the group. Yet existing group memory benchmarks rely on synthetic conversations and inherit constrained evaluation from general memory benchmarks, focusing on recall against fixed answers or decisions over a closed set. We introduce GroupAssistBench, the first memory benchmark from real-world group conversations with a deployed group assistant, covering 80,299 messages in 50 groups and costing over $53K. Over the same histories, GroupAssistBench evaluates memory recall (L1) with reference answers and memory use (L2) with service requests without fixed answers across single, distributed, and conflicting memories (M1–M3). Requests are verified to require group history, memories are anchored to their sources, and L2 uses tailored rubrics with penalties derived from real user feedback. Across evaluations of frontier models with full context, retrieval methods, and memory systems, the strongest memory system rivals full context on recall but falls far behind on use, especially for distributed and conflicting memories. Together, two additions that relate retrieved memories and bind remembered facts as response constraints bring consistent gains, showing that group memory systems must improve use, not recall alone.

[13] TraceRouter++: Causal Semantic Routing Realignment via Optimal Transport cs.CLPDF

Chuancheng Shi, Shangze Li, Wenhua Wu, Hengzhi Chen, Fei Shen

TL;DR: 论文提出 TraceRouter++，一种面向大型基础模型的“路由级”安全干预框架，用于抵御对抗性攻击和越狱。方法遵循 Discover-Trace-Realign 范式：先定位安全敏感起始层，并用稀疏自编码器将激活解耦为可解释路由特征；再通过特征影响分数追踪下游因果电路；最后用基于约束最优传输的软路由重对齐，将有害输入下的不安全激活边缘分布重定向到安全参考分布。

Details

Motivation: 现有安全干预多基于局部性假设，刚性抑制孤立神经元或特征；但有害语义表现为分布式、跨层路由电路，局部消融既易被复杂越狱攻击突破，又损害模型通用能力。

Result: 摘要在代表性 LLM、多模态大模型和扩散模型上进行大量实验，称 TraceRouter++ 优于所评估基线，在提升对抗鲁棒性的同时保持较好的安全-效用权衡；未给出具体数值或 benchmark 名称。

Insight: 创新点在于从路由电路层面进行安全干预，而非刚性抑制；结合注意力发散定位起始层、SAE 特征解耦、特征影响分数追踪因果电路，并引入约束最优传输的坐标级边缘近似软重对齐，在重定向恶意语义流的同时保留正交计算路径。

Abstract: Despite their remarkable capabilities, large foundation models (LFMs) remain susceptible to adversarial manipulation. Current safety intervention methods primarily rely on the \textit{locality hypothesis}, rigidly suppressing isolated neurons or features. However, harmful semantics manifest as distributed, cross-layer circuits, rendering localized ablations both brittle against complex jailbreaks and detrimental to general utility. To bridge this gap, we propose TraceRouter++, a routing-level safety intervention framework that explicitly traces safety-relevant causal routing pathways activated under harmful inputs and realigns their unsafe activation states. Our framework operates in a \textit{Discover-Trace-Realign} paradigm: it first pinpoints a sensitive onset layer via attention divergence and disentangles dense activations into interpretable routing features using sparse autoencoders (SAEs). Subsequently, it maps these features to downstream causal circuits via feature influence scores (FIS). Crucially, departing from rigid suppression paradigms, TraceRouter++ introduces a constrained optimal transport (OT)-based soft routing realignment. By formulating safety intervention as a tractable coordinate-wise marginal optimal transport approximation, our method redirects unsafe activation marginals toward safe-reference marginals within causally identified routing subspaces while preserving non-selected coordinates at the intervention point. This mechanism safely redirects malicious semantic flows while leaving orthogonal computational routes intact. Extensive experiments across representative large language models (LLMs), multimodal large language models (MLLMs), and diffusion models (DMs) show that TraceRouter++ outperforms evaluated baselines, improving adversarial robustness while maintaining a favorable safety-utility trade-off. WARNING: This paper contains unsafe model responses.

[14] Shuffled OPSD: When Semantic Consistency Hurts On-Policy Self-Distillation? cs.CL | cs.AIPDF

Qingyuan Zheng, Zhihao Dou

TL;DR: 论文研究On-Policy Self-Distillation（OPSD）中输入结构的作用，提出Shuffled OPSD：把学生与教师共享的问题陈述中的词序随机打乱，同时保持仅教师可见的完整解题示例不变。在三个数学推理基准和三个规模的Qwen3模型上，该方法比保留原句的对照以及标准OPSD的准确率更高；而在不同规模上打乱教师专属示例都会导致性能下降。

Details

Motivation: OPSD中，同一模型分别作为学生和教师；二者共享题目，学生仅依据题目生成轨迹，教师则额外看到完整解题示例并给出token级监督。直观上，保持题目语义结构一致应有益，但论文质疑这一假设，旨在探究共享输入的语义连贯性是否总是有益。

Result: 在3个数学推理benchmark、3个Qwen3模型规模上，平均准确率超过语义完整的对照与标准OPSD；对标准OPSD的最大提升为Qwen3-1.7B上的3.52个百分点。进一步对照显示，保留更大词块相比词级打乱无优势，而打乱教师专属解题示例在所有规模上都降低准确率。

Insight: 创新点在于揭示并利用了OPSD中“破坏共享问题表述反而有益、破坏教师专属参考却有害”的不对称现象，并通过跨问题混合和移除目标等实验，区分问题内容、词序和教师上下文三者的作用，挑战了语义连贯性普遍有益的假设，启示后续研究输入结构如何影响学生生成与教师监督。

Abstract: Knowledge distillation improves a student model’s reasoning ability through supervision from a teacher. On-policy self-distillation (OPSD) uses the same model as both teacher and student under different contexts: the student generates reasoning trajectories from the question alone, while the teacher additionally receives a complete worked example and provides token-level supervision along those trajectories. Because the problem statement is shared, its content and ordering shape both student generation and teacher feedback. Preserving its semantic structure therefore seems natural, but does it always improve learning? We introduce Shuffled OPSD, which shuffles the words of the shared problem statement while keeping the teacher-only worked example intact. Across three mathematical reasoning benchmarks and three Qwen3 model scales, the method improves mean accuracy over both an intact-target control and Standard OPSD under matched training configurations. The largest gain over Standard OPSD is 3.52 percentage points on Qwen3-1.7B. Further controls show that preserving larger word blocks offers no advantage over word-level shuffling at the evaluated scales, whereas shuffling the teacher-only worked example reduces accuracy at every scale. Cross-problem mixing and target-removal experiments further distinguish the roles of problem content, word order, and teacher context. These results show that disrupting the shared problem statement can improve learning, while disrupting the teacher-only reference hurts performance. They challenge semantic coherence as a universally beneficial principle for OPSD and motivate closer study of how input structure shapes student generation and teacher supervision.

[15] Who is the Pronoun Pro? Comparing LLMs and Humans at Number-Disambiguation of ‘they’ cs.CL | cs.AI | cs.HCPDF

Carlos Hartmann

TL;DR: 论文在受控标注任务中比较LLM与人类对英语代词“they”的单复数消歧能力，使用174条来自Reddit的真实用例，涉及创新和依赖语境的情况。73名人类标注者和40多种LLM配置参与，最佳LLM Kimi K3达到94.5%准确率，超过人类多数投票和上半部分人类标注者均值，仅略低于最佳单个人类标注者。论文还分析成本、速度与错误类型，主张LLM可用于大规模语言学标注辅助，但需人类批判性监督，而非简单替代。

Details

Motivation: 关注在语言复杂、语义要求高的标注任务中，LLM能否可靠替代或补充人类研究助理，并评估其实际可用性与局限性。

Result: 在174条真实“they”用例上，最佳LLM Kimi K3达到94.5%准确率，高于人类多数投票86.2%和上半部分人类标注者均值78.5%，仅最佳单个人类标注者95.7%更高；LLM在成本和速度上有明显优势。定性错误分析显示低一致性案例中人类推理能暴露标注类型学的不一致，体现人机互补性。

Insight: 摘要宣称的创新包括面向语义复杂任务的基准式人机比较、透明的成本与时间证据，以及反对纯替代式自动化的方法论论证。客观看，其可借鉴点在于将LLM评估与标注框架反思结合，并强调低一致性样本对改进标注类型学和人类监督的价值。

Abstract: This study evaluates whether large language models (LLMs) can reliably replace or complement human research assistants in linguistically complex annotation. In a controlled annotation task, 73 human annotators (local students and crowdworkers with and without a background in linguistics) and more than 40 LLM configurations number-disambiguated 174 authentic uses of ‘they’ from Reddit, including innovative and context-sensitive cases. The best LLM, Kimi K3, reached 94.5% accuracy, above the human majority vote (86.2%) and the mean of the top half of human annotators (78.5%); only the single best human annotator scored higher (95.7%). LLMs also offer substantial practical advantages in cost and speed. However, qualitative error analysis reveals a crucial complementarity: human reasoning in low-agreement cases exposes inconsistencies in the annotation typology and thereby improves the methodological framework itself. The study contributes (i) a benchmark-style human-LLM comparison for a semantically demanding task, (ii) transparent cost and time evidence, and (iii) a methodological argument against a purely replacement-oriented view of automation. LLMs are viable for large-scale annotation support in linguistics, but their responsible use requires human critical oversight. The dataset with its gold-standard labels, all human and LLM annotations, and the fully reproducible code are openly available as a Quarto-based, public repository.

[16] When Self-Verification Hurts: Calibrated Answer Repair for Financial Question Answering cs.CLPDF

Celine Guo, Weida Liang, Yujian Fu, Anqi Dang, Yanzheng Jin

TL;DR: 论文研究金融问答中模型即使抽取到看似合理的数字仍可能因实体、期间、单位或会计概念错配而出错的问题。作者将答案修复建模为“可能恢复信息”与“可能破坏正确内容”之间的权衡决策，提出校准答案修复方法 CAR：先判断草稿是否值得修复，再选择针对性检查、定位相关证据或草稿片段，并限制允许的编辑。

Details

Motivation: 自验证/开放式修订可能改坏原本正确的内容，或检查错误关系并强化原错误；因此需要一种更可控的答案修复机制，在提升答案正确性的同时降低对正确内容的损害。

Result: 在 FinRank、Fin-RATE 和 FinQA 上，使用三个开放权重模型时，CAR 在所有数据集-模型设置中均优于通用验证方法；系统层面可恢复 14.6% 初始遗漏的答案要求，仅损害 2.1% 初始正确内容，且后草稿阶段 token 成本比迭代修订方案低 30% 以上。

Insight: 主要创新在于将答案修复形式化为收益-损害权衡，并引入校准的修复决策、针对性检查、证据/草稿片段定位与有界编辑；消融表明证据定位和有界编辑是性能提升的主要贡献因素。

Abstract: Financial question answering can fail even when a model extracts a plausible number from the supplied evidence: the answer may attach it to the wrong entity, period, unit, or accounting concept. Asking the model to verify and revise its draft can help, but open-ended revision may alter content that was already correct or inspect the wrong relation and reinforce the original error. We formulate repair as a decision that balances the information it may recover against the correct content it may damage. Based on this trade-off, we introduce Calibrated Answer Repair (CAR), which first estimates whether a draft is likely to benefit from repair, then selects a targeted check, localizes the relevant evidence or draft span, and limits the permitted edit. Across FinRank, Fin-RATE, and FinQA with three open-weight models, CAR outperforms generic verification in all dataset-model settings. At the system level, it recovers 14.6% of initially missed answer requirements while damaging 2.1% of initially correct ones. Its post-draft token cost is over 30 percent lower than the iterative revision alternatives. Ablations on two dataset-model settings show that evidence localization and bounded editing are the main contributors to CAR’s gains.

[17] Beyond Accuracy: What Language-Subspace Interventions Capture, Remove, and Improve cs.CL | cs.AIPDF

Shaanil Punglia, Dadhichi Telwadkar, Vaibhav Rathore, Siddharth Sureka, Neha Bhargava

TL;DR: 论文系统评估多语言LLM中的语言子空间干预：在MATH和MGSM上，用Mean-SVD与DiffPCA构建语言子空间，检验其是否捕获翻译相关变化、投影移除后是否去除语言信息，以及在Qwen2.5-7B上不同干预配置对准确率和响应语言保真度的影响。结果表明子空间只捕获部分语言信息，最提升MGSM准确率的干预会牺牲语言保真度，仅一种干预使联合准确率小幅提升1.60点。

Details

Motivation: 语言子空间干预常被用于抑制中层激活中的语言相关成分以改善多语言推理，但现有评价多关注准确率，忽略了这些干预究竟捕获、移除和改进了什么，以及是否损害响应语言。

Result: 两个子空间均捕获大量翻译相关变化，但投影后线性探针仍几乎完美识别语言，说明未穷尽移除语言信息；在Qwen2.5-7B上，最提升MGSM准确率的干预牺牲响应语言保真度，仅一种干预使联合准确率提升1.60点，且仅改提示也可复现准确率增益但显著损害语言保真度。

Insight: 创新点在于从“捕获什么、移除什么、改进什么”三方面审计语言子空间干预，并强调因果干预不能只看准确率，还需纳入语言替换权衡；客观看，其对现有子空间方法的局限性和评价盲点进行了实证揭示。

Abstract: Language-subspace interventions aim to improve multilingual reasoning by suppressing language-related components in mid-layer activations. Using MATH and MGSM across four open-weight LLMs, we ask three questions: do estimated language subspaces capture translation-associated variation? Does projecting them out remove language information? And do they improve accuracy without compromising response language? For the first question, we build subspaces using the existing Mean-SVD construction and our DiffPCA. Both capture substantial translation-associated variation and show similar patterns across model depth. For the second, we project out each subspace and train linear probes to classify language from the projected activations. The probes still identify language almost perfectly, showing that neither subspace exhaustively captures language information. For the third, we evaluate several intervention configurations on Qwen2.5-7B, by varying subspace creation methods and intervention operators. We find that the intervention that most improves MGSM accuracy sacrifices response-language fidelity, with only one intervention improving joint accuracy–answers that are correct and in the target language–slightly, by 1.60 points. Interestingly, subtle changes in the prompt alone can reproduce the accuracy gains albeit with significant damage to language fidelity. We argue that causal interventions must be judged on more than accuracy, to fully account for language substitution trade-offs.

[18] MDSKILL: Learning to Internalize Skill Memory through Multi-View Distillation for Efficient Dynamic Agent Self-Evolution cs.CLPDF

Dian Jin, Jinluan Yang, Kairong Han, Baohong Li, Zijing Hu

TL;DR: 论文提出 MDSKILL，一种面向自进化智能体的多视角蒸馏方法，用于将技能记忆内化到策略中。它将技能记忆组织为状态锚定、动作规划和执行控制三个互补功能视角，并用动态指标筛选高质量技能，构建视角特定与整合推理分支，再通过 On-Policy Self-Distillation 将互补经验迁移并巩固到共享策略中，从而提升强化学习效率和智能体持续自进化能力。

Details

Motivation: 现有方法用技能记忆作为文本提示辅助强化学习，并通过 On-Policy Self-Distillation 内化技能引导轨迹中的知识；但自生成记忆常含噪声监督以及功能纠缠、冗余或冲突知识，直接蒸馏难以有效内化，甚至在有限训练预算下导致性能下降。

Result: 论文在三个智能体基准上评估 MDSKILL，并与不依赖更强外部教师模型的 RL、技能增强和自蒸馏基线比较；在不增加额外计算开销的情况下，于 ALFWorld 和 WebShop 上分别将 SOTA 性能提升最多 7.1% 和 8.1%，消融和进化分析进一步验证了知识内化与策略改进效果。

Insight: 创新点在于从多视角组织技能记忆，结合动态质量筛选构建视角特定与整合推理分支，并利用 OPSD 跨分支蒸馏互补经验到共享策略；客观来看，其价值在于缓解自生成技能记忆的噪声、冗余和冲突问题，在无需外部强教师和额外计算开销的条件下实现高效技能内化与智能体自进化。

Abstract: Skill memories encode past experiences as textual prompts to improve reinforcement learning (RL) for self-evolving agents. Complementing RL, On-Policy Self-Distillation (OPSD) provides dense, fine-grained supervision to internalize knowledge from skill-guided trajectories and accelerate policy learning. However, self-generated memories often contain noisy supervision and functionally entangled, redundant, or conflicting knowledge, making direct distillation difficult. Under limited training budgets, these issues can hinder knowledge internalization and even degrade performance. To address this problem, we propose MDSKILL, a multi-view distillation approach for skill memory internalization. MDSKILL organizes skill memories into three complementary functional views (state grounding, action planning, and execution control) and selects high-quality skills using dynamic metrics to construct view-specific and integrated reasoning branches. OPSD then transfers and consolidates complementary experiential knowledge across these branches into a shared policy, enabling efficient skill internalization and continuous agent self-evolution. We evaluate MDSKILL on three agent benchmarks against RL, skill-augmented, and self-distillation baselines that do not rely on stronger external teacher models. Without additional computational overhead, MDSKILL improves upon SOTA performance by up to 7.1% and 8.1% on ALFWorld and WebShop. Extensive ablations and evolution analyses further validate its effectiveness in knowledge internalization and policy improvement. The code is available at https://anonymous.4open.science/r/MDSKILL-CF53.

[19] Suppressing Misaligned Behaviors in Language Models, Durably cs.CL | cs.AI | cs.LGPDF

Luxi He, Pengcheng Jiang, Jifan Zhang, Jiawei Han, Danqi Chen

TL;DR: 论文研究如何持久地抑制语言模型中的错位行为。作者构建了两个触发方式不同的模型生物（窄触发：仅在特定提示线索下作弊；宽触发：在不安全代码上微调后广泛产生有害回复），将事后行为抑制形式化为立即抑制、进一步训练下的持久性和通用能力保持三个评估维度，并对比SFT、GRPO和梯度上升遗忘。在此基础上提出PEARL（Prefix-Expanded Adversarial Reinforcement Learning），一种GRPO变体，通过缓存生物轨迹中中间状态采样的延续来增强rollouts，从而提升抑制的持久性。

Details

Motivation: 不良模型行为可能在训练任何阶段出现或重新出现，现有对齐方法往往只实现浅层抑制，行为在后续训练中容易复发。论文旨在研究一旦观察到不良行为，如何有效且持久地将其抑制，同时不损害模型的通用能力。

Result: 在两种模型生物设置和两个模型家族（Qwen3-4B、GPT-OSS-20B）上，PEARL实现了较强的立即抑制并保持任务准确率，且在生物数据或无关数据上进一步微调后仍保持持久；当不良行为分布广泛、无法通过定向SFT解决时，PEARL收益最大。

Insight: 论文的创新点在于：一是构建窄触发和宽触发两类模型生物，系统性地刻画行为抑制问题；二是提出立即抑制、持久性和能力保持三维评估框架；三是基于对齐浅层性假设，认为持久抑制需要模型能从更广泛的错位状态中恢复，并据此设计PEARL，利用缓存轨迹中间状态的延续增强rollouts，使抑制更深入、更持久。

Abstract: Undesired model behaviors can emerge or re-emerge at any stage of training. Once observed, how effectively can we suppress them? To study this question, we first construct two model organisms–language models deliberately trained to exhibit a specific misalignment–that differ in what triggers the behavior. The narrow-trigger organism misbehaves only when the prompt contains a specific cue: it cheats on multiple-choice questions by following a planted hint that points to a wrong answer, and behaves normally otherwise. The broad-trigger organism has no such cue: after fine-tuning on insecure code, it gives harmful responses across a wide range of open-ended prompts. We then formalize post-hoc behavioral suppression and evaluate methods along three criteria: immediate suppression of the target behavior, durability under further training, and preservation of general capability. We compare standard methods–SFT, GRPO, and gradient-ascent unlearning–along these dimensions. Building on prior work on alignment shallowness, we hypothesize that more durable suppression requires the ability to recover from a wider range of misalignment states. We introduce Prefix-Expanded Adversarial Reinforcement Learning (PEARL), a GRPO variant that augments rollouts with continuations from intermediate states sampled along cached organism trajectories. Across both organism settings and two model families (Qwen3-4B, GPT-OSS-20B), PEARL achieves strong immediate suppression while preserving task accuracy, and remains durable under further fine-tuning on either the organism’s data or unrelated data. It obtains the greatest benefit when the undesirable behavior is broad and cannot be addressed through targeted SFT. Together, our framework and method offer a step toward more durable post-hoc removal of undesired behaviors.

cs.CV [Back]

[20] Physics-Informed Functional Tucker Method with RKHS Factors for Sparse Spatiotemporal Reconstruction cs.CV | cs.AIPDF

Zhengyang Liu, Jianguo Huang, Junxi Hu, Yucheng Mi, Haojie Cai

TL;DR: 论文提出 PhysFTM，一种物理信息函数化 Tucker 模型，用于从固定稀疏传感器和缺失帧中重建时空物理场。该方法在 RKHS 中参数化 Tucker 模态因子，并在传感器外的配点上通过有限差分残差引入 PDE 约束；先在支撑帧上交替优化 Tucker 核心张量与 RKHS 因子以重建连续空间场，再在 Tucker 核心空间构造连续轨迹以解码缺失帧。实验在 Allen-Cahn、Navier-Stokes、三维热方程和 Copernicus 海洋再分析数据上验证了 2D/3D 稀疏时空重建效果。

Details

Motivation: 稀疏传感器和缺失帧导致物理场大区域缺乏约束，逆问题病态，模型可能拟合观测点却产生不准确的传感器外结构或帧间预测。论文旨在利用 PDE 物理约束补充无直接观测区域的监督信息，提升稀疏时空重建可靠性。

Result: 在 Allen-Cahn、Navier-Stokes、三维 Heat 方程和三维网格化 Copernicus 海洋再分析数据上，PhysFTM 在两种缺失帧协议下均实现了准确的 2D/3D 重建；摘要未给出具体数值或明确 SOTA 对比。

Insight: 创新点在于将 Tucker 模态因子参数化到 RKHS 中，并在传感器外配点用有限差分 PDE 残差提供物理约束；同时通过在 Tucker 核心空间构造连续轨迹来实现缺失帧重建，将空间表示学习与时间补全解耦。

Abstract: Reconstructing physical dynamics from fixed sparse sensors and missing frames is a fundamental challenge in scientific research. Sparsity leaves large regions of the field unconstrained, making the inverse problem ill-posed: a model may fit the observed sensors while producing inaccurate off-sensor structures or unreliable predictions between support frames. To this end, we propose PhysFTM—a physics-informed functional Tucker model that parameterizes the Tucker mode factors in reproducing kernel Hilbert spaces (RKHSs) and incorporates PDE constraints through finite-difference residuals evaluated at off-sensor collocation points. These residuals provide additional physical information in regions without direct observational supervision. PhysFTM first reconstructs continuous spatial fields at support frames via an alternating minimization scheme, optimizing the Tucker core tensor and the RKHS-based factors separately. To enable missing-frame reconstruction, PhysFTM constructs a continuous trajectory in Tucker-core space, which is decoded by the learned spatial RKHS-FTM representation to recover the full spatiotemporal field. Experiments on Allen–Cahn, Navier–Stokes, three-dimensional Heat equations, and a three-dimensional gridded Copernicus ocean reanalysis demonstrate accurate 2D/3D reconstruction under two missing-frame protocols. Code is available at https://github.com/liusirzzz/PhysFTM.

[21] Beyond Proposals: Region-Consistent Open-Set Object Detection cs.CVPDF

Jiehui Tang, Yunlong Yu, Zhao Wang, Jungong Han

TL;DR: 论文针对开放集目标检测中密集候选框导致同一物体重叠proposal被赋予矛盾的已知/未知预测问题，提出OCSI-Net，将OSOD重构为以对象为中心的区域级语义推理问题。该方法聚合相关proposal为结构化区域簇并进行区域级开放集语义推断，同时引入几何感知不确定性机制稳定开放集边界附近的优化，从而获得一致且稳定的检测结果。

Details

Motivation: 现有开放集目标检测方法对每个proposal独立进行开放集推理，忽略了高度重叠且语义相关的proposal之间的共享信息，导致同一物体区域出现碎片化检测和不稳定的已知/未知判断。论文旨在解决密集proposal机制下开放集决策不一致的问题。

Result: 在多个benchmark上的实验表明，OCSI-Net一致达到state-of-the-art性能，已知和未知检测的harmonic mean相比第二优方法平均提升6.83%。

Insight: 论文的核心创新是将开放集目标检测从proposal级独立推理转变为以对象为中心的区域级语义一致决策，通过结构化区域簇聚合语义证据，并利用几何感知不确定性机制稳定开放集过渡边界。客观来看，其可借鉴之处在于显式建模相关proposal的语义一致性，以缓解密集候选框带来的矛盾预测和开放集决策不稳定问题。

Abstract: Open-set object detection (OSOD) suffers from a fundamental limitation under dense proposal mechanisms: highly overlapping proposals associated with the same object are often assigned contradictory known/unknown predictions. Existing methods perform open-set reasoning independently for each proposal, while ignoring the shared semantics among correlated proposals, resulting in fragmented detections and unstable open-set decisions. In this paper, we reformulate OSOD as an object-centric semantic inference problem, where correlated proposals jointly support a coherent region-level semantic decision. Based on this perspective, we propose OCSI-Net, a unified framework that combines object-centric semantic consistency with geometry-aware uncertainty stabilization. Specifically, OCSI-Net aggregates correlated proposals into structured region clusters and infers region-level semantics from aggregated semantic evidence, thereby enforcing coherent open-set predictions within the same object region. Furthermore, a geometry-aware uncertainty mechanism is introduced to stabilize optimization near open-set transition boundaries. Extensive experiments on multiple benchmarks demonstrate that OCSI-Net consistently achieves state-of-the-art performance, improving the harmonic mean of known and unknown detection by an average of 6.83% over the second-best method.

[22] FlowBench: Counterfactual Imagination Training for Spatial Intelligence cs.CVPDF

Yifan Liu, Baihan Zhang, Khai Loong Aw, Klemen Kotar, Daniel L. K. Yamins

TL;DR: 论文将「假设运动下想象场景变化」的能力形式化为反事实空间想象（CSI），并构建了 FlowBench：一个含 42.8 万训练样例的数据集与评测基准，覆盖仿真与真实场景、家庭环境与桌面操作等多类问题。作者设计了可扩展的全自动流水线，为每个假设运动配对空间问题，并在仿真中提供精确的结果帧和光流作为监督。FlowBench 同时用于训练和评测 VLM 与世界模型，实验显示微调后的 VLM 和世界模型均有显著提升。

Details

Motivation: 人类能够在假设运动下想象场景如何变化，这种能力是空间推理和机器人规划的核心，但在视觉语言模型和世界模型中尚未被充分探索。论文旨在填补这一空白，解决该能力缺乏系统性定义、训练数据和评测基准的问题。

Result: 专有 VLM 存在「想象差距」，即推断运动结果时表现不如直接观察；开源模型接近随机水平。Qwen3-VL-8B 在 FlowBench-Sim-Train 上微调后，仿真场景准确率提升 29.5 个百分点，真实图像提升 16.9 个百分点，sim-to-real 迁移效果良好；世界模型 Psi-0.6 的文本到光流预测准确率从 52.1% 提升至 61.7%，四个模型家族均有提升。

Insight: 创新在于将反事实空间想象定义为一个明确任务，并利用仿真提供的精确结果帧与光流作为可扩展的自动监督信号，构建统一的数据与评测资源，同时服务于 VLM 和世界模型两条技术路线。其全自动数据生成流水线和 sim-to-real 验证思路具有较强的可复用性。

Abstract: Humans imagine scene changes under hypothetical motions, an ability central to spatial reasoning and robotic planning. However, this capability remains underexplored in vision-language models and world models. We formalize this capability as Counterfactual Spatial Imagination (CSI) and introduce FlowBench, a resource for training and evaluating both, with 428K training examples spanning simulated and real-world scenes, household environments and tabletop manipulation, and diverse question types. A scalable, fully automatic pipeline pairs each hypothetical motion with spatial questions and, in simulation, exact outcome frames and optical flow. For VLMs, proprietary models exhibit an imagination gap, performing worse when motion outcomes must be inferred than when observed, and open-source models perform near chance; fine-tuning Qwen3-VL-8B on FlowBench-Sim-Train raises its accuracy by 29.5 percentage points in simulation and 16.9 on real images, a strong sim-to-real transfer. For world models, training on FlowBench-Sim improves text-to-flow prediction for all four model families we test, raising the accuracy of Psi-0.6 from 52.1% to 61.7%. FlowBench thus serves as a training and evaluation resource for both. Code and datasets are released at https://github.com/liuyifan22/FlowBench.

[23] SurgEye: Instruction Tuning-Free Perception-Reasoning Framework for Surgical Scene Understanding cs.CVPDF

Qing Xu, Xiechen Shen, Xianjian He, Guqiao Liang, Zixiao Chai

TL;DR: 论文提出SurgEye，一个无需指令微调的手术场景感知-推理框架。它通过路由式多任务感知RMP从视觉基础模型中提取空间与时间证据，再利用冻结大语言模型驱动的智能体组合推理ACR，将RMP专家和时间记忆作为可调用工具，组合跨任务、跨时间证据生成有依据的答案。作者还构建了覆盖八种手术流程、带感知标签可验证评估的SurgEye-Bench，并声称在手术推理上优于现有SOTA。

Details

Motivation: 现有手术视觉语言模型依赖大规模问答对进行指令微调，数据整理成本高且重新训练代价大；论文希望免指令微调地实现器械、解剖结构和工作流层面的组合推理。

Result: 大量实验表明SurgEye在手术推理任务上优于现有最先进方法，并在自建的SurgEye-Bench（涵盖八种手术流程、每题配有感知标签）上进行可验证评估；摘要未给出具体定量数值。

Insight: 创新点在于用任务特定路径路由视觉基础模型以产生多任务时空证据，并用冻结LLM将感知专家与时间记忆组织为可调用工具进行组合推理，从而避免指令微调；同时构建带感知标签的可验证手术推理基准，提升评估可信度。

Abstract: Surgical scene understanding is pivotal for computer-assisted intervention, yet demands compositional reasoning across instruments, anatomy, and workflow. Existing surgical vision-language models rely on instruction tuning with large-scale question-answer pairs, entailing laborious data curation and costly retraining. To bridge this gap, we propose SurgEye, a perception-reasoning framework that grounds reasoning in a vision foundation model without instruction tuning. Specifically, we devise Routed Multi-Task Perception (RMP) that routes the vision foundation model through task-specific pathways learned from public perception labels, producing spatial and temporal evidence. We further introduce Agentic Compositional Reasoning (ACR) that employs a frozen large language model to coordinate RMP specialists and temporal memory as callable tools, composing evidence across tasks and time into grounded answers. Moreover, we construct SurgEye-Bench spanning eight surgical procedures, pairing each question with perception labels for verifiable evaluation. Extensive experiments demonstrate that SurgEye outperforms state-of-the-arts in surgical reasoning. The source code will be available.

[24] Compression Forcing: Hierarchical History Tokenization Enables Flexible World Models cs.CVPDF

Siqiao Huang, Partha Kaushik, Michael Chen, Hengkai Pan, Kaiwen Geng

TL;DR: 论文提出 Compression Forcing（CF）通用世界模型框架，将历史观测压缩与世界建模的预测目标对齐。其核心是 Dynamics-Aware Tokenization（DAT），按动态显著性递减组织历史 token，并结合扩散骨干实现 Hierarchical History Guidance（HG-H），在推理时选择性放大动态显著信息。实验表明该方法能提升预测保真度、物理真实性和下游决策表现。

Details

Motivation: 现有世界模型多沿用视频生成的 tokenization 和条件机制，更关注感知真实而非预测效用；在视觉干扰和任务无关背景占主导的环境中，这会限制世界模型效果。论文旨在让历史压缩直接服务于未来预测。

Result: 摘要给出定性结果：在多种环境中，Compression Forcing 提升了预测保真度、物理真实性和下游决策性能，并宣称其是一种灵活且有效的世界建模方法；未提及具体 benchmark、量化指标或 SOTA 对比。

Insight: 创新点在于将历史压缩与世界模型的预测目标显式对齐，提出按动态显著性排序的层次化历史 tokenization，并在扩散骨干上通过推理时分层引导选择性增强显著动态信息。客观看，其可借鉴之处是将 token 组织方式与预测效用结合，并用层次化条件机制提升对视觉干扰的鲁棒性和世界模型的灵活性。

Abstract: Effective world modeling requires compressing salient history information that facilitates future prediction. However, most existing approaches inherit tokenization and conditioning mechanisms from video generation, where objectives are centered around perceptual realism rather than predictive utility. This discrepancy limits the world model’s effectiveness, especially in environments dominated by visual nuisances or task-irrelevant background details. To this end, we introduce Compression Forcing (CF), a general framework that aligns history compression with the predictive objective of world modeling. Compression Forcing features Dynamics-Aware Tokenization (DAT), a novel tokenization method that compresses history observations into a sequence ordered by decreasing dynamics saliency. Built on a diffusion backbone, this structure enables Hierarchical History Guidance (HG-H), allowing selective amplification of dynamics-salient information during inference. Extensive experiments across various environments demonstrate improved prediction fidelity, physical realism, and downstream decision-making performance, establishing Compression Forcing as a highly flexible and effective approach for world modeling.

[25] Do Pretrained Backbones Transfer to Physical Sensors? A Systematic Study of Multimodal Initialization Across the Sensor Stack cs.CVPDF

Fabio Montello, Ronja Güldenring, Lucas Mogensen Sandby, Christoffer Koo Øhrstrøm, Lazaros Nalpantidis

TL;DR: 论文系统研究了预训练骨干网络向真实物理传感器多模态场景的迁移问题。作者从16个公开自动驾驶数据集中构建了包含282万图像的多传感器数据集，涵盖RGB、LWIR、NIR、Event等物理模态以及Stereo Depth、Segmentation等模型派生模态，并基于MultiMAE进行域内预训练。通过适配稀疏样本损失、研究编码器瓶颈和SSIM/FFT稠密重建损失，论文在多个下游传感器融合任务上比较了ViT、MAE、DINO和MultiMAE的迁移效果。

Details

Motivation: 论文关注两个开放问题：哪些预训练权重最适合多传感器设置，以及域内预训练能否优于现成预训练模型。现有MultiMAE等方法依赖稠密、模型派生模态，尚不清楚其在真实物理传感器和模态稀疏场景下的鲁棒性。

Result: 论文构建了2.82M图像、16个公开数据集的多传感器自动驾驶集合，并在下游传感器融合任务上系统比较。结果显示MultiMAE是出乎意料的强即用基线，甚至能有效迁移到未见过的模态；作者的目标性修改在多传感器迁移和整体鲁棒性上带来增量但一致的提升。

Insight: 创新点在于面向真实物理传感器栈系统评估预训练迁移，构建大规模多传感器域内预训练数据，并针对模态稀疏适配MultiMAE损失、分析编码器瓶颈及稠密重建损失的作用。客观看，其价值更偏向系统性实证与数据/方法适配，而非全新架构。

Abstract: Transfer learning from pretrained backbones has become standard practice in multi-sensor computer vision—yet two questions remain open: which pretrained weights transfer best to multi-sensor setups, and can in-domain pretraining outperform off-the-shelf solutions? Frameworks like MultiMAE demonstrate the utility of masked autoencoding in a multi-modal context, but rely entirely on dense, model-derived modalities. It is unclear how robustly these methods translate to real-world applications that use physical sensors, where the aggregation of diverse public datasets for large-scale pretraining inherently introduces modality sparsity. To investigate this systematically, we curate a multi-sensor autonomous driving collection of $2.82M$ images from 16 publicly available datasets, incorporating both physical hardware modalities (RGB, LWIR, NIR, Event) and model-derived modalities (Stereo Depth, Segmentation). We use this collection for in-domain pretraining based on the MultiMAE approach. We adapt MultiMAE’s loss formulation to accommodate sparse samples, and study the effect of an enforced encoder bottleneck for modality alignment and of additional losses specific to dense reconstruction (SSIM and FFT). We then systematically compare against common pretrained backbones (ViT, MAE, DINO and MultiMAE) across downstream sensor fusion tasks. We find that MultiMAE is a surprisingly strong out-of-the-box baseline, transferring effectively even to modalities it has never seen, while our targeted modifications yield incremental but consistent improvements on multi-sensor transfer and overall robustness.

[26] From Visual Matching to Textual Conditioning: Cross-Domain Few-Shot Segmentation via Reference Inversion cs.CVPDF

Jiahao Nie, Yun Xing, Wenbin An, Qixin Zhang, Yap-Peng Tan

TL;DR: 论文提出 RI-SAM3，用于跨域小样本分割（CD-FSS）。该方法利用 SAM3 的文本条件分割能力，通过参考反演为每个目标类别学习合适的文本 token，替代对目标外观敏感的参考-目标视觉匹配。RI-SAM3 包含双层参考反演、层自适应文本条件和存在感知置信监督三个设计。冻结 SAM3 时，每类仅优化约 12K 参数，单次前向即可完成分割且文本条件可复用。

Details

Motivation: 现有基于 SAM 的 CD-FSS 方法依赖参考-目标视觉匹配，对目标外观变化高度敏感；而 SAM3 直接使用类别名作为文本条件在严重域偏移下性能不稳定。论文希望学习更合适的类别文本条件，以提升跨域小样本分割的鲁棒性。

Result: 论文称在多个广泛采用的 CD-FSS 基准上取得一致更优性能，并表现出较强的鲁棒性、效率和泛化能力；摘要未给出具体数值或明确 SOTA 表述。

Insight: 创新点在于将 CD-FSS 从逐样本视觉匹配转向可复用的文本条件：双层参考反演学习文本编码器前后互补 token，层自适应文本条件按解码层组合 token，存在感知置信监督联合约束查询置信与目标存在。客观亮点是冻结 SAM3、仅调极少参数并实现单次前向高效分割。

Abstract: Cross-Domain Few-Shot Segmentation (CD-FSS) aims to segment objects from unseen domains using only a few annotated references. Recent Segment Anything Model (SAM)-based methods have advanced CD-FSS, yet their reference-target visual matching is highly sensitive to target appearance variations. SAM3 introduces text-conditioned segmentation, enabling natural-language descriptions to guide segmentation across target images, but directly using the category name as the condition still yields unstable performance under severe domain shifts. We propose Reference Inversion for SAM3 (RI-SAM3) that leverages SAM3’s large-scale vision-language pre-training to learn suitable textual tokens for each target category. RI-SAM3 comprises three core designs: (i) Bi-level Reference Inversion that learns complementary tokens before and after the frozen text encoder: the former remain anchored in SAM3’s text space while the latter represent the category more flexibly; (ii) Layer-adaptive Textual Conditioning combines these tokens differently as object queries evolve across decoder layers; and (iii) Presence-aware Confidence Supervision that jointly supervises query confidence and target presence for reliable predictions. With SAM3 frozen, RI-SAM3 optimizes around 12K parameters only per category and it efficiently segments each target in a single forward pass with reusable textual conditioning without per-sample reference-target visual matching. Extensive experiments over multiple widely adopted benchmarks show that RI-SAM3 achieves superior performance consistently in CD-FSS tasks, together with strong robustness, efficiency, and generalization.

[27] ScienceVision: Can AI Observe and Reason Like a Scientist? cs.CV | cs.AIPDF

Tianyi Ma, Jing Gu, Haozhe Zhao, Liang Chen, Yunfan Zhang

TL;DR: 论文提出 ScienceVision，一个跨学科的科学 3D 理解与推理基准，通过专家级视觉问答评估 AI 是否能像科学家一样观察和推理。基准包含 250 道题，覆盖生物、化学、医学影像、机械 CAD 和建筑五个领域，每题结合原生 3D 数据、科学问题、指定回答格式和参考答案。论文在三种设置下评测 12 个模型：直接通过 agent 工具访问原生 3D 文件、增加基于 MCP 的专家式查看/测量工具、以及基于预设渲染 2D 图像答题。

Details

Motivation: 现有 multimodal benchmark 多集中于文本和 2D 图像，科学 3D 数据评测在表示和协议上跨学科不统一，难以比较模型是否具备从 3D 结构、空间关系到领域知识的科学观察与推理能力。

Result: 主实验中最佳模型平均得分仅 0.53/1；在无额外 MCP 工具的原生 3D 工件工作流中，所有模型均优于渲染 2D 图像工作流，平均高出 17.2 个百分点。

Insight: 创新点在于构建跨五大学科、专家审核的科学 3D VQA 基准，并系统比较原生 3D 访问、MCP 专家式查看/测量工具与渲染 2D 图像三种交互范式，强调几何/拓扑证据与领域知识结合的科学推理评测。

Abstract: Scientific discovery builds on accurate observations interpreted through domain knowledge and reasoning. For scientific 3D data, this requires understanding structures and spatial relationships, from protein contacts to anatomical organization. Existing multimodal benchmarks largely evaluate text and 2D images, while scientific 3D evaluations vary in representations and protocols across disciplines, limiting cross-disciplinary comparison. To assess these foundational capabilities, we introduce ScienceVision, a cross-domain benchmark for scientific 3D understanding and reasoning through expert-level visual question answering (VQA). We construct 250 questions, with 50 each from Biology, Chemistry, Medical Imaging, Mechanical CAD, and Architecture, through a pipeline combining synthetic question generation with domain-expert review and annotation. Each task pairs a native 3D artifact with a scientific question, a specified response format, and a task-specific reference answer, requiring domain knowledge to be applied to geometric or topological evidence. We evaluate 12 models on the same questions in three settings: visual question answering with native 3D artifacts accessed through an agent harness’s file and code-execution tools; the same workflow augmented with expert-inspired viewer inspection and measurement tools exposed through the Model Context Protocol (MCP); and visual question answering from prescribed rendered 2D images using an agent harness. All settings use the same task-specific deterministic final-answer scoring. The inspection environment supports view changes, target isolation, and slice inspection. In this main comparison, the best mean answer score is 0.53 out of 1. Mean scores in the native-artifact workflow without additional MCP tools exceed those in the rendered-image workflow for every model, by 17.2 percentage points on average.

[28] Generalizable AI-Generated Image Detection with Embedding-Space Explanations cs.CV | cs.AI | cs.MMPDF

Nikos Giakoumoglou, Paschalis Giakoumoglou, Tania Stathaki, Symeon Papadopoulos

TL;DR: 论文提出FOX，用于可泛化且可解释的AI生成图像检测。它先构建并锁定一个基于DINOv3的专家检测器，再用其视觉表示引导轻量级VLM生成取证解释，并通过联合嵌入预测器将锁定视觉证据映射到冻结文本编码器的语义空间。这样解释训练不会改变检测判决，同时兼顾跨生成器泛化和自然语言解释。

Details

Motivation: AI生成图像日益逼真，指令驱动编辑器可局部重合成且无拼接边界，现有检测器多为黑箱、判决不可验证，并因记忆特定生成器指纹而泛化差。纯视觉专家检测器（如DINOv3）泛化较好，但可解释方法联合训练分类与语言生成会损害视觉表示，因此需要保证解释训练不能改变检测判决。

Result: FOX在14个未见生成器上取得97.4平均AUC，在XPlainVerse上BERTScore为69.8，在LOKI上为56.1。

Insight: 创新点在于解耦检测与解释：锁定专家检测器以保留判决和泛化能力，再用其表示引导VLM，并通过联合嵌入预测器对齐冻结文本编码器语义空间来生成解释。客观看，可借鉴之处是让解释模块不反向影响检测表示，从而在可解释性与跨生成器泛化之间取得平衡。

Abstract: AI-generated images have reached unprecedented realism, while instruction-driven AI editors re-synthesize images for local changes, leaving no splice boundary to localize. Most detectors are black boxes with unverifiable verdicts and generalize poorly because adapting a backbone on the detection distribution memorizes generator-specific fingerprints that quickly become obsolete. Specialist detectors built on pure-vision representations, particularly DINOv3, generalize well across generators, but explainable methods forgo this advantage: jointly training one multimodal model for classification and language generation subjects its visual representation to the language objective. We therefore require that explanation training cannot alter the verdict. We introduce FOX, which first builds and locks a specialist DINOv3 detector, then uses its representation to guide a lightweight VLM toward forensic explanation. A joint-embedding predictor maps locked visual evidence into a frozen text encoder’s semantic space, producing guidance that the VLM verbalizes on demand. FOX achieves 97.4 mean AUC across 14 unseen generators, 69.8 BERTScore on XPlainVerse, and 56.1 on LOKI.

[29] Co-Occurrence Priors Override Visual Spatial Evidence: Understanding False Alarms in Vision-Language Models for Construction Struck-By Hazard Detection cs.CV | cs.LGPDF

Xiaoshan Zhou

TL;DR: 论文研究视觉语言模型（VLM）在建筑施工撞击（struck-by）危险检测中频繁误报的成因。作者用355组空间接近度受控的工人-挖掘机反事实场景对，对Qwen2.5-VL-7B和InternVL3.5-8B进行行为分析、内部表征探测、因果干预与决策校正，发现模型能正确识别实体并编码空间关系，但危险判断被“工人-挖掘机共现即危险”的先验主导，误报主要源于决策环节而非感知环节。

Details

Motivation: VLM越来越多用于施工危险监测，但误报严重且机理不清。论文要厘清误报究竟来自目标识别失败、空间关系编码失败，还是共现先验主导的危险决策。

Result: 在被控空间距离场景上，两模型对实体识别与空间关系编码的out-of-fold AUROC分别为0.82和0.91，且在79%和73%的场景对中给近距离场景更高的危险分；但仍有74%和100%的远距离场景被误判为危险。移除工人后Qwen的危险判定率降至11%，工人存在对危险分的影响是空间接近度的2.4倍和3.2倍。注意力头、神经元级干预和激活引导均无法消除共现偏差，而将危险决策条件化于模型自身空间关系表征后，Qwen在等敏感度下特异度从0.36提升到0.58，优于阈值重校准。

Insight: 创新点在于用反事实场景对结合行为、表征探测、因果干预和决策校正，系统定位VLM危险检测误报的根因，并提出“以已编码的空间证据约束决策”的思路。其可借鉴之处是：模型感知正确不等于决策正确，应针对决策层做条件化与对齐，而非一味提升视觉识别能力。

Abstract: Vision-language models (VLMs) increasingly support construction hazard monitoring, yet their frequent false alarms remain poorly understood. This study investigates whether false struck-by hazard warnings arise from failures in object recognition, spatial relation encoding, or hazard decisions dominated by learned co-occurrence priors. We evaluated Qwen2.5-VL-7B and InternVL3.5-8B using 355 counterfactual worker-excavator scene pairs with controlled spatial proximity and worker-removal variants, combining behavioral analysis, internal representation probing, causal interventions, and decision correction. Both models recognized the entities and encoded their spatial relations (out-of-fold AUROC: 0.82 and 0.91), assigning higher hazard scores to near than far scenes in 79% and 73% of pairs. Nevertheless, 74% and 100% of far scenes were falsely classified as hazardous. Removing the worker reduced Qwen’s hazard rate to 11%, while worker presence exerted 2.4- and 3.2-times greater influence on hazard scores than spatial proximity in the respective models. Attention-head interventions, neuron-level interventions, and activation steering failed to eliminate this co-occurrence bias. However, conditioning hazard decisions on the model’s own spatial-relation representation improved Qwen’s specificity from 0.36 to 0.58 at matched sensitivity, outperforming threshold recalibration. These findings reveal that VLM struck-by false alarms stem primarily from decision-making failures rather than deficiencies in object recognition or spatial encoding, suggesting that future efforts should prioritize conditioning hazard decisions on encoded spatial evidence rather than further improving visual recognition alone.

[30] What Remains in Sight? Autoregressive Video Decoding as Representation-Guided Context Rewriting cs.CV | cs.AI | cs.LGPDF

Lixuan He, Jihyeon Je, Yang You, Leonidas Guibas

TL;DR: 论文将长时程自回归视频生成中的“哪些历史帧/片段应作为可见上下文输入解码器”建模为显式的测试时决策，提出免训练方法ReCR，将自回归视频解码表述为表示引导的上下文重写。ReCR在每个解码步利用解码器内部表示从已生成历史中选择可见上下文单元，并将其映射到紧凑的因果轴上以解码下一块。

Details

Motivation: 现有自回归视频生成多通过最近窗口、压缩KV缓存或外部记忆隐式决定可见过去，限制了长视频生成的稳定性；论文关注能否在推理阶段显式选择对解码器可见的历史内容。

Result: 在长时程文本到视频基准上，ReCR在匹配可见token预算下持续提升多个自回归骨干；摘要未给出具体数值或SOTA声明，并通过表示空间诊断和transition-count提示切换等分析验证有效性。

Insight: 创新点在于免训练地将上下文选择视为测试时决策，并利用解码器内部表示引导上下文重写，综合匹配全局rollout状态、避免与近期内容冗余、保留时间边界覆盖和扩大空间布局多样性，再映射到紧凑因果轴。客观看，其可借鉴之处是把“可见历史”从隐式缓存策略转为可解释、可优化的上下文选择机制。

Abstract: Long-horizon video generation is increasingly moving toward autoregressive rollout, where each newly generated segment becomes part of the evidence used for future prediction. Recent causal decoders, long-context training, and KV-cache mechanisms have extended feasible duration, but most inference pipelines still decide the visible past implicitly through recent windows, compressed caches, or external memory. This raises a direct question: \textbf{can the choice of which past frames or segments to show to the decoder be treated as an explicit test-time decision?} We answer this question with \textsc{ReCR}, a training-free method that formulates autoregressive video decoding as representation-guided context rewriting. At each step, \textsc{ReCR} builds a selected visible context from the generated history by using internal decoder representations to favor context units that match the global rollout state, avoid redundancy with the recent continuation, preserve temporal boundary coverage, and widen the range of spatial layouts in view. The selected units are then mapped onto a compact causal axis before decoding the next block. Across long-horizon text-to-video benchmarks, \textsc{ReCR} consistently improves multiple autoregressive backbones under matched visible-token budgets. Empirical analyses such as representation-space diagnostics and transition-count prompt switching further show that improving what remains visible is an effective path toward more stable long-video decoding.

[31] ReconDreamer#: Self-Evolving Closed-Loop Synergy of Reconstruction and Generation for Dynamic Driving Scenes cs.CVPDF

Songling Liu, Guosheng Zhao, Xiaofeng Wang, Chaojun Ni, Guan Huang

TL;DR: 论文提出ReconDreamer#，一个面向动态驾驶场景的自演化闭环框架，通过重建模型与生成模型迭代互增强，缩小新轨迹合成数据与真实传感器观测之间的差距，并结合几何先验自适应地面模型提升重建渲染质量。

Details

Motivation: 动态驾驶场景重建虽借助生成先验提升了新轨迹渲染质量，但生成数据与真实传感器观测仍存在明显差距，限制了重建保真度的进一步提升。

Result: 在Waymo、nuScenes、PandaSet和EUVS上进行了评估；在Waymo 6米横向偏移设置下，相较ReconDreamer++，NTA-IoU和NTL-IoU分别提升8.8%和3.0%，FID降低15.1%。

Insight: 核心创新是构建重建与生成相互迭代增强的自演化闭环，使生成的新轨迹数据更贴近真实观测并反哺3D表示能力；同时引入基于几何先验的自适应地面模型以改善渲染质量。

Abstract: Dynamic driving scene reconstruction has made remarkable strides in recent years by leveraging generative priors, particularly in the quality of novel trajectory rendering. Nevertheless, a significant gap remains between the generated data and real-world sensor observations, severely hindering further improvements in reconstruction fidelity. In this paper, we present ReconDreamer#, a self-evolving framework that synergizes reconstruction and generation for dynamic driving scenes, which substantially bridges the gap between synthesized novel trajectory data and recorded real-world observations. Within this self-evolving closed-loop framework, the generative and reconstruction models mutually reinforce each other through iterative refinement, ultimately yielding a significant enhancement in the 3D representation capability of the reconstruction model. Additionally, we introduce a geometry-prior-based adaptive ground model to further enhance the rendering quality of the reconstruction model. To validate the effectiveness of the proposed method, we conduct comprehensive evaluations on Waymo, nuScenes, PandaSet, and EUVS benchmarks. Specifically on Waymo, at a 6-m lateral shift, ReconDreamer# improves NTA-IoU and NTL-IoU by 8.8% and 3.0%, respectively, and reduces FID by 15.1% relative to ReconDreamer++.

[32] Reliability-Role Collaborative Learning for Semi-Supervised Cardiac Image Segmentation cs.CVPDF

Zhongrui Xie, Wenfeng Zhang, Wei Hu, Ruitao Xie, Wenjie Wang

TL;DR: 论文提出R2CL框架用于半监督心脏图像分割，通过建模双网络间置信度关系，将未标注区域分为相互不确定、置信度差异和相互置信三类，分别用于熵正则化、方向性知识迁移和可靠监督锚点。结合区域混合训练、原型对比学习和上下文细化，缓解伪标签监督可靠性空间不均匀问题。

Details

Motivation: 半监督分割中伪标签监督在空间区域上可靠性高度不均匀，现有方法对所有未标注像素施加统一监督，忽略边界、低置信度和网络分歧等噪声区域，限制了困难区域判别学习。

Result: 在多个数据集上区域重叠和边界相关指标取得稳定提升，在前景稀疏的MVAA-CT数据集上优势更明显，但摘要未给出具体数值和SOTA对比。

Insight: 创新点在于按双网络置信度关系为未标注区域分配不同角色进行协作学习，并引入可靠性过渡区域原型对比与高置信上下文细化。客观看，角色化监督和区域混合跨网络训练为半监督分割提供了更精细的可靠利用方式。

Abstract: Semi-supervised cardiac image segmentation aims to exploit abundant unlabeled images together with limited annotations, but the reliability of pseudo-label supervision is highly uneven across spatial regions. In a dual-network setting, interior regions are usually more stable, whereas boundary regions, low-confidence areas, and disagreements between networks can introduce noisy supervision. However, existing methods typically impose uniform supervision on all unlabeled pixels, overlooking reliability differences and limiting discriminative learning in challenging areas. To address this issue, we propose R2CL (Reliability-Role Collaborative Learning), a framework for semi-supervised cardiac image segmentation. R2CL models confidence relationships between two networks to assign roles to unlabeled regions: mutually uncertain regions are used for entropy regularization, confidence-discrepant regions for directional knowledge transfer, and mutually confident regions as reliable supervision anchors. Reliability-Role Mixed Cross Learning (RR-MCL) implements these roles through region-mixed training, directional cross-network supervision, and entropy regularization over mutually uncertain regions. Reliability-Transition Prototype Contrastive Learning (RT-PCL) constructs query features from foreground-adjacent reliability-transition regions and learns network-specific class prototypes from mutually confident regions, improving feature discrimination around ambiguous boundaries. High-confidence Context Refinement (HCR) exploits reliable interior pixels and trusted dense blocks within mutually confident regions to complement stable contextual supervision. R2CL achieves stable improvements in region-overlap and boundary-related metrics across multiple datasets, with more evident advantages on the foreground-sparse MVAA-CT dataset.

[33] CounterSketch: Physics-Grounded Counterfactual Reasoning for Sketch-to-Video Generation cs.CVPDF

Hien Trinh, Tan-Hiep To, Dinh-Khoi Vo, Tam V. Nguyen, Minh-Triet Tran

TL;DR: 论文提出 CounterSketch，一个基于物理约束的反事实草图动画框架，可将自然语言“如果……会怎样？”问题转化为物理上合理的草图视频。该框架结合 VLM 场景理解、因果事件链推理、基于约束的轨迹生成与验证，并将验证后的轨迹投影为对象级运动边界框以指导草图视频合成。

Details

Motivation: 现有草图到视频生成难以处理假设性干预下的反事实演化，且直接依赖 VLM 预测运动缺乏物理与几何合理性。论文旨在解决如何根据自然语言反事实查询生成物理可信的草图动画。

Result: 在 CounterSketchBench（80 个问题）上实验表明，基于物理的轨迹比直接 VLM 预测提供更强的运动引导。

Insight: 创新点在于将反事实推理、物理约束验证与草图视频生成结合，通过因果事件链和约束轨迹生成提升运动合理性，并利用对象级运动边界框连接高层推理与下游视频合成。

Abstract: Counterfactual sketch animation explores how a static sketch evolves under a hypothetical intervention. We present CounterSketch, a physics-grounded framework that transforms natural-language ``What if…?’’ queries into physically plausible sketch animations. The framework integrates VLM-guided scene grounding, causal event-chain reasoning, constraint-based trajectory generation, and verification. It models physical transitions to update the scene graph, computes continuous object trajectories, and validates them against physical, geometric, and semantic constraints. The verified trajectories are then projected into object-level motion bounding boxes to guide downstream sketch-video synthesis. Experiments on CounterSketchBench (80 questions) demonstrate that physics-grounded trajectories provide stronger motion guidance than direct VLM predictions.

[34] ProGeo-RC: Prototype-guided Semantic Reliability and Geometric Error Calibration for Semi-supervised Cardiac Image Segmentation cs.CVPDF

Yunfei Zhu, Wenfeng Zhang, Pan Zeng, Xin Huang, Tao Jia

TL;DR: 论文提出ProGeo-RC半监督医学图像分割框架，利用概率类原型评估无标注预测的语义可靠性并校准教师伪标签，同时引入基于符号距离场的几何约束来细化模糊边界并抑制目标外虚假前景。该方法在ACDC和MVAA-CT数据集上优于代表性半监督分割方法，表明联合语义可靠性估计与几何误差校准能提供更可靠的伪监督。

Details

Motivation: 现有半监督医学图像分割方法多依赖预测置信度或模型一致性估计伪标签可靠性，难以同时刻画语义可靠性和几何合理性，容易产生错误伪监督。

Result: 在ACDC和MVAA-CT上取得优于代表性半监督分割方法的表现；在ACDC 10%标注设置下，DSC和Jaccard分别比第二优结果提升0.37%和0.49%，95HD和ASD分别降低0.61和0.21。

Insight: 创新点在于用概率类原型估计无标注预测的语义可靠性以校准像素级教师伪标签，并结合符号距离场几何约束纠正模糊边界和目标外虚假前景。客观看，其可借鉴之处是将语义级可靠性评估与结构化几何先验联合用于伪标签纠错，而不仅依赖置信度或一致性。

Abstract: Semi-supervised medical image segmentation aims to train reliable segmentation models using limited labeled data and abundant unlabeled data. However, existing methods typically rely on prediction confidence or model consistency to estimate pseudo-label reliability, but these criteria are insufficient to characterize both semantic reliability and geometric plausibility. To address this issue, we propose ProGeo-RC, a prototype-guided semantic reliability and geometric error calibration framework for semi-supervised medical image segmentation. ProGeo-RC consists of two key modules: Prototype-guided Semantic Reliability Estimation (PSRE) and Structured Semantic-Geometric Error Calibration (S-SGEC). PSRE uses probabilistic class prototypes to estimate the semantic reliability of unlabeled predictions and calibrate pixel-wise teacher pseudo labels, while S-SGEC introduces signed-distance-field-based geometric constraints to refine ambiguous boundaries and suppress spurious foreground responses outside target structures. Experiments on the ACDC and MVAA-CT datasets show that ProGeo-RC achieves better performance compared with representative semi-supervised segmentation methods. Under the 10% labeled setting on ACDC, ProGeo-RC improves DSC and Jaccard by 0.37% and 0.49% over the corresponding second-best results, while reducing 95HD and ASD by 0.61 and 0.21, respectively. These results indicate that joint semantic reliability estimation and geometric error calibration can provide more reliable pseudo supervision for semi-supervised medical image segmentation.

[35] Sketch-Omni: Instance-Level Sketch-Based Visual Prompting for Perception MLLMs cs.CVPDF

Dongyu Wang, Dar-Yen Chen, Zhiming Hong, Arkaprabha Basu, Yi-Zhe Song

TL;DR: 论文提出 Sketch-Omni，这是首个面向草图（sketch）的感知型多模态大模型（MLLM）。它支持将草图作为图像进行检测、指向、指代解析与可供性定位，也支持将草图作为视觉提示，在照片中定位所画对象的具体实例。为克服照片训练模型缺乏草图纹理线索和缺少草图-照片实例配对数据的问题，作者用边缘检测自动生成草图式训练数据，并通过照片引导的关系对齐训练草图分支。

Details

Motivation: 现有可提示感知模型主要面向照片，缺乏对草图输入的支持；同时草图缺少颜色和纹理，且没有数据集将一张草图和照片中对应同类中的特定实例配对，导致草图作为实例级视觉提示难以实现。

Result: 在草图提示定位任务上达到 91.4 F1，显著优于最强通用 MLLM 的 81.8；当照片中存在多个同类实例时优势翻倍，并在六类任务共 40 个评测单元中的 39 个取得最佳结果。作者还将发布包含 2754 个草图-照片链接的人工标注 benchmark。

Insight: 创新点在于将草图同时作为图像输入和实例级视觉提示，并通过自动边缘检测生成保留标签的草图数据、构造草图-照片实例对，以及用照片引导的关系对齐让草图分支学习照片分支的短语-图像块关系；核心可借鉴之处是无需手工绘制训练标签的跨模态数据构造与对齐策略。

Abstract: Promptable perception, from GroundingDINO to Rex-Omni, puts one model behind one prompt interface and returns coordinates for any localisation task. Every model in that line was built for photographs; none has been given a sketch. We present Sketch-Omni, to our knowledge the first perception MLLM built to work on sketches and from them. On a sketch, the sketch is the image, and text prompts detect, point, resolve object and spatial referring expressions, and ground affordances in it. From a sketch, the sketch is the prompt, and a drawing of one object localises that object in a photograph, even among several of the same category, a query that neither a category name nor a photo exemplar can pose. Two obstacles stand in the way. A photo-trained model’s attention is tuned to colour and texture, which a sketch does not have, and no dataset pairs a drawing with the one instance it depicts among others of its kind. We build the training data instead of collecting it. An edge detector turns every labelled photograph into a sketch-like drawing that keeps the photograph’s labels, and two constructions produce sketch-to-photo instance pairs from existing data. On top of this, photo-guided relation alignment passes each photograph and its edge sketch through the same model and trains the sketch pass to reproduce how the photograph pass relates phrases to patches and patches to each other. Not one training label is drawn by hand. Sketch-Omni reaches 91.4 F1 on sketch-prompted localisation against 81.8 for the strongest general-purpose MLLM, with a lead that doubles when the photograph holds several instances of the queried category, and it is best in 39 of 40 evaluation cells across six tasks. Code, model, and a hand-annotated benchmark of 2754 sketch-to-photo links will be released.

[36] Weight-Space Programming for Improving Image-Restoration Diffusion Models: Technical Conjectures and a Recursive Improvement Loop cs.CVPDF

Zhuoran Zheng, Doubao

TL;DR: 论文探讨能否让LLM在没有梯度、配对数据和微调的情况下，直接编辑扩散式图像复原模型的权重以提升复原质量。作者提出将权重空间编程（WSP）从“调试结构错误”推进到“性能改进”，并给出三个技术猜想：LLM可在有限前向预算内定位质量差距所在层、稀疏权重方向编辑可提升PSNR和感知质量、诊断-编辑-评估-再诊断的递归循环可累积收益。论文属于猜想性工作，提出部署流程和基准设想，未报告实际实验结果。

Details

Motivation: 现有WSP实验表明LLM能通过修改极少参数修复结构bug，但难以赋予模型新能力；论文关注在图像复原这一精确回归场景中，WSP能否从调试走向真正提升去模糊、超分、去噪、修复等任务的表现。

Result: 摘要未报告定性或定量实验结果，也未给出benchmark上的SOTA比较；论文主要是技术猜想、能力分析、PyTorch风格部署流程和基准设计。

Insight: 创新点在于提出面向图像复原扩散UNet的权重空间编程框架：利用低层信号处理层较集中、适合定向权重编辑，而高层语义层分布化且难稀疏修改；并提出诊断→编辑→评估→再诊断的递归自改进循环，以及针对像素级精确回归的挑战分析。

Abstract: Can an LLM directly edit the weights of a diffusion-based image-restoration model to \emph{improve} restoration quality—without gradients, paired data, or fine-tuning? Recent experiments show frontier LLMs can fix structural bugs by modifying as little as (0.016%) of parameters, but cannot instill new capabilities. We ask whether weight-space programming (WSP) can move from \emph{debugging} to \emph{improvement} in the precise-regression setting of image restoration (deblurring, super-resolution, denoising, inpainting). Restoration diffusion UNets present a middle ground: low-level signal-processing layers are concentrated enough for targeted weight-direction edits, while high-level semantic layers remain distributed and resistant to sparse modification. We propose three technical conjectures: (C1) an LLM can localize the layer behind a quality gap (over-smoothing, color shift, hallucinated texture) within a 200 forward-pass budget; (C2) editing a small set of identified weight directions (e.g., scaling high-frequency filter norms, correcting output-bias channels) can measurably improve PSNR and perceptual quality; (C3) a recursive loop—diagnose (\rightarrow) edit (\rightarrow) evaluate (\rightarrow) re-diagnose—can compound gains across rounds, forming a minimal recursive self-improvement (RSI) loop for vision. We analyze the required AI capability (arguing targeted edits need not superintelligence, while general capability generation does) and detail the challenges of precise pixel-level regression. We provide a PyTorch-style deployment pipeline and outline a benchmark. This manuscript is conjectural, intended to open a new subtopic for image restoration.

[37] GroundSeek-CXR: Retrospective Evaluation-Guided Active Observation Framework for Multi-Label Chest Radiography Diagnosis cs.CVPDF

Yihang Fu, Xinyu Liu, Qing Xu, Xiaofeng Yang, Zhen Chen

TL;DR: 论文提出 GroundSeek-CXR，一个面向多标签胸片诊断的主动观察框架。该方法不再用每次局部观察对当前诊断的即时影响来监督，而是借助最终诊断回顾性评估观察与停止决策，并结合离线蒙特卡洛树搜索和回顾性差距加权偏好优化来学习主动观察策略。实验表明其优于现有最优方法。

Details

Motivation: 胸片诊断需识别细微病灶和多区域共现异常；现有基于偏好的视觉语言模型主动观察方法仅按观察对当前诊断的即时效果进行监督，无法衡量每次观察及停止决策对最终诊断的贡献。

Result: 大量实验表明 GroundSeek-CXR 优于当前最先进方法；相较即时分数偏好学习，在异常图像上的 sample-F1 提升 4.48%。

Insight: 创新点在于提出回顾性诊断评估 RDE，用离线 MCTS 通过独立延续为观察和停止候选打分；并设计回顾性差距加权偏好优化 RGPO，按分数差距加权学习偏好，优先选择能提升诊断准确性的动作。客观看，其核心是将主动观察决策的监督信号从即时收益转向最终诊断回报，并融合树搜索与偏好优化。

Abstract: Chest radiography (CXR) diagnosis is critical for thoracic disease assessment, yet challenged by the need to identify subtle lesions and co-occurring abnormalities across regions. While recent vision-language models can actively observe local regions, existing preference-based methods supervise each observation by its immediate effect on the current diagnosis, failing to capture how each observation and stopping decision contributes to the final diagnosis. To address this issue, we propose GroundSeek-CXR, an active CXR diagnosis framework that learns observation and stopping decisions by retrospectively evaluating them with the final diagnosis. Specifically, we first devise a Retrospective Diagnostic Evaluation (RDE) that scores observation and stop candidates from offline Monte Carlo tree search through independent continuations. Moreover, we design a Retrospective Gap-weighted Preference Optimization (RGPO) that learns from retrospective-score preferences weighted by their score gaps, prioritizing actions that improve diagnostic accuracy. Extensive experiments demonstrate that our GroundSeek-CXR framework outperforms state-of-the-art methods, and improves abnormal-image sample-F1 by 4.48% over immediate-score preference learning.

[38] Mitigating Spatial Blindness in Multiple Instance Learning for Whole-Slide Image Analysis cs.CVPDF

Xiangyu Li, Jincheng Li, Xing Xie, Bin Guan, Shaohua Kevin Zhou

TL;DR: 论文研究全切片图像多实例学习中的“空间盲”问题，即模型虽使用坐标或图结构，却未必真正利用组织空间邻域。作者提出 ResTopoMIL：先冻结已学习的组织成分预测器作为参考，再用坐标重分配分离损失训练图残差以学习空间修正。通过受控任务和多种 WSI 分类/生存分析基准验证该方法能缓解空间盲并提升性能。

Details

Motivation: 提供坐标或图并不保证训练后的预测器利用组织邻域；论文指出早期成分学习可能削弱空间通路的监督，从而造成空间盲。目标是让 MIL 在保留组织成分预测能力的同时学习有效的空间校正。

Result: 在受控空间任务上 AUC 为 0.987，同一架构联合训练为 0.684；在 4 个 WSI 分类和 5 个生存分析基准上，8 个任务的平均主指标提升，其中 4 个经 Holm 校正的患者簇检验后仍显著。破坏邻域会使性能接近仅成分锚点，而保邻域变换和分层扰动进一步表明方法依赖空间邻域信息。

Insight: 创新在于将“冻结成分预测器+图残差+坐标重分配分离损失”作为显式空间校正学习框架，并用置换/信息分析区分架构对称性、可用空间信息与优化机制的作用。客观看，其可借鉴点是把成分学习作为稳定参考锚点，再单独监督空间通路，避免联合训练中的空间监督被弱化。

Abstract: Whole-slide multiple instance learning can predict diagnostic and prognostic labels from tissue composition while making limited use of spatial organization. Providing coordinates or a graph does not establish that a trained predictor uses tissue neighborhoods. We examine this gap as spatial blindness and hypothesize that early composition learning can weaken supervision of a spatial pathway. We introduce ResTopoMIL, which freezes a learned composition predictor before training a graph residual with a coordinate-reassignment separation loss. Permutation and information analyses distinguish architectural symmetry, available spatial information and the proposed optimization mechanism. On a controlled spatial task, staged training achieved an AUC of 0.987, compared with 0.684 for the same jointly trained architecture. Performance remained high on a composition-only task. Across four WSI classification and five survival benchmarks, ResTopoMIL improved mean primary performance on eight tasks; four improvements remained significant after Holm-adjusted patient-cluster tests. Coordinate-free residual controls recovered some of the gains, while neighborhood destruction reduced performance close to the composition anchor. Neighborhood-preserving transformations and anatomy-stratified perturbations further characterized this dependence. These results support fixing a composition reference as a practical intervention for learning spatial corrections, while retaining additional appearance capacity as a contributing explanation.

[39] ReSMap: Recasting Satellite Priors for Robust and Accurate Online HD Map Construction cs.CV | cs.ROPDF

Kyungmin Kim, Sumin Lee, Sungoh Jeong, DoHyun Lim, Seonghyun Park

TL;DR: 论文提出ReSMap，一个冗余感知的摄像头-卫星融合框架，用于鲁棒的在线高清地图构建。它通过Predict-then-Fuse模块（利用预测置信度作为逐像素权重融合BEV特征）和Object-Centric Cooperative Multi-modal Query解码器（在查询级别进行模态分离与门控合并）来明确利用卫星图像作为冗余，以应对摄像头故障。在nuScenes上所有设置达到SOTA，并在摄像头故障场景下也达到SOTA。

Details

Motivation: 现有基于摄像头的在线高清地图构建系统依赖车载摄像头，易受传感器故障影响；而卫星图像可离线缓存，不受实时传感条件限制。现有方法仅在正常条件下利用卫星图像，未将其作为传感器退化时的冗余。因此，需要一种融合框架来增强鲁棒性。

Result: 在nuScenes数据集的所有分割和范围设置上达到最先进性能，并且在摄像头故障场景下也达到了新的最先进水平。

Insight: 创新点包括：将卫星图像作为冗余而非仅先验；提出PTF模块，使用各分支预测置信度作为逐像素可靠性权重进行融合，无需学习路由器；提出OCMQ解码器，在查询级别实现逐模态分离和门控合并。这些设计有效提升了摄像头故障下的鲁棒性，为多模态融合提供了新思路。

Abstract: Despite recent advances in camera-based online HD map construction, these systems remain susceptible to sensor failures due to their reliance on onboard cameras. In contrast, satellite images remain available independently of real-time sensing conditions as they are cached offline. Existing methods utilizing satellite images, however, primarily leverage them under nominal conditions, without explicitly addressing them as redundancy when a sensor degrades. To address this gap, we propose ReSMap, a redundancy-aware camera-satellite fusion framework for robust online HD map construction. Our framework explicitly leverages satellite images as a complementary source to improve robustness against camera failures. A Predict-then-Fuse (PTF) module fuses camera and satellite BEV features by using each branch’s own prediction confidence as a per-pixel reliability weight, with no learned router. An Object-Centric Cooperative Multi-modal Query (OCMQ) decoder then extends this per-modality separation to the query level: each modality cross-attends only to its own BEV, and per-modality updates are merged through a learned per-token gate. Our model achieves state-of-the-art performance on nuScenes across all splits and range settings. More importantly, it also sets the new state-of-the-art under camera-failure scenarios, demonstrating that satellite images serve not only as a strong prior but as practical redundancy for robust online HD mapping.

[40] IGSW: Multi-Weather Urban Scene Reconstruction with Intrinsic-Decoupled Gaussian Splatting cs.CVPDF

Zhongquan Wang, Long Cao, Peijia Xie, Senlei Ma, Jie Li

TL;DR: 论文提出 IGSW，一个两阶段框架，用于多天气城市场景重建。它结合内禀解耦的 4D 高斯场景建模与结构保持的天气细化：第一阶段将背景高斯外观分解为天气共享的反照率/法线及天气特定的辐照度/残差，并利用扩散编辑图像与逆渲染内禀图进行特征级软监督；第二阶段通过结构保持门控抑制不可靠教师区域，并用逐高斯天气传输变量和目标天气分支增强目标天气外观。

Details

Motivation: 现有方法将生成式图像编辑与动态高斯表示结合进行多天气重建，但天气编辑图像常含结构扭曲，直接拟合其外观会把教师图像误差传播到 4D 场景中，破坏原始结构和时序连续性。

Result: 在 nuScenes 和 Waymo 上，IGSW 在对比方法中取得最高的内容相似度和语义一致性、最低的时序误差，同时保留明显的天气相关光照与大气效果。

Insight: 创新点在于将 4D 高斯外观解耦为天气共享的内禀属性与天气特定光照/残差，并以特征级软监督和结构保持门控减少不可靠编辑图像的影响；客观来看，该内禀解耦与门控机制有助于在天气编辑中兼顾结构稳定性和外观真实性。

Abstract: High-fidelity weather editing of 4D scenes can support end-to-end training and closed-loop simulation for autonomous driving. Combining generative image editing with dynamic Gaussian representations enables multi-weather scene reconstruction, but weather-edited images often contain structural distortions. Directly fitting their appearance may propagate errors from teacher images into the reconstructed 4D scene, compromising the original scene structure and temporal continuity. To address these challenges, we propose IGSW, a two-stage framework that combines intrinsic-decoupled 4D Gaussian scene modeling with structure-preserving weather refinement. In Stage I, background Gaussian appearance is decomposed into weather-shared albedo and normals and weather-specific irradiance and residuals. Weather teacher images generated by a diffusion-based image editing model and intrinsic maps estimated by a diffusion-based inverse renderer provide feature-level soft supervision for stable structure and illumination changes. In Stage II, a structure-preservation gate combines semantic conflicts with cross-weather intrinsic discrepancies to suppress unreliable teacher regions. Per-Gaussian weather transport variables and target-weather branches then strengthen the desired weather appearance over the intrinsic-decoupled scene. Experiments on nuScenes and Waymo show that IGSW achieves the highest content similarity and semantic consistency and the lowest temporal error among the compared methods, while retaining pronounced weather-dependent illumination and atmosphere. The project page is available at https://akaxi6.github.io/IGSW/ .

[41] SphereFlow: Missing Modality Imputation via Geometric Transport on Hypersphere cs.CVPDF

Huihan Wang, Juncheng Wang, Yuxiang Feng, Wenlong Hou, Guoqi Yu

TL;DR: 论文提出SphereFlow，用于缺失模态插补。它通过冻结的自监督编码器将所有模态投影到共享超球潜空间，使跨模态插补转化为球面上邻近点之间的短程几何传输，并用流匹配实现。针对标准高斯噪声源与单位范数流形不匹配、线性插值路径偏离球面导致解码器无训练信号的问题，作者采用观测模态潜变量作为源、时间加权解码器损失等几何感知修改。

Details

Motivation: 现有方法将各模态视为独立潜流形，迫使生成器学习长且无结构的跨流形跳跃，插补困难。论文希望利用共享超球潜空间，将跨模态插补简化为邻近点间的短程几何传输。

Result: 在BraTS多模态MRI插补和CT-MRI转换实验中，方法在所有设置下取得SOTA生成保真度；用插补数据进行下游脑肿瘤分割时，缩小了约80%与全模态oracle之间的差距。

Insight: 创新点在于共享超球潜空间上的跨模态几何传输视角、以观测模态潜变量替代噪声源的data-as-source流匹配、扩展解码器域的时间加权损失，以及关于离流形偏差维度无关界的理论分析。客观上，其价值在于把流匹配的几何约束与缺失模态插补问题结合，减少跨模态生成中的非结构化长程映射。

Abstract: Missing modality imputation aims to synthesize unobserved modalities from observed ones. Existing methods treat each modality as living on its own latent manifold, forcing the generator to learn a long, unstructured jump between them. We present SphereFlow, which takes a fundamentally different view: by projecting all modalities into a shared hyperspherical latent space via a frozen self-supervised encoder, cross-modal imputation reduces to short-range geometric transport between nearby points on the sphere. Realizing this idea with flow matching, however, exposes two geometric conflicts: the standard Gaussian noise source is catastrophically mismatched with the unit-norm data manifold, and the linear interpolation path departs the sphere surface into a region where the decoder has no training signal. We resolve both with simple, geometry-aware modifications: replacing the noise source with the observed-modality latent so that the velocity field learns only the cross-modal displacement, and training the decoder with a time-weighted loss that extends its domain to the near-surface region traversed by the flow. We prove that this data-as-source formulation reduces the worst-case off-manifold deviation from a quantity that grows with the latent dimension to a small, dimension-independent bound governed by inter-modality similarity. Experiments on BraTS multi-modal MRI imputation and CT-MRI translation demonstrate state-of-the-art generation fidelity across all settings, while downstream brain tumor segmentation with our imputed data closes roughly eighty percent of the gap to the full-modality oracle. Code is available at https://github.com/WTHH031230/SphereFlow.

[42] VidUEU-Agent: A Data-Curation Agent for Multimodal Understanding, Editing, and Unified Tasks cs.CVPDF

Tengju Ru, Weitong Lian, Zecong Tang, Lingyi Meng, Haoran Li

TL;DR: 论文提出 VidUEU-Agent，一个面向多模态理解、编辑和统一任务的自动化数据策展智能体，可从原始视频合成高质量指令数据。该框架采用全局感知过滤、关键帧与多模态信号挖掘、元数据构建与路由、指令样本合成四阶段流水线，并基于下游训练结果闭环迭代数据分布。作者据此构建了 10 万样本的 VidUEU-100K 数据集，实验表明在其上微调基线模型可带来显著且一致的性能提升。

Details

Motivation: 多模态学习受高质量指令数据稀缺和标注成本高的严重制约，而原始视频虽丰富动态，却难以转化为精确的多模态指令数据。论文旨在自动化地从原始视频构建覆盖理解、编辑和统一任务的高质量指令数据，并优化数据质量与训练效果。

Result: 论文构建了跨多领域、多任务的 10 万训练样本数据集 VidUEU-100K。实验显示，在多个代表性 benchmark 上，使用该数据集微调基线模型均取得显著且一致的性能提升，验证了智能体框架和数据集的有效性；摘要未给出具体数值或 SOTA 对比。

Insight: 创新点在于将数据策展建模为自动化智能体流水线，从视频感知、关键帧与多模态信号挖掘到指令合成形成完整链路，并利用下游训练反馈自主迭代数据分布。客观来看，其可借鉴之处是闭环数据优化机制以及面向理解、编辑和统一任务的多模态指令数据构建范式。

Abstract: The advancement of multimodal learning is heavily constrained by the scarcity and high annotation cost of high-quality instruction data. While raw videos offer an abundant and dynamic source of visual knowledge, transforming them into precise instruction data remains highly challenging. To bridge this gap, we present VidUEU-Agent, an automated pipeline that synthesizes instruction data from raw videos for multimodal understanding, editing, and unified tasks. VidUEU-Agent follows a four-stage pipeline: it performs global perception to filter low-quality shots, mines keyframes and multimodal signals, constructs metadata and routes the metadata to synthesize instruction samples. More importantly, our agent framework can autonomously iterate the data distribution based on downstream training results, thereby optimizing data quality, balance, and training effectiveness. Scaling this pipeline, we construct VidUEU-100K, a large-scale dataset of 100K training samples spanning multiple domains and tasks, providing a robust foundation for developing comprehensive multimodal capabilities. Extensive experiments demonstrate that fine-tuning baseline models on VidUEU-100K yields significant and consistent performance gains across diverse representative benchmarks, validating the effectiveness of our proposed agent framework and dataset.

[43] GVT3D: Gesture-Vision-Text Token Enrichment for 3D Embodied Reference Understanding cs.CVPDF

Vimarsh Shah, Dulanga Weerakoon, Ajinkya Deshpande, Saksham Agarwal, Nilesh Bhatia

TL;DR: 论文提出GVT3D，用于3D具身指代理解任务：在包含人类指向手势和语言描述的点云场景中定位目标物体的三维位置。模型通过Embodied Fusion在共享嵌入空间联合建模语言、手势和视觉，并利用注视化（foveation）策略仅处理指向方向附近粗区域以降低计算量。

Details

Motivation: 现有3D-ERU模型缺乏跨模态表示对齐，且处理成本随点云规模增大而显著增加，影响自然人机交互中的高效目标定位。

Result: 相较已有3D-ERU模型，GVT3D准确率提升约8%；引入foveation后在保持相当准确率的情况下实现2.95倍加速。

Insight: 创新点在于将语言、手势和视觉在共享嵌入空间中进行联合对齐融合，并借鉴注视机制只处理指向方向周围的粗区域点云，从而兼顾跨模态理解与计算效率。

Abstract: Natural human-robot interaction requires robots to interpret language and gestures within a 3D environment to support tasks such as pick-and-place and object retrieval. 3D Embodied Reference Understanding (3D-ERU) supports this by identifying the 3D location of a target object from a scene (point-cloud) containing a human pointing gesture and an accompanying language description. However, current 3D-ERU models lack alignment of cross-modal representations and incur high processing costs that increase with the point-cloud’s size. To address this, we propose GVT3D, which employs an Embodied Fusion module that jointly models language, gesture, and vision inputs within a shared embedding space, promoting cross-modal alignment. We additionally introduce a foveation approach that selectively processes the point cloud within a coarse region around the estimated pointing direction to reduce computations. GVT3D achieves ~8% higher accuracy and foveation enables 2.95x faster execution while maintaining comparable accuracy, relative to prior 3D-ERU models.

[44] Entropy-Gated Dynamic Rectification for Object Hallucination in Large Vision-Language Models cs.CVPDF

Riasad Alvi, Md. Asaduzzaman Shuvo, Md. Ahashan Habib, Md. Faiyaz Abdullah Sayeedi

TL;DR: 论文针对大型视觉语言模型中的物体幻觉问题，提出熵门控动态矫正（EGDR）。该方法利用前一步已可用logits的预测熵作为token局部控制信号，动态调整潜在引导的门控阈值：高熵时降低干预阈值，低熵时保持校准工作点。方法无需额外前向，在LLaVA系列上降低CHAIR幻觉指标且基本不影响POPE和通用多模态能力。

Details

Motivation: 现有潜在引导虽可在单次前向中缓解物体幻觉，但固定门控对早期和晚期解码步骤采用同一标准，忽略了自由生成过程中预测熵的显著时序变化。论文希望利用不确定性这一因果可用信号实现更精细的动态干预。

Result: 在LLaVA-1.5-7B上，静态门控将CHAIR CS/CI从53.00/15.30降至32.00/14.30，EGDR进一步降至29.00/13.60；在匹配58%干预率下静态控制为30.50/14.00。增益在LLaVA-NeXT-7B上保持并随生成预算增长，POPE和通用多模态能力基本不变，额外延迟约2%。

Insight: 创新点在于将前一步已可用logits的熵作为轻量、因果可用的token局部控制信号，动态调节潜在引导门控阈值，实现时序自适应矫正。客观上看，其价值在于无需额外模型前向即可复用现有引导机制，并以低开销改善物体幻觉。

Abstract: Large Vision-Language Models (LVLMs) routinely name objects that are absent from the image. Latent steering can reduce this failure in one model forward pass, but a fixed gate treats early and late decoding steps by the same criterion. We observe that predictive entropy varies sharply within free-running generation, while states that commit to short answers occupy a narrow, confident band. We exploit this temporal mismatch by using uncertainty as a causally available, token-local control signal for an existing steering gate. We propose Entropy-Gated Dynamic Rectification (EGDR), which lowers the intervention threshold when entropy derived from the preceding step’s already available logits is high and retains the calibrated operating point when it is low. On LLaVA-1.5-7B, the underlying static gate lowers CHAIR CS/CI from 53.00/15.30 to 32.00/14.30; replacing only its constant threshold with EGDR further lowers them to 29.00/13.60. At a matched 58% intervention rate, the corresponding static control obtains 30.50/14.00. The gain over the static gate holds on LLaVA-NeXT-7B and grows with the generation budget, while POPE scores and general multimodal capability remain effectively unchanged. EGDR requires no additional model forward pass and adds roughly 2% measured latency. Our code is available at https://github.com/alvi-uiu/EGDR.

[45] DynOSeg: Dynamic Object Segmentation via Motion-Geometry Fusion cs.CVPDF

Xu Han, Yiduo Wang, Yiran Wang, Viorela Ila

TL;DR: 论文提出DynOSeg，一种用于视频动态对象分割（MOS）的两阶段运动-几何融合方法。该方法先将几何感知残差图与深度编码融合为运动的几何表示，以区分相机运动与物体动态，再引入语义特征细化对象理解；同时构建了包括CARLA-MOS在内的多样训练语料。

Details

Motivation: 现有MOS方法多依赖光流或点轨迹等纯视觉运动线索，在相机与物体运动混合、阴影和反射等运动诱导外观变化下存在歧义；简单融合深度和语义信息也无法充分利用互补性。

Result: 实验表明DynOSeg在多种数据集上持续优于已发表的SOTA MOS方法，并提出了新的合成城市驾驶MOS训练数据集CARLA-MOS。

Insight: 创新点在于两阶段融合框架：先以几何感知残差图和深度编码构建运动几何表示，再结合语义特征进行对象级细化；同时通过新数据集增强训练多样性，体现了运动、几何与语义互补利用的价值。

Abstract: Moving Object Segmentation (MOS) detects and separates dynamic objects from the static background in a video, which plays a fundamental role in dynamic-scene understanding. Most MOS methods rely on pure visual motion cues alone, such as optical flow or point tracks, which is inherently ambiguous due to the compounded camera and object motions as well as motion-induced appearance changes, e.g. shadows and reflections. Naive fusion of additional information such as depth and semantics, however, cannot leverage their complementarity to the fullest extent. We therefore propose DynOSeg, a novel MOS approach that takes on a two-stage fusion structure. DynOSeg first combines a geometry-aware residual map and depth encoding into a geometric representation of motion, delineating camera motions and object dynamics. We then further incorporate semantic features to refine object understanding. We have additionally curated a diverse training corpus including CARLA-MOS, a new synthetic urban-driving dataset for MOS training. Our experiments show that DynOSeg consistently outperforms published state-of-the-art MOS methods across a diverse range of datasets. Our code and pretrained models can be found at https://github.com/ACFR-RPG/DynOSeg

[46] Is Forgetting Unified? Exploring Concept Unlearning in Unified Multimodal Models cs.CVPDF

Zhongkai Xue, Junkai Chen, Mingcong Lei, Rumo Zhang, Yixiao Liu

TL;DR: 论文研究统一多模态模型中的跨通路概念遗忘问题，即同时阻止模型生成目标概念图像并在输入图像中识别该概念，同时保留其他能力。作者提出诊断基准 UUBench 和统一遗忘方法 UniForget，并系统评估现有单通路遗忘方法在生成与理解两条通路上的局限。

Details

Motivation: 现有概念遗忘方法多只针对图像生成或视觉理解单一通路，难以在统一多模态模型中同时抑制目标概念的生成与识别，因此需要研究跨通路联合遗忘并保持模型其他能力。

Result: UUBench 覆盖 30 个目标概念和 60 个语义相关邻近概念；实验表明单通路方法及直接组合的联合遗忘效果有限。UniForget 在 BAGEL-7B-MoT 和 Janus-Pro-7B 上分别达到 80.0% 和 92.7% 的联合遗忘率，优于代表性基线，并揭示了联合遗忘与能力保持之间的权衡。

Insight: 论文的创新点在于提出跨通路概念遗忘任务与诊断基准 UUBench，并设计统一遗忘方法 UniForget 以同时作用于生成和理解通路；从客观角度看，其价值在于将单通路遗忘的不足系统化，并给出联合评估与联合干预框架。

Abstract: Unified multimodal models (UMMs) integrate image generation and visual understanding, but are existing single-pathway unlearning methods sufficient to suppress target concepts across both? In this work, we study cross-pathway concept unlearning. Our goal is to prevent a model from both generating images of a target concept and identifying it in input images, while preserving other capabilities. To this end, we introduce UUBench, a diagnostic benchmark covering 30 target concepts and 60 semantically related neighboring concepts. Experiments on representative UMMs show that the evaluated single-pathway methods achieve limited joint forgetting: generation-side editors strongly suppress target generation but yield limited forgetting in visual understanding, while understanding-side baselines struggle to achieve substantial forgetting even in their own pathway. Their direct composition also yields limited joint forgetting. We therefore propose UniForget, achieving joint forgetting rates of 80.0% on BAGEL-7B-MoT and 92.7% on Janus-Pro-7B, outperforming representative baselines. Further analysis reveals trade-offs between joint forgetting and capability preservation.

[47] Beyond Token Representations: Explicit Visual Object Grounding for Video Reasoning Segmentation cs.CVPDF

Jun Huang, Junyang Zhang, Zijie Yue, Bingkun Wang, Yong Luo

TL;DR: 论文针对视频推理分割中现有分割token语义单调、空间模糊的问题，提出显式视觉目标定位框架EVOG。方法先利用多模态大模型生成并多步精炼目标的边界框和点集等空间提示，再通过查询多样化与跨查询一致性对齐增强提示可靠性，并用基于提示的关键帧选择支持掩码解码与传播。在八个数据集上显著优于现有方法。

Details

Motivation: 现有VRS方法依赖少量分割token指导掩码预测，但这些token难以充分表达目标，且缺乏明确空间定位，导致分割不准确。论文希望用显式空间提示替代或补充隐式token表示。

Result: 在八个数据集上的实验表明该方法显著优于SOTA；摘要未给出具体指标数值。

Insight: 创新点在于将MLLM生成的显式空间提示（框和点集）引入VRS，并通过多步对话精炼、查询多样化与跨查询一致性约束，以及基于提示的关键帧选择提升掩码解码与传播的可靠性。

Abstract: Video Reasoning Segmentation (VRS) aims to infer and segment target objects specified by reasoning queries in videos. Existing methods typically first infer the target and generate a few segmentation tokens for it, which are then fed into a mask decoder for mask prediction. However, such segmentation tokens tend to be both semantically monotonous and spatially ambiguous, making them largely insufficient to represent the target to guide mask prediction. In this paper, we propose an explicit visual object grounding framework for VRS. Specifically, we first propose a spatial-aware prompt generation and refinement scheme, in which we query a multimodal large language model to generate explicit spatial prompts (i.e., bounding box and point set) for the target and then refine them through multi-step dialogue. Furthermore, we introduce a query diversification and alignment module to generate auxiliary queries that describe the same target from different perspectives. We then enforce cross-query consistency between their predicted spatial prompts to alleviate the training bias caused by query limitations. Finally, we select several reliable frames using a prompt-based keyframe selection strategy to support mask decoding and propagation. Extensive experiments on eight datasets demonstrate that our method significantly outperforms state-of-the-art methods. The code is available at https://github.com/viscom-tongji/EVOG.

[48] KeyWise: Efficient Video-Only Piano Transcription with a Lightweight Per-Key Architecture cs.CV | cs.LGPDF

Anastasios Chatzigiannis, Natalia Panagou, Maria Koziri

TL;DR: 论文提出 KeyWise，一种仅使用视频进行钢琴转录的轻量级逐键架构。它利用键盘已知几何结构，将校正后的键盘划分为 88 个等分 ROI，并用所有琴键共享的小型循环网络逐键跟踪按键状态，从而避免全局特征让网络自行学习每个键的位置。

Details

Motivation: 基于音频的钢琴转录受延音踏板、噪声或缺失音轨影响，而顶视角视频能直接显示每个琴键的按下与释放。现有视觉转录网络多用单个全局特征描述整个键盘，并需学习每个键的位置，参数和计算开销较大。

Result: 在 PianoVAM 上，KeyWise 以 1.3M 参数达到 95.0% onset F1@50ms 和 97.8% F1@100ms，优于 V2N 的 94.7% 和 97.0%，参数从 28.2M 大幅减少，同 GPU 上快 4.7 倍；100ms 下 release F1 为 97.0% vs 95.5%，velocity F1 为 89.4% vs 84.5%。在 PianoYT 上，PianoVAM 预训练的 KeyWise 达到 69.4% onset F1@100ms，与最佳公开结果相当，但新录音条件需微调。

Insight: 核心创新是利用键盘固定几何先验，将每个琴键显式建模为 ROI，并通过所有键共享的轻量循环网络逐键跟踪，避免学习键位并显著降低参数量与计算量；客观上看，其创新点在于把结构化先验与共享时序建模结合，实现高效且准确的纯视频钢琴转录。

Abstract: Audio-based piano transcription is hampered by the sustain pedal and by noisy or missing soundtracks, problems that video-only transcription avoids, since a top-view video shows directly when each key is pressed and released. Most visual transcription networks, however, describe the whole keyboard with one global feature vector and must learn where each key is. In this paper, we propose KeyWise, a lightweight per-key architecture that exploits the known geometry of the keyboard instead of learning it. The rectified keyboard is divided into 88 equal tiles that serve as regions of interest (ROIs), and one small recurrent network, shared by all keys, follows each key over time. On PianoVAM, KeyWise reaches the onset accuracy of the state-of-the-art V2N (95.0% vs. 94.7% onset F1 at 50 ms and 97.8% vs. 97.0% at 100 ms) with 1.3 M instead of 28.2 M parameters, and it runs 4.7 times faster on the same GPU. At 100 ms, it is also significantly more accurate when the key releases (97.0% vs. 95.5% F1) or the velocities (89.4% vs. 84.5% F1) must be correct as well. On PianoYT, KeyWise pre-trained on PianoVAM reaches 69.4% onset F1 at 100 ms, comparable to the best published result, although it needs fine-tuning for new recording conditions.

Taehoon Kim, Wonjun Hwang, Hyunsouk Cho

TL;DR: 论文针对CLIP在测试时遇到图像损坏时产生的模态偏差问题，提出一种在线测试时适应方法CDIM。该方法结合冻结的自监督骨干DINOv2与CLIP，通过交叉协方差logit集成、门控信息最大化优化和在线logit调整，在表示、优化和预测三个层面打破高置信错误带来的确认偏差循环。

Details

Motivation: CLIP在测试时损坏下会将扭曲视觉特征高置信地映射到错误文本锚点，而标准熵最小化会强化这些错误并导致预测标签分布崩塌，因此需要一种能缓解CLIP文本偏差的在线TTA方法。

Result: CDIM在ImageNet-C、CIFAR10-C和CIFAR100-C上优于所比较的基于CLIP的在线TTA基线，并在持续TTA和混合域TTA中保持竞争力；消融实验验证了各组件贡献。

Insight: 创新点在于从表示、优化和预测三层协同解决模态偏差：用无文本的DINOv2分支与CLIP文本原型投影进行交叉协方差集成，以门控信息最大化兼顾样本锐化和批次多样性并过滤不可靠样本，再用在线logit调整纠正类别分布偏斜。客观上看，将自监督视觉特征引入CLIP-based TTA以削弱文本偏差，是该论文较有借鉴价值的思路。

Abstract: Vision-language models such as CLIP enable zero-shot classification from arbitrary class names, but under test-time corruptions, their visual encoder projects distorted features onto incorrect textual anchors with high confidence: a phenomenon we term modality bias. Standard entropy minimization reinforces these overconfident errors, which creates a confirmation-bias loop and collapses the predicted label distribution. We propose an online test-time adaptation (TTA) method that breaks this loop at three levels. At the representation level, a frozen self-supervised backbone (e.g., DINO v2) is combined with CLIP through a cross-covariance logit ensemble. CLIP text prototypes are projected into the feature space of this text-free backbone, so the self-supervised branch classifies native image features and is less affected by the text bias of CLIP. At the optimization level, a gated test-time information-maximization objective pairs per-sample sharpening with batch-level diversity, while an adaptive gate excludes unreliable samples from the gradient. At the prediction level, an online logit adjustment corrects the residual class-distribution skew. CDIM outperforms the compared CLIP-based online TTA baselines on ImageNet-C, CIFAR10-C, and CIFAR100-C, and remains competitive in continual and mixed-domain TTA. Extensive analyses and ablation studies verify the contribution of each of its components. Our code is available at https://github.com/B1ueC1oud/CDIM

[50] 3DGS-VBench+ and 3DGSV-Assessor: A Multi-Dimensional Subjective Benchmark and a Reference-Flexible Objective Model for 3D Gaussian Splatting Video Quality Assessment cs.CVPDF

Yuke Xing, William Gordon, Jiarui Wang, Yujie Zhang, Guangtao Zhai

TL;DR: 论文针对3DGS视频质量评估中缺乏对齐参考和多维失真问题，构建了包含3036个重建、9108个MOS并覆盖整体、几何、颜色质量的大规模主观基准3DGS-VBench+，并提出参考灵活的多模态评估模型3DGSV-Assessor，可在一次生成中输出三个维度的质量分数并支持RR/NR评估。

Details

Motivation: 3DGS存储开销大，训练与压缩会引入表示特定以及几何/颜色维度特定失真；真实捕获又通常缺乏与连续合成相机轨迹对齐的GT视频，仅有稀疏捕获视图可作参考，因此需要稀缺对齐参考下的多维3DGS视频质量评估。

Result: 在3DGS-VBench+上与人类感知的相关性达到SOTA，并对其他NVS基准表现出较强零样本泛化；摘要未给出具体相关系数数值。

Insight: 创新点在于构建多维主观基准，并将基于排序的参考线索通过有界门融入冻结回归器，实现一次生成多维评分和RR/NR兼容；客观看，其价值在于把3DGS压缩失真分解为几何与颜色维度，并缓解真实场景参考稀缺问题。

Abstract: 3D Gaussian Splatting (3DGS) enables real-time novel view synthesis (NVS), but its heavy storage makes compression essential, and both training and compression introduce representation-specific distortions. Since generative compression in many methods treats geometric and color attributes via different parameters, it can induce dimension-specific distortions, motivating multi-dimensional assessment. Moreover, real-world captures generally lack an aligned ground-truth video for a synthesized, continuous camera trajectory, leaving only sparse captured views as references. To enable multi-dimensional quality assessment of 3DGS videos with scarce aligned references, we first present 3DGS-VBench+, a large-scale subjective video quality assessment (VQA) benchmark comprising 3{,}036 3DGS reconstructions, each paired with a rendered video, across 11 real-world and 22 synthetic scenes, generated by 6 representative compression algorithms at systematically designed levels and annotated with 9{,}108 MOSs covering overall, geometry, and color quality. Based on 3DGS-VBench+, we propose 3DGSV-Assessor, a reference-flexible, large multimodal framework that scores all three dimensions in one generation pass and fuses a ranking-based reference cue into the frozen regressor through a bounded gate, supporting both reduced-reference (RR) and no-reference (NR) assessment. Extensive experiments demonstrate state-of-the-art correlation with human perception on 3DGS-VBench+ and strong zero-shot generalization to other NVS benchmarks. The dataset and code will be released at https://github.com/YukeXing/3DGSV-Assessor.

[51] E2I-Flow: EEG-to-Image Reconstruction via Wavelet-Temporal Cross-Attention and Conditional Flow Matching cs.CV | cs.LGPDF

Haowei Pan, Hwa Hui Tew, Junn Yong Loo, Fuad Noman, Hernando Ombao

TL;DR: 论文提出E2I-Flow，一个EEG到图像的视觉解码与重建框架。它先用结合平稳小波频率分支和时间混合分支的双向交叉注意力编码器学习EEG时频表示，并通过对比损失和MSE损失将EEG表示与真实图像嵌入对齐；随后将对齐后的EEG表示作为Transformer条件流匹配模型的条件，预测对应视觉嵌入并合成图像。

Details

Motivation: 基于EEG的视觉解码与重建旨在从非侵入式脑信号恢复视觉信息，但EEG具有复杂的时频特性，且离散去噪框架难以将神经活动映射为有意义的视觉表示。

Result: 实验显示该方法具有较强零样本EEG到图像检索能力，并提升了视觉重建性能，优于多个当前先进基线；摘要未给出具体数值或benchmark名称。

Insight: 创新点在于用平稳小波频率分支和时间混合分支经双向交叉注意力与自适应融合学习EEG时频表示，并以条件流匹配替代离散去噪框架完成EEG条件下的视觉嵌入预测与图像合成。

Abstract: Electroencephalography (EEG)-based visual decoding and reconstruction aims to recover visual information from non-invasive brain signals. But there are challenges due to the complex temporal-frequency characteristics of EEG signals and the difficulty of discrete denoising framework in mapping neural activity to meaningful visual representations. Here, we propose E2I-Flow, an EEG-to-image decoding and reconstruction framework that combines a novel EEG encoder for time-frequency representation learning from EEG features, and conditional flow matching for visual decoding. A stationary-wavelet frequency branch and a temporal-mixing branch are integrated through bidirectional cross-attention and adaptive fusion, with the resulting EEG representations aligned to ground-truth image embeddings using contrastive and mean-squared-error loss supervisions. The aligned EEG representation is then used as a condition for a Transformer-based flow-matching model to predict the corresponding visual embedding for image synthesis. Experimental results show strong zero-shot EEG-to-image retrieval and improved visual reconstruction performance, outperforming several state-of-the-art baselines.

[52] BtsPruner: Role-Aware Visual Token Pruning for Large Vision-Language Models cs.CV | cs.AIPDF

Zihao Qi, Zhibo Zhang, Miaoxin Zhang, Jincheng Cai, Zengqiang Yan

TL;DR: 论文提出 BtsPruner，一种免训练的角色感知视觉 token 剪枝框架，用于降低大型视觉语言模型推理时的视觉 token 开销。方法观察到视觉 token 在编码过程中会从局部细节载体逐步转变为全局引用代表，并据此在固定预算下保留互补角色 token。

Details

Motivation: 现有免训练剪枝方法通常用语言模型注意力、视觉注意力或特征相似度统一衡量视觉 token 重要性，忽略了不同 token 在视觉编码过程中承担的角色差异。论文旨在解决 LVLM 处理大量视觉 token 带来的计算和内存开销问题。

Result: 在 LLaVA-1.5、LLaVA-NeXT、Qwen-VL、InternVL 和 CogVLM 上实验表明，BtsPruner 在激进 token 削减下仍能较好保持性能。LLaVA-1.5-7B 仅用 576 个视觉 token 中的 32 个（5.6%）即可保留稠密模型 91.6% 的性能；LLaVA-NeXT-7B 高分辨率设置下实现最高 5.95 倍 CUDA 延迟加速。

Insight: 创新点在于发现视觉 token 角色随编码推进由局部细节向全局代表演变，并基于注意力集中度演化定位角色转换点。方法同时保留后期全局代表 token，并递归回溯早期局部细节 token，以固定预算恢复互补的细粒度证据。

Abstract: Large Vision-Language Models (LVLMs) often process hundreds or thousands of visual tokens, resulting in substantial computational and memory overhead during inference. Existing training-free pruning methods typically estimate token importance using language-model attention, visual attention, or feature similarity. However, these methods treat visual tokens with uniform criteria and overlook that tokens may serve different roles during visual encoding. In this work, we investigate the evolution of visual token roles in LVLMs and observe that tokens progressively transition from local detail carriers to globally referenced representatives as encoding proceeds. Based on this observation, we propose BtsPruner, a training-free role-aware visual token pruning framework that preserves complementary token roles under a fixed budget. Specifically, BtsPruner first identifies the token-role transition point based on the evolution of attention concentration. It then discovers globally referenced representatives from later encoding stages and recursively retrieves local detail tokens from earlier stages to recover complementary fine-grained evidence. Extensive experiments across LLaVA-1.5, LLaVA-NeXT, Qwen-VL, InternVL, and CogVLM demonstrate that BtsPruner achieves superior performance retention under aggressive token reduction. On LLaVA-1.5-7B, BtsPruner preserves 91.6% of the dense model’s performance using only 32 of 576 visual tokens (5.6%). In the high-resolution LLaVA-NeXT-7B setting, it achieves up to a 5.95x CUDA-latency speedup.

[53] SEEK-VAU: Towards Evidence-Faithful Video Anomaly Understanding via Agentic Search cs.CVPDF

Minghe Wang, Weijun Zeng, Dapeng Luo, Xinyue Wang, Nong Sang

TL;DR: 论文提出SEEK-VAU，将视频异常理解重构为带有限搜索预算的智能体搜索-验证交互协议：模型不再被动推理预收集证据，而是主动收集异常事件链中前兆、触发和后果三阶段证据，并在最终判断前验证证据充分性。为训练该行为，论文提出证据忠实的反事实验证奖励EFCV，并构建SEEK-Bench基准。

Details

Motivation: 现有VAU系统通常按预定义规则以前馈方式收集采样帧或事件证据，一旦收集到的证据不完整，推理开始后模型无法继续搜索缺失证据，从而限制异常判断的可靠性。

Result: 摘要未报告具体定量结果或SOTA水平；论文主要提出SEEK-Bench基准，包含视频级episode、时间区间标注、语义QA和事件链阶段标签，用于评测证据忠实且可主动验证的VAU。

Insight: 创新点在于将VAU从被动证据推理转变为带预算约束的主动搜索与验证协议，显式建模异常事件链完整性和证据充分性；EFCV通过反事实验证奖励使所选证据具有支持性、紧凑性和必要性，是较可借鉴的证据质量约束思路。

Abstract: Most video anomaly understanding (VAU) systems still infer anomaly judgments from evidence collected in a feed-forward way under predefined rules, typically sampled frames, or events clustered from such frames. This protocol fails when the collected evidence is incomplete under imperfect predefined rules, because once reasoning begins the model can no longer search for the missing evidence. We address this limitation with SEEK-VAU, which reframes VAU as an agentic search-and-verify interaction protocol under a bounded search budget: instead of passively reasoning over pre-collected evidence, the policy actively gathers evidence across the three stages of an anomaly event chain (precursor, trigger, and aftermath) and verifies its sufficiency before finalization, turning event-chain completeness into an explicit design rather than an outcome of predefined rules. A bounded budget caps the cost of this search, but does not by itself teach the policy where to look or when to stop. To make this behavior learnable, we introduce evidence-faithful counterfactual verification (EFCV), which rewards selected evidence that remains supportive, compact, and necessary under counterfactual verification. We further introduce SEEK-Bench, featuring video-level episodes with temporal interval annotations, semantic QA and event-chain stage labels. Together, SEEK-VAU and SEEK-Bench establish a strong foundation for evidence-faithful and actively verifiable VAU.

[54] Gradient-Guided Orientation Learning for Weakly Supervised Oriented Object Detection cs.CVPDF

Junyong Jang, Juheon Park, Minyoung Jeon, Jaehyup Lee

TL;DR: 论文面向弱监督旋转目标检测（WS-OOD），指出水平框或点等弱标注缺乏显式方向信息，现有间接角度推断在光照、颜色、阴影和背景杂乱等外观变化下容易模糊且不稳定。作者提出梯度引导方向学习：从目标区域提取主导梯度方向并转换为伪角度标签做辅助监督，使检测器无需旋转框标注或额外可学习参数即可学习方向感知表示；同时引入角度感知评价指标，将角度偏差纳入标准AP。实验表明该方法提升检测性能和方向估计质量，所提指标比常规AP更能反映角度预测误差。

Details

Motivation: 论文旨在降低旋转框标注成本，利用更廉价的水平框或点等弱标注进行旋转目标检测；但弱标注不提供显式方向，现有依赖几何一致性或等变约束的间接监督在外观变化下易产生模糊、不稳定的角度估计。

Result: 在弱监督旋转检测基准上的大量实验表明，伪角度监督同时提升检测性能和方向估计质量；所提角度感知指标比单独使用常规AP更能真实评估角度预测。摘要未给出具体数值或明确SOTA声明。

Insight: 创新点在于利用输入图像自身包含的方向先验，从目标区域提取主导梯度方向生成伪角度标签进行辅助监督，无需旋转框标注或额外可学习参数；同时指出高空间IoU下AP可能掩盖显著角度错误，并提出融合角度差异的评价指标。

Abstract: Weakly supervised oriented object detection (WS-OOD) aims to reduce the annotation cost of rotated bounding boxes by leveraging cheaper annotations such as horizontal boxes or points. However, these weak annotations provide no explicit orientation information, forcing existing methods to infer object angles through indirect geometric consistency or equivariant constraints. Such indirect supervision often becomes ambiguous and unstable under real-world appearance variations, including illumination changes, color shifts, shadows, and background clutter. In this paper, we present a simple yet effective gradient-guided orientation learning strategy for WS-OOD. Motivated by the observation that the input image itself contains useful orientation priors, our method extracts dominant gradient-based orientation cues from object regions and converts them into pseudo-angle labels for auxiliary supervision, enabling the detector to learn orientation-aware representations without requiring rotated bounding box annotations or extra learnable parameters. Furthermore, since AP can mask substantial angular errors when spatial IoU remains high, we introduce an orientation-aware evaluation metric that incorporates angular discrepancies into the standard AP protocol. Extensive experiments on weakly supervised oriented detection benchmarks demonstrate that the proposed pseudo-angle supervision improves both detection performance and orientation estimation quality, while the proposed metric provides a more faithful assessment of angle prediction than conventional AP alone.

[55] ReXMo: End-to-End Video Character Replacement with Expression and Motion Preservation cs.CVPDF

Hualiang Wei, Wenhui Li

TL;DR: 论文提出 ReXMo，一个端到端的视频角色替换框架，可在替换演员身份的同时保留面部表情、动作和场景动态，支持单人、多人和选择性角色替换。方法通过合成驱动视频与原始视频共享表演但身份不同的配对三元组监督，利用 In-Context Diffusion Transformer 联合建模驱动视频、参考身份和文本提示，并用 Expression Enhancement Adapter 增强多人场景中的表情保留。

Details

Motivation: 现有视频角色替换方法多依赖掩码生成和姿态引导，在复杂场景中难以细粒度保留表情和动作。论文旨在减少这种依赖，实现更统一、更高质量的单人、多人和选择性角色替换。

Result: 摘要称大量实验表明 ReXMo 在身份保真度、运动一致性、表情保留和视觉质量上优于现有方法，但未给出具体 benchmark 名称或定量指标。

Insight: 创新点包括端到端统一支持多类角色替换、利用共享表演不同身份的合成配对三元组进行可扩展训练、用 In-Context Diffusion Transformer 联合条件学习身份转换，以及针对多人场景的表情增强适配器。客观上看，其核心可借鉴之处在于用合成配对监督替代对掩码和姿态引导的强依赖，并统一建模身份、动作与表情保持。

Abstract: Video character replacement aims to replace actors while preserving facial expressions, motions, and scene dynamics. Existing methods typically rely on masked generation with pose guidance, limiting fine-grained expression and motion preservation in complex scenes. We propose ReXMo, an end-to-end framework that supports single-, multi-, and selective character replacement scenarios. ReXMo introduces paired triplet supervision, where synthesized driving videos and original videos share identical performances but different identities, enabling scalable training through synthetic data generation. An In-Context Diffusion Transformer learns identity transformation by jointly conditioning on driving videos, reference identities, and text prompts, while an Expression Enhancement Adapter improves facial expression preservation in multi-person scenes. Extensive experiments demonstrate that ReXMo outperforms existing methods in identity fidelity, motion consistency, expression preservation, and visual quality. Our dataset and code will be publicly released.

[56] Therapy-FIS: A Dataset for Assessing Facilitative Interpersonal Skills in Psychotherapeutic Setting cs.CV | cs.AI | cs.CYPDF

Qijian Zheng, Wanqing Lin, Yuanfang Wang, Yijun Ge, Jordan Bate

TL;DR: 论文提出 Therapy-FIS，这是首个公开的、用于心理治疗场景中促进性人际技能（FIS）自动评估的数据集，包含基于 890 段治疗师视频和 8 段模拟患者视频提取的面部、声学和文本多模态信号，并提供 8 个 FIS 子量表评分和总体 FIS 评分。作者基于该数据集建立了包含治疗师表现评分和联合治疗师—模拟患者表现评分两个任务的基准，并提出 DyadicFIS 基线模型。

Details

Motivation: 现有 FIS 评估依赖人工观察评分，成本高、劳动密集且难以规模化；同时缺乏公开的心理治疗专业数据，限制了自动化 FIS 评估的发展。

Result: 论文构建了 Therapy-FIS 数据集和两个下游任务基准，并提出 DyadicFIS 基线；该模型在治疗师单人和双人设置下均取得最佳宏平均 CCC 分数，分别为 0.2807 和 0.3299。

Insight: 主要创新在于发布首个公开的标准化心理治疗 FIS 多模态数据集，并建立可复现的自动评估基准；方法上提出结合门控融合与路由读出机制的 DyadicFIS，用于治疗师—模拟患者联合表现建模。

Abstract: Facilitative Interpersonal Skills (FIS) is a performance-based assessment method that provides a structured framework for evaluating the interpersonal competencies of mental health professionals. However, existing FIS assessment relies heavily on human observer ratings, making the evaluation process labor-intensive, costly, and difficult to scale. Progress toward automated FIS assessment has also been constrained by the lack of publicly available professional data in psychotherapy settings. To address this limitation, we introduce Therapy-FIS, a dataset for automated FIS assessment in standardized psychotherapy tasks. To our knowledge, Therapy-FIS is the first publicly available dataset in this domain. It contains facial, acoustic, and textual signals derived from standardized therapist responses to standardized mock patient stimulus clips, based on 890 therapist video recordings and 8 patient videos, and is rated with eight FIS subscale ratings and an overall FIS score. Built on Therapy-FIS, we further establish a benchmark with two downstream tasks: Therapist Performance Rating and Joint Therapist–mock-Patient Performance Rating. We also introduce a competitive baseline, DyadicFIS, with gated fusion and routed readout under therapist-only and dyadic settings. DyadicFIS obtains the best macro CCC(Concordance Correlation Coefficient) scores on both tasks (0.2807 and 0.3299). Therapy-FIS offers a reproducible resource for data-driven automated assessment of therapeutic interpersonal skills in mental health.

[57] Phase-DGS: Phase-guided Dynamic Gaussian Splatting from Unsynchronized Multi-view Video cs.CV | cs.GRPDF

Hosung Jeon, Jun Young Jeong, Sangwoon Kwak, Joonsoo Kim, Sangmin Kim

TL;DR: 论文提出 Phase-DGS，用每帧在运动周期中的语义相位替代传统时间戳，为无同步多视角视频建立跨相机的时空对应，从而实现动态高斯泼溅重建。方法可处理时间偏移、丢帧和相机冻结，并能集成 RealTime-4DGS、FreeTimeGS 等现有骨干。

Details

Motivation: 动态高斯泼溅的时空重建通常依赖多相机严格时间同步，但实际中专用硬件成本高，音频对齐、几何匹配、姿态跟踪等数据驱动方法又受环境和场景假设限制。论文旨在摆脱对时间戳同步的依赖，解决非同步多视角视频下的动态场景重建问题。

Result: 在时间偏移、丢帧和相机冻结等设置下，Phase-DGS 相比基线最高提升 +8.5 dB PSNR，并可无缝集成 RealTime-4DGS 和 FreeTimeGS 等现有 DGS 骨干。

Insight: 核心创新是用表示帧在运动周期中内在状态的语义相位替代时间戳，使不同相机、不同时刻拍摄的同一运动状态帧自然对齐，从而建立非同步相机的时空对应。客观来看，该思路将同步问题转化为运动状态/相位匹配问题，减少对专用同步硬件和场景特定假设的依赖，并具备较好的骨干可插拔性；但其有效性可能依赖运动周期或相位可估计的假设。

Abstract: Spatiotemporal scene reconstruction with Dynamic Gaussian Splatting (DGS) fundamentally depends on perfect temporal synchronization across all cameras, an assumption rarely achievable in practice. Industry-standard synchronization requires expensive specialized hardware and complex setup, while data-driven alternatives including audio-based alignment, geometry matching, and pose tracking remain limited by restrictive environmental or scene-specific assumptions. To overcome these limitations, we propose Phase-DGS, a framework that replaces conventional timestamps with a semantic phase representing each frame’s inherent state within the underlying motion cycle. Frames capturing the same motion state share the same phase regardless of which camera records them or when, naturally establishing spatiotemporal correspondence across unsynchronized cameras. We validate Phase-DGS under temporal offsets, frame drops, and camera freezes, achieving up to +8.5 dB PSNR improvement over baselines while enabling seamless integration with existing DGS backbones including RealTime-4DGS and FreeTimeGS.

[58] Learning from Unlabeled Photographs for General Oral Image Analysis cs.CVPDF

Zhong-Shi Zhang, Zi-Li Yu, Xiao Xiao, Huan Liu, Gui-Song Xia

TL;DR: 论文提出DOVE，利用约53万张公开无标注口腔照片进行域自适应自监督预训练，构建可迁移的通用口腔视觉编码器。作者在5个分类和4个分割数据集上与3个基线比较，覆盖冻结骨干和端到端微调，验证其在少样本迁移和标签效率上的优势，表明无标注口腔照片可支撑通用且标签高效的口腔视觉研究。

Details

Motivation: 现有口腔图像AI模型多针对单一任务，且依赖昂贵的专家标注，缺乏可跨异质视觉任务迁移的通用表示。论文试图验证无标注口腔照片的自监督预训练能否产生可迁移的视觉表征。

Result: DOVE在5个分类和4个分割数据集上整体表现最强，在多数benchmark的少样本迁移和标签效率上最优。20-shot冻结骨干设置下，分类平均macro-F1比第二好的DINOv3高4.37个百分点，分割平均macro-Dice高4.49个百分点；端到端微调时仅用10%标注数据，在多数数据集达到最强全数据基线至少80%的性能。

Insight: 论文宣称的创新点是面向口腔照片的大规模域自适应自监督预训练，以及跨分类/分割任务的通用性和标签效率验证。客观看，其贡献更偏向领域数据构建、领域适配预训练与系统评测，而非全新的自监督算法；但展示了无标注口腔图像学习通用骨干的可行性和实用价值。

Abstract: Oral photographs contain rich clinical information and hold significant potential for artificial intelligence applications; however, most models are trained for individual tasks and rely on costly expert annotations. We hypothesized that self-supervised training on unlabeled oral photographs could produce transferable visual representations for heterogeneous vision tasks. To test this hypothesis, we developed DOVE (Dental Optical Visual Encoder) by domain-adaptive self-supervised pretraining on approximately 530,000 unlabeled oral photographs collected from publicly accessible sources. DOVE was benchmarked against 3 established baselines across 5 classification datasets and 4 segmentation datasets, covering both frozen backbone evaluation and end-to-end finetuning. DOVE achieved the strongest overall performance, ranking best in few-shot transfer and label efficiency on most benchmarks. Under the 20-shot frozen-backbone protocol, DOVE outperformed the second best DINOv3 by 4.37 percentage points in mean macro-F1 across classification datasets and by 4.49 percentage points in mean macro-Dice across segmentation tasks. Under end-to-end finetuning, DOVE achieved at least 80% of the strongest full-data baseline’s performance across most evaluation datasets using only 10% of the labeled training data. Qualitative retrieval and feature analyses further indicated better organization of image-level and spatial representations. These findings support DOVE as a generalizable and label-efficient backbone for oral vision research.

[59] DenseTwins: Dense Modality Decoupling for Multimodal Self-Supervision cs.CV | cs.AI | cs.LGPDF

Carlo Sgaravatti, Daniele Kopyshevskiy, Davide Secco, Giacomo Boracchi

TL;DR: 论文提出 DenseTwins，一种面向多模态自监督学习的稠密表征框架。该方法将稠密特征空间显式解耦为共享成分与模态特有成分，并结合稠密跨模态对应关系与冗余减少目标，同时优化增强视图下的模态内目标和跨模态对齐/去相关目标，从而在保留空间结构的同时捕获共享语义与模态特有线索。

Details

Motivation: 现有多模态自监督方法要么通过强制模态间特征对齐而忽略模态特有信息，要么只学习全局表征而破坏空间结构；论文旨在学习同时保留共享语义和模态特有线索的稠密多模态表征。

Result: 论文在多个多模态基准和传感设置上评估 DenseTwins，取得与现有多模态自监督基线相比的 SOTA 或具有竞争力的性能。

Insight: 创新点在于将稠密特征空间显式解耦为共享与模态特有部分，并用稠密跨模态对应和冗余减少目标联合约束；客观看，其可借鉴之处是结构化稠密嵌入设计，在跨模态对齐的同时保留模态特异性和空间结构。

Abstract: Multimodal perception is central to many computer vision applications, including robotics and autonomous driving, where different sensing modalities provide complementary cues. Self-supervised learning offers a scalable solution for exploiting large amounts of unlabeled multimodal data. However, existing methods either neglect modality-specific information by enforcing feature-space alignment between modalities or learn only global representations without preserving spatial structure. We propose DenseTwins, a multimodal self-supervised framework that learns dense representations by explicitly decoupling the dense feature space into shared and modality-specific components. The proposed approach combines dense cross-modal correspondences with redundancy-reduction objectives, enabling the model to capture both shared semantics and modality-specific cues within a structured dense embedding space. DenseTwins jointly optimizes an intra-modal objective across augmented views and a cross-modal objective that aligns shared features while decorrelating modality-specific information. We evaluate DenseTwins across multiple multimodal benchmarks and sensing setups, achieving state-of-the-art or competitive performance across multimodal self-supervised baselines. Code available at: https://github.com/CarloSgaravatti/DenseTwins.

[60] Cycle WAM: Coupling Action and Video through Cycle Consistency cs.CV | cs.AI | cs.ROPDF

Qinglin Yang, Jiepeng Wang, Yuxing Huang, Haibin Huang, Kai Xu

TL;DR: 论文提出 Cycle WAM，将视频与动作视为由因果关系联系的一对模态，通过统一训练框架学习二者对应关系。方法在视频-动作联合训练中加入双向转换，并引入可微的循环重建目标来耦合前向与逆向动态映射。模型同时支持前向动态预测和逆向动态推断，并能更有效地利用预训练视频先验。

Details

Motivation: 现有世界动作模型虽利用预训练视频先验，但训练目标本身不能保证动作与预测出的视觉后果一致，可能限制预训练能力的发挥。论文希望利用动作导致场景变化、视频变化反推动作这一因果联系，强化视频与动作的跨模态一致性。

Result: Cycle WAM 在 LIBERO-Pro 上取得同类方法中的 SOTA，在 DROID 上显著改进 DreamZero；与 UWM 的对比显示其逆向动态性能具有竞争力，且前向动态预测呈现更清晰的视觉细节。

Insight: 创新点在于用双向转换联合学习前向与逆向动态，并通过可微循环重建目标显式约束动作与视频预测之间的一致性。客观上看，将视频-动作对应关系建模为可闭环验证的因果配对，是更充分利用视频预训练先验的可借鉴思路。

Abstract: Recent advances in video modeling have driven substantial progress in world action models (WAMs). Existing approaches exploit pretrained video priors by decoding actions from latent representations or inferring them from generated videos. The underlying link is causal: robot actions produce changes in the scene, and video observations of these changes provide evidence for inferring the actions that produced them. However, existing training formulations do not by themselves ensure that actions and their predicted visual consequences are consistent, potentially limiting the use of pretrained capabilities. We treat video and action as paired modalities linked by this causal relationship and learn their correspondence through a unified training framework designed to better exploit pretrained video priors. Cycle WAM first augments video-action co-training with bidirectional conversion to reinforce cross-modal correspondence and recover forward and inverse dynamics. It then introduces a differentiable cycle-reconstruction objective to further couple the two mappings. The resulting model supports both forward and inverse dynamics, with the potential to improve performance through more effective use of pretrained video priors. Cycle WAM achieves state-of-the-art performance among comparable methods on LIBERO-Pro and substantially improves DreamZero on DROID. Comparisons with UWM characterize its inverse-dynamics performance and show sharper visual detail in the displayed forward-dynamics predictions.

[61] SPOD: Recovering Suppressed Temporal Evidence in Video-LLMs via Single-Pass Orthogonal Decomposition cs.CVPDF

Yangguang Lin, Quan Fang, Xiaoyan Liu, Zefan Xu

TL;DR: 论文提出 SPOD，一个无需训练的视频大语言模型推理框架，通过单次前向传播中的隐状态正交分解来恢复被抑制的时间证据。SPOD 包含帧显著性分析 FDA 和时间差分正交子空间 TDS，将最终隐状态分解为时间、先验和残差分量并分别缩放，以增强动态信息、抑制静态外观偏置。方法在多个时间推理基准上优于强推理时基线，并保持接近原始推理的效率。

Details

Motivation: Video-LLM 在场景转换理解和事件排序等时间敏感推理中易出现时间幻觉，而现有方法要么需要昂贵重训练，要么依赖低效多轮推理和测试时外部监督。论文指出现有时间证据已编码在模型隐表示中但未被充分利用，因此希望以无需训练、单次推理的方式恢复这些证据。

Result: 在 VidHalluc、EventHallusion 和 VideoMMMU 上，SPOD 在多种时间推理任务和模型架构上持续优于强推理时基线，同时保持接近原始推理的效率。

Insight: 创新点在于发现并利用 Video-LLM 隐表示中已存在但被抑制的时间证据，通过 FDA 识别显著帧、TDS 构造正交时间与先验子空间，并对隐状态分量进行解析式重缩放，实现无需训练的单次前向时间推理增强。

Abstract: Video Large Language Models (Video-LLMs) have achieved strong performance on many video understanding tasks, yet remain vulnerable to temporal hallucination in time-sensitive reasoning scenarios such as scene transition understanding and event ordering. Existing solutions either require costly retraining or introduce inefficient multi-pass inference and external supervision at test time. In this work, we show that temporal evidence is already encoded in Video-LLM hidden representations but is not fully exploited during standard inference. Motivated by this observation, we propose Single-Pass Orthogonal Decomposition (SPOD), a training-free framework that recovers suppressed temporal evidence through analytical hidden-state manipulation in a single forward pass. SPOD consists of two components: Frame Distinctiveness Analysis (FDA), which identifies salient frames from middle-layer hidden-state geometry, and Temporal Delta Subspace (TDS), which constructs orthogonal temporal and prior subspaces from weighted inter-frame hidden-state differences. By decomposing final hidden states into temporal, prior, and residual components and rescaling them independently, SPOD enhances dynamic information, suppresses static appearance bias, and preserves complementary information. Experiments on VidHalluc, EventHallusion, and VideoMMMU show that SPOD consistently outperforms strong inference-time baselines across temporal reasoning tasks and model architectures while retaining near-vanilla inference efficiency. These results suggest that improving temporal reasoning in Video-LLMs may depend not only on additional supervision, but also on recovering temporal evidence already present in their internal representations. Our code is available at https://github.com/VioAgnes/SPOD.

[62] iMED: A Multi-Endoscope Dataset for Surgical 3D Perception cs.CV | cs.ROPDF

Sierra Bonilla, Fengyi Jiang, Chinedu Nwoye, Jingpei Lu, Kailey Reardon

TL;DR: 论文介绍了iMED，首个用于机器人辅助微创手术的同步多内窥镜数据集，旨在解决现有手术视觉基准缺乏独立留出视角的问题。该数据集包含340个序列约17万个同步时间点，来自两个独立立体内窥镜的4个视角，涵盖离体、尸体和活体手术场景。论文提出了基于一内窥镜训练另一内窥镜测试的评估协议，基准测试了23种最先进方法，发现依赖光度监督的方法在跨内窥镜设置下性能显著下降。

Details

Motivation: 现有手术视觉数据集仅从单一内窥镜捕获单轨迹序列，无法区分几何泛化与沿狭窄前向路径的光度插值，缺乏独立留出视角，导致无法公正评估模型的几何泛化能力。

Result: 基准测试了23种最先进方法，实验表明：依赖光度监督的方法在留出内窥镜设置下性能显著下降，而具有显式几何正则化的方法更鲁棒；大规模基础模型在手术图像上迁移效果出奇地好，常优于领域特定模型，暗示当前微调协议可能存在缺陷。

Insight: 创新点在于构建了首个同步多内窥镜数据集，提供独立留出视角，并提出了新的评估协议用于手术视觉的几何泛化；客观分析表明，数据集的多视角同步记录和丰富标注为评估方法泛化性提供了新基准，基础模型的优异表现也提示了领域微调策略的改进空间。

Abstract: We introduce iMED, the first synchronized multi-endoscope dataset for robot-assisted minimally invasive surgery (RAMIS), designed to address a fundamental limitation in surgical vision benchmarking: the absence of independent held-out viewpoints. Existing surgical datasets capture single-trajectory sequences from a single endoscope, making it impossible to distinguish geometric generalization from photometric interpolation along a narrow forward-facing path. iMED provides 340 sequences ~170K synchronized timepoints with 4 views per timepoint) recorded simultaneously from two independent stereo endoscopes across ex vivo, postmortem, and live surgical settings, spanning 14 specimens with diverse anatomical regions and motion regimes. The dataset includes calibrated camera intrinsics, ArUco-based frame-wise pose estimates with uncertainty quantification, instrument segmentation masks, and rich clinical metadata. Using a train-on-one-endoscope, test-on-another evaluation protocol, we benchmark 23 state-of-the-art methods across rigid and deformable novel view synthesis, pose estimation, feature matching, and monocular depth estimation. Our experiments show that methods relying primarily on photometric supervision degrade substantially under this held-out-endoscope setting, while methods with explicit geometric regularization are more robust across views in our benchmark. Notably, large-scale foundation models transfer surprisingly well to surgical imagery, often outperforming domain-specific models, suggesting pitfalls in current fine-tuning protocols. iMED establishes a new evaluation protocol for geometric generalization in surgical vision.

Qingguo Hu, Xinyu Ma, Yujie Lin, Yaoxiang Wang, Qi Shi

TL;DR: 论文研究大范围航拍区域下的精细跨视角定位问题，指出传统方法依赖狭窄地理先验，扩展到更大区域后会面临地面与航拍视图空间范围差异大、相似地点多导致匹配困难与视觉歧义。作者提出 RingMatch 搜索-验证框架，先利用预测不确定性逐步缩放寻找候选位置，再通过深度环几何表示和 LVLM 验证器消除残余歧义，并构建 VIGOR-Pro 基准。

Details

Motivation: 现有精细跨视角定位方法通常假设上游定位已给出较窄的地理先验，当初始定位较粗时适用性受限；直接扩展到更大航拍区域会因尺度差异和相似地点增多而显著增加匹配难度与视觉歧义。

Result: 在构建的 VIGOR-Pro 基准上（由 VIGOR 航拍瓦片组成、空间范围可达 400 m），RingMatch 在航拍区域扩大时仍保持较强定位性能，并持续优于代表性的端到端方法和检索式方法。

Insight: 论文宣称的创新点包括基于预测不确定性的渐进式搜索与验证框架，以及用于 LVLM 验证器的深度环几何表示，可作为视角共享的空间签名解决残余歧义。客观来看，其价值在于将大范围跨视角定位建模为“候选搜索+几何签名验证”，并把不确定性引导与视觉语言模型验证结合，以应对广域下的尺度差异和相似地点歧义。

Abstract: Fine-grained cross-view localization aims to precisely localize a ground camera against geo-referenced aerial imagery, but current methods typically assume a narrow geographic prior, substantially limiting their applicability when upstream localization remains coarse. To bridge this gap, we study this task over much larger aerial regions. However, this seemingly simple extension fundamentally changes the localization problem: the growing disparity in spatial extent between ground and aerial views makes reliable cross-view matching difficult to establish directly, while the increasing number of look-alike locations in larger regions introduces substantial visual ambiguity. To address these challenges, we propose RingMatch, a search-and-verification framework for wide-area cross-view localization. RingMatch first leverages predictive uncertainty to uncover plausible locations and progressively zooms into their surroundings to obtain increasingly fine-grained evidence, yielding a compact set of promising candidates. We then introduce the depth ring, a novel geometric representation that reveals subtle variations in the spatial layout around a location and serves as a view-shared spatial signature for an LVLM verifier to resolve residual ambiguity. To benchmark this setting, we construct VIGOR-Pro by composing VIGOR aerial tiles into contiguous regions with spatial extents of up to 400 m. Experiments show that RingMatch maintains strong localization performance as the aerial region expands, consistently outperforming representative end-to-end and retrieval-based methods.

[64] Pixel-space video generation with U-JiT cs.CVPDF

Théophane Vallaeys, Jakob Verbeek, Matthieu Cord

TL;DR: 论文探索不依赖预训练自编码器的像素空间视频生成，分析现有基于基础Transformer的像素空间图像生成方法在高维视频生成中的不足，并提出UNet-like Transformer架构U-JiT，结合训练配方与采样策略改进。作者在Kinetics-600上从零训练潜空间和像素空间模型进行受控实验，验证像素空间视频生成的可行性及性能优势。

Details

Motivation: 现有视频生成主流采用潜空间扩散/流匹配，但依赖预训练自编码器导致两阶段训练、重建与生成性能权衡以及信号损失。论文旨在探索直接在像素空间训练端到端视频生成模型的可行性，并解决高维像素视频生成中现有Transformer设计的局限。

Result: 在Kinetics-600的128分辨率frames-to-video任务上，U-JiT相比先前SOTA将生成FVD提升27%，相比JiT像素空间基线提升14%；在256分辨率下，随时空压缩率提高至64×64×8，优势进一步扩大。在class-to-video生成中，每帧FID相比潜空间模型提升27%、相比像素空间模型提升61%，FVD分别降低9%和56%。

Insight: 创新点在于系统评估像素空间视频生成路线，并提出UNet-like Transformer U-JiT及配套训练和采样改进，以克服基础Transformer在高维视频生成中的不足。其可借鉴之处是绕过预训练自编码器实现更端到端的视频生成，并通过受控实验比较像素空间与潜空间模型在不同分辨率和压缩率下的表现。

Abstract: Latent diffusion and flow matching models are ubiquitous in state-of-the-art generative video models. While highly effective, relying on pretrained autoencoders to compress the data into a lower-dimensional latent space also comes with drawbacks. First, training proceeds in two stages: autoencoder pretraining, with compression and reconstruction objectives, followed by latent generative model training, precluding fully end-to-end training. Second, autoencoders also lead to an inherent loss of signal, and the trade-off between reconstruction quality and generative performance in the latent space is still an active area of research. In this paper we explore the viability of generative video models directly in pixel space. We analyze recent approaches in pixel-space generative modeling of images that rely on basic transformer designs, and identify several shortcomings that limit their effectiveness for the high-dimensional pixel-space video generation. To overcome these limitations, we propose a UNet-like transformer architecture along with improvements to the training recipe and sampling strategy. We conduct controlled experiments, training latent and pixel-space models from scratch on Kinetics-600. For the standard frames-to-video task at 128 resolution, our U-JiT model improves generative FVD by 27% over prior state-of-the-art results, and by 14% against the JiT pixel-space baseline. Scaling to higher 256 resolution, we find the improvement over pixel-space and latent-based models grows as we increase the spatiotemporal compression rate up to $64 \times 64 \times 8$. Finally, we find that U-JiT improves per-frame FID by 27% over latent models and 61% over pixel-space models on class-to-video generation, with respectively 9% and 56% reduction in FVD.

[65] NightIQA: A Comprehensive Dataset and Benchmark for MLLM-Based Nighttime Image Quality Assessment cs.CVPDF

Yiding Tian, Jiaojiao Yi, Jiayuan Lu, Pufan Xu, Jinlin Wu

TL;DR: 论文提出 NightIQA，一个面向多模态大模型（MLLM）的夜间图像质量评估数据集与基准套件。它包含自采真实夜间图像数据集 NightIQA-Real、覆盖感知/技术评估/定位/推理四层级的视觉 grounded 多选题基准 NightIQA-Bench，以及可扩展的场景感知夜间退化合成管线 NightIQA-Syn。基于该数据对 Qwen3-VL-8B 进行 SFT 和 GRPO 微调后，在夜间 IQA 上优于开源 IQA 方法，并取得高于领先闭源模型的总体准确率。

Details

Motivation: 夜间图像存在复杂的复合退化，传统基于分数的 IQA 方法缺乏多维、语言化推理能力；现有真实 IQA 数据集以白天为主且较陈旧，合成数据又缺少夜间特定退化建模，导致 MLLM 夜间评估存在关键数据缺口。

Result: 构建了包含 1,769 张真实高分辨率图像、6 类典型失真和多维标注的 NightIQA-Real，1,078 道四层级多选题的 NightIQA-Bench，以及初始 8 万张高保真合成图像和生成代码的 NightIQA-Syn。微调后的 Qwen3-VL-8B 在基准上优于开源 IQA 方法，总体准确率超过领先闭源模型。

Insight: 创新在于面向夜间 IQA 构建真实+合成+基准的一体化数据体系，引入结构化多维标注和四层级视觉 grounded 评估，并用场景感知合成同时建模全局与实例级退化；训练上结合 SFT 与 GRPO，体现领域特定数据和结构化监督对 MLLM 夜间质量评估的价值。

Abstract: Nighttime images are inherently susceptible to complex, compound degradations, making accurate quality assessment exceptionally challenging. Score-based IQA methods condense quality into a single number, lacking the multi-dimensional, language-grounded reasoning required for nighttime evaluation. Multimodal Large Language Models (MLLMs) offer a more comprehensive alternative, yet remain constrained by a critical data gap: existing real-world IQA datasets are overwhelmingly daytime-centric and often built from outdated captures, while synthetic alternatives lack nighttime-specific degradation modeling, yielding artifacts far removed from authentic nighttime conditions. To bridge this gap, we present NightIQA, a dataset and benchmark suite purpose-built for MLLM-based nighttime IQA. It comprises three core components: NightIQA-Real, a real-world dataset of 1,769 high-resolution images that we captured ourselves with modern equipment, encompassing six canonical distortion types and structured multi-dimensional annotations; NightIQA-Bench, a benchmark of 1,078 visually grounded multiple-choice questions tailored to MLLM-based nighttime IQA, organized around a four-level evaluation hierarchy spanning Perception, Technical Assessment, Localization, and Reasoning; and NightIQA-Syn, an inherently scalable, scene-aware synthesis pipeline that jointly models global-level and instance-level degradations, with an initial release of 80k high-fidelity images alongside the full generation code. Fine-tuned on NightIQA with Supervised Fine-Tuning (SFT) and Group Relative Policy Optimization (GRPO), Qwen3-VL-8B outperforms open-source IQA methods and achieves higher overall accuracy than leading closed-source models, highlighting the value of domain-specific data and structured supervision.

[66] OCR in the Foundation Model Era: From Specialist Models to Unified Models cs.CVPDF

Zhiheng Li, Zongyang Ma, Jiaxian Chen, Cheng Cui, Yi Liu

TL;DR: 本文是一篇面向基础模型时代的OCR综述，围绕“忠实转录”重新定义OCR，即应在给定视觉范围内完整重建图像中的文本、符号和文档结构，且输出均可由图像视觉证据验证。文章梳理OCR从专用模型到统一模型的发展，比较检测识别联合建模、结构化序列生成和预训练视觉语言模型等路线，并将当代OCR系统归纳为五类，分析生成式OCR的幻觉、遗漏、空间定位不足和推理成本，最后总结评估缺口与未来方向。

Details

Motivation: OCR应用广泛且正从专用模型转向统一模型，但基础模型时代的系统架构、训练数据、生成式风险和评估方式差异较大，缺乏统一定义与系统梳理。论文旨在厘清从专用OCR到统一OCR的发展脉络，比较不同技术路线并指出关键挑战。

Result: 论文为综述性工作，未报告定量实验结果；其主要结果是提出忠实转录视角下的OCR定义，将 contemporary OCR 系统组织为模块化工具与任务专用模型、专用文档VLM、混合流程、通用VLM、商业产品与托管服务五类，并梳理从字符/场景文本识别到结构元素解析和整页转录的评估体系，指出输出定义不一致、性能难以归因到单个组件、真实采集条件与可见内容结构覆盖不足等缺口。

Insight: 创新点在于以“忠实转录”统一OCR目标，强调输出可被视觉证据验证，并系统比较五类共存OCR范式的架构、数据与训练；同时聚焦生成式OCR的幻觉、遗漏、空间定位和推理成本，并从评估定义、组件归因和真实场景覆盖角度揭示评测缺口。其可借鉴之处是提供了基础模型时代OCR的分类框架、风险清单和评估反思，并维护持续更新的相关工作总结。

Abstract: Optical Character Recognition (OCR) is among the oldest and most widely deployed applications of artificial intelligence. This survey reviews its development, focusing on the Foundation Model Era. We define OCR through faithful transcription: within a specified visual scope, OCR should completely reconstruct the text, notation, and document structure observable in the source image, and every textual, notational, or structural element in the output must be verifiable from visual evidence in that image. We trace the development from specialist OCR models toward unified models, examining joint text detection and recognition, structured sequence generation, and the use of pretrained vision-language models (VLMs). We organize contemporary OCR systems into five coexisting families: modular toolkits and task-specific models, specialized document VLMs, hybrid workflows, general-purpose VLMs, and commercial products and managed services. We compare their architectures, data, and training, examining hallucination and omission, insufficient spatial grounding, and inference cost in generative OCR. We review evaluation from character and scene text recognition to parsing individual structural elements and full-page transcription, highlighting gaps in consistent output definitions, attribution of performance to individual components, and coverage of real capture conditions and visible content and structure. Finally, we summarize the challenges facing OCR and discuss directions for future research. We continuously track related work at https://github.com/zhihengli-casia/Awesome-OCR-in-the-Foundation-Model-Era.

[67] VisSearch-Bench: Evaluating Long Horizon Visual Search in Multimodal Agents cs.CV | cs.AIPDF

ZhengXian Wu, Qianhao Nie, Zhuohong Chen, Junjie Gao, Yuyao Zhang

TL;DR: 论文提出 VisSearch-Bench，用于评估多模态智能体在长程视觉搜索中的视觉上下文管理能力。该基准包含 204 个人工编写任务，覆盖 7 个主领域、25 个子领域和五类视觉搜索任务，并为每个任务提供参考答案、可追踪证据和评分规则。论文在 16 个多模态模型上进行评测，发现即使配合搜索工具，最佳系统也仅完全解决 26.47% 的任务。

Details

Motivation: 现有搜索任务常只用图像识别初始实体，随后依赖文本检索，难以考察智能体如何跨图像、物体和搜索步骤组织并连接视觉信息。论文因此关注视觉上下文管理，构建更强调外观、局部细节、空间关系与外部信息结合的视觉密集搜索评测。

Result: 在 VisSearch-Bench 上评测 16 个多模态模型，最佳系统即使使用搜索工具也只完全解决 26.47% 的任务。这表明将视觉证据与搜索结果链接起来并形成完整答案仍具有显著挑战。

Insight: 论文的创新点在于提出覆盖多类长程视觉搜索任务的基准，并用 Accuracy 衡量完整任务成功、用 Process Score 衡量中间结果与子问题答案的正确性和完整性。客观来看，其价值在于细粒度过程评估和对视觉上下文管理能力的系统化任务设计。

Abstract: Multimodal agents use web and image tools to answer questions beyond their internal knowledge. Some search tasks use images mainly to identify an initial entity, then rely on text retrieval. These tasks offer limited insight into how agents connect information across images, objects, and search steps. We therefore focus on visual context management. This involves organizing and using visual information and linking it to other findings during search. Visually intensive tasks require agents to combine object appearance, local details, or spatial relations with external information. We introduce VisSearch-Bench, with 204 human-written tasks across 7 main domains and 25 subdomains. It covers Set-wise Visual Search, Sequential Visual Search, Hybrid Wide-Deep Search, Competitive Visual Disambiguation, and Visual-Structure-Guided Search. Each task provides a reference answer, traceable evidence, and scoring rules. Accuracy measures complete task success, while Process Score measures the correctness and completeness of individual results and subquestion answers. Across 16 multimodal models, the best system fully solves only 26.47% of tasks, even with search tools. These results highlight the challenge of linking visual evidence with search results to produce complete answers.

[68] Heads That Write, Not Just Point: Image Retrieval Heads in Vision-Language Models cs.CV | cs.CLPDF

Junsung Park, Sangwon Yu, Donghun Kang, Yeongtak Oh, Jaewon Jeong

TL;DR: 论文研究大型视觉语言模型在多图上下文中如何定位与问题相关的图像，将这一过程视为上下文图像检索，并提出ICIR-MCQ探针进行注意力头级分析。作者先根据注意力质量识别稀疏的pointer heads，再在输出侧训练轻量检索器识别writer heads，发现二者选择出的头集合差异很大；通过集合划分消融表明二者交集处的头对多图推理最关键。

Details

Motivation: LVLM可处理交错文本和多图，但其内部如何识别与问题相关的图像仍不清楚；论文希望从头级机制揭示这一grounding/检索步骤。

Result: 在ICIR-MCQ上发现稀疏pointer heads；单个pointer head在自然多图检索基准上与外部视觉语言检索器相当。top-100中pointer与writer仅重叠35–47个头；消融显示交集头对标准多图推理最重要，因果效应最一致。

Insight: 创新点在于区分“指向哪里读”的注意力pointer与“输出什么信息”的writer，并用输出侧轻量检索器筛选writer head；客观看，双准则交集和集合划分消融为LVLM内部图像检索机制提供了可解释、可因果验证的分析框架。

Abstract: Large vision-language models (LVLMs) can process interleaved text and multiple images in a single context, but how they internally identify the image relevant to a question remains poorly understood. We study this grounding step as in-context image retrieval and introduce ICIR-MCQ, a controlled probe for head-level analysis of image retrieval. Using each head’s attention mass over image-token spans, we first identify sparse pointer heads that reliably point to the target image. These heads transfer beyond the probe: a single pointer head is competitive with external vision-language retrievers on a naturalistic multi-image retrieval benchmark, showing that LVLMs contain a strong internal image retrieval signal. Attention alone, however, shows where a head reads from, not what its output passes on to later layers, so we introduce a complementary output-side analysis. For each head, we train a lightweight image retriever on its value-weighted output and use it to identify writer heads, whose outputs contain enough information to identify the target image. Surprisingly, the pointer and writer criteria select substantially different head sets, with only 35 to 47 heads overlapping among the top-100 heads under each criterion. Targeted set-partition ablations show that the heads most important for standard multi-image inference are concentrated in this pointer-writer intersection, while heads selected by only one criterion show less consistent separation from random-head controls. The most consistent causal effects arise from heads selected by both the attention-based pointer and output-side writer criteria. Code and data: https://github.com/parkquasar/Image-Retrieval-Heads.

[69] EP-LVMQA: Evidence-Preserving Coarse-to-Fine Localization for Low-Quality Long Video Moment Question Answering cs.CVPDF

Guoning Chen, Zhexun Hu, Yifan Wei, Yixuan Weng, Bin Li

TL;DR: 论文研究低质量长视频中的时刻问答（LVMQA），要求系统返回带时间戳且可支撑答案的证据，而非仅返回标签或自由文本。作者提出 EP-LVMQA 证据保留式由粗到细框架，包含不确定性感知多模态证据时间线、粗粒度证据规划器和证据条件细粒度定位器，并通过证据卡在两个模块间传递规划分数、匹配线索、置信度代理、模态证据、邻域上下文和时间先验。

Details

Motivation: 长教学视频中的证据常不完整或不可靠，如字幕缺失、ASR 噪声、OCR 稀疏、关键动作短暂且无解说；现有 VideoQA 系统输出难以检查答案依据，因此需要可定位、可核验的时间戳证据。

Result: 在 MedVidQA 上达到 26.1 R1@0.5 和 31.8 mIoU，相较最强非本方法分别提升 6.3 和 5.1 点；Suff.@2 与 Ver.@2 提升至 25.4 和 22.6，分别增益 9.5 和 8.8 点。匹配低质量扰动下，平均相对退化从最强非本方法的 14.6% 降至 11.1%；在 Ego4D-NLQ 和 QVHighlights 上与既有任务专用定位方法总体相当。

Insight: 创新点在于将证据保留和由粗到细定位结合，用证据卡显式传递规划器分数、匹配线索、源置信度代理、模态证据、邻域上下文与时间先验；客观看，其价值主要在低质量长视频的鲁棒证据定位、可检查性以及跨模态不确定性建模。

Abstract: Long instructional videos are increasingly used in medical education, industrial training, and procedural knowledge transfer, yet their evidence is often incomplete or unreliable: subtitles may be missing, ASR noisy, OCR sparse, and decisive actions briefly shown without narration. Existing VideoQA systems typically return labels or free-form text, making answer support difficult to inspect. We study Long Video Moment Question Answering (LVMQA), where a system returns timestamped answer-supporting evidence. We propose EP-LVMQA, an evidence-preserving coarse-to-fine framework comprising an Uncertainty-Aware Multimodal Evidence Timeline, a Coarse Evidence Planner, and an Evidence-Conditioned Fine Localizer. An evidence card transfers planner scores, matched cues, source-confidence proxies, modality-specific evidence, neighboring context, and a temporal prior between the latter two modules. On MedVidQA, EP-LVMQA achieves 26.1 R1@0.5 and 31.8 mIoU, improving the strongest non-ours results by 6.3 and 5.1 points. It raises Suff.@2 and Ver.@2 to 25.4 and 22.6, gains of 9.5 and 8.8 points. Under matched low-quality perturbations, average relative degradation is reduced from the strongest non-ours value of 14.6% to 11.1%. On Ego4D-NLQ and QVHighlights, EP-LVMQA remains broadly comparable to established task-specific localization methods.

[70] Behind Video Generation: Observations and Insights on Multimodal Understanding from Kling cs.CVPDF

Kling Team, Dong Chen, Jialu Chen, Xinlong Chen, Yifan Dai

TL;DR: 论文是Kling视频生成团队关于多模态理解的经验总结报告，提出生成质量是多模态理解在数据、条件、对齐、评估全链路上的投影。作者逐模块介绍共享多模态基础模型Kling-MLLM、面向DiT的captioning语义契约、提示增强、文本编码器选择、奖励建模与自动评估协议，旨在为可控视频生成背后的理解栈提供实践指导。

Details

Motivation: 视频生成瓶颈正从渲染质量转向语义控制，需要系统理解多模态能力如何贯穿数据、条件、对齐和评估链路，从而提升可控生成质量。

Result: 摘要未给出具体定量结果、benchmark对比或SOTA声明；主要结果是总结Kling多代实践中的模块化观察与设计经验，并提出可操作的工程指导。

Insight: 创新在于将视频生成质量系统性地映射为多模态理解全链路问题，并给出Kling-MLLM统一多模态输入、reference-grounded captioning、基础与agentic提示增强、感知训练文本编码器、双奖励模型及可追溯评估等实践洞见。客观看，其贡献更偏向工业经验总结和方法论框架，而非单一新算法或定量突破。

Abstract: Video generation has progressed from isolated clips to coherent short-form audiovisual sequences, and the bottleneck is shifting from rendering to semantic control. This report summarizes the observations and insights on multimodal understanding gained from successive generations of Kling. We argue that generation quality is the projection of multimodal understanding through the entire data–conditioning–alignment–evaluation chain, and we examine this chain module by module. Kling-MLLM, a shared multimodal foundation model that unifies image, video, audio, synchronized audio–video, and interleaved inputs, is specialized for each downstream module. Because a DiT is trained on far fewer training samples than a language model, we treat captioning as a semantic contract that maximizes description density, and extend it to reference-grounded captioning that makes reference bindings explicit. Prompt enhancement translates user intent into this caption distribution: basic PE is derived from captions of compelling videos and refined with text-level and generator-in-the-loop reinforcement learning, while agentic PE refines the story and retrieves compelling examples. We show that a reliable text encoder comes from perceptual training rather than from stronger language ability, and that candidates can be screened from text alone before end-to-end training. Reward modeling pairs a general reward model for explainable, localized critiques with a decision reward model for fast multidimensional scoring, treating perceptual bandwidth as a first-class design parameter. Finally, we review task-specific automatic evaluation, focusing on protocols that expose scoring shortcuts and help trace failures to their upstream source. We hope these lessons offer practical guidance for building the understanding stack behind controllable video generation.

[71] VISTA: Vision-Language Integrated Semantic Triage Architecture for Online Face Clustering in Long Videos cs.CV | cs.MMPDF

Jiashu Liao, Joanne Garde-Hansen, Sanjay Sharma, Tanaya Guha

TL;DR: 论文提出VISTA，这是首个面向长视频在线人脸聚类的视觉-语言集成语义分诊框架。它通过受控且不确定性感知的语义推理，在长视频中维持高聚类纯度和长程身份连续性，从而应对外观变化。

Details

Motivation: 现有长视频人脸聚类多为离线方法，难以满足数据不完整或计算受限的实际场景；少量在线方法依赖几何视觉特征和简单阈值，在长且无约束视频中面对光照、姿态和服装变化时容易失效。

Result: 在VPCD基准（包含《生活大爆炸》《吸血鬼猎人巴菲》《神探夏洛克》共15集）上，VISTA取得较稳健的精确率-召回率平衡，并在相同检测器、嵌入和评估协议下持续优于最强匹配在线无监督基线COFC。

Insight: 创新点在于首次将视觉-语言框架引入在线人脸聚类，并用受控、不确定性感知的语义推理进行身份关联决策；客观看，其价值在于用语义信息补充纯视觉嵌入，在在线约束下兼顾簇纯度和长程身份连续性。

Abstract: Clustering faces by identity in long, continuous video streams is a challenging task that serves as a building block for human-centric video analysis. The majority of existing face clustering models are offline, i.e., they assume the entire data to be available altogether. In many practical scenarios, however, the entire data is either unavailable or poses computational challenges to be processed in its entirety. A handful of online methods that exist rely on geometric visual features and simple thresholding techniques, which fail when the faces exhibit significant appearance variations in long, unconstrained videos. To this end, we present VISTA, the first vision language framework for online face clustering in long videos. VISTA leverages semantic reasoning in a controlled, uncertainty aware fashion to maintain both high cluster purity and long range identity continuity. This makes it robust to the variations in illumination, pose, and costume changes that the faces and characters exhibit in long real world videos. Extensive experiments on the VPCD benchmark dataset (with 15 full episodes from The Big Bang Theory [2007-2019], Buffy the Vampire Slayer [1997-2003], and Sherlock [2010-2017]) demonstrate that VISTA achieves a robust precision and recall balance, consistently improving over the strongest matched online unsupervised baseline (COFC) under identical detector, embeddings, and evaluation protocol. Code and supplementary material: https://github.com/SkyJiashu/VISTA.

[72] Learning Where to Look: Observation Policy Optimization for Thinking with Images cs.CVPDF

Junfeng Wang, Jiawei Liu, Yongchao Xu, Tao Jiang, Jiangbo Ai

TL;DR: 论文针对视觉语言模型在“用图像思考”时需主动放大高分辨率图像关键区域的问题，指出许多失败源于观察动作不准确，且边界框token具有更高不确定性。为此提出观察策略优化OPO，通过不确定性引导的观察分支和证据局部化的优势归因，对放大/裁剪等观察动作进行细粒度强化学习优化，从而提升视觉证据获取与视觉推理性能。

Details

Motivation: 复杂视觉问答中模型需要主动定位并裁剪信息区域，但现有基于最终答案奖励的强化学习方法只提供轨迹级反馈，难以为关键观察动作分配精确信用，导致定位错误或裁剪证据不足。

Result: 实验表明OPO在Qwen3-VL不同规模模型上均带来一致提升，并在与更大模型及工具增强基线的比较中取得有竞争力的表现；摘要未给出具体benchmark数值。

Insight: 创新点在于将图像放大工具调用视为可优化的观察事件，利用生成不确定性选择分支探索替代裁剪，并用局部证据充分性奖励只在对应观察动作token上做信用分配。客观看，其价值在于把观察策略从轨迹级结果监督中解耦，实现更细粒度的工具调用信用归因。

Abstract: Complex visual question answering often requires vision-language models to think with images, i.e., actively acquire local visual evidence by zooming into informative regions of high-resolution images. Our error analysis shows that many failures in this setting originate from incorrect observation actions, where the model fails to localize key regions or obtain crops that contain sufficient evidence for answering. We further observe that bounding-box tokens exhibit substantially higher generation entropy than ordinary language tokens, indicating high uncertainty when the model decides where to look. However, existing reinforcement learning methods based on final-answer rewards provide only trajectory-level feedback, making it difficult to assign precise credit to the observation actions. We propose Observation Policy Optimization (OPO), an RL algorithm for optimizing observation actions in tool-augmented thinking with images. OPO consists of two core modules. First, Uncertainty-Guided Observation Branching (UOB) treats each image zoom-in tool call as an observation event and selectively branches at high-uncertainty, low-redundancy events during rollout, enabling efficient exploration of alternative cropping decisions and their resulting visual evidence. Second, Evidence-Localized Advantage Attribution (ELAA) compares sibling branches using local evidence-sufficiency rewards and assigns credit only to the corresponding observation-action tokens. By decoupling observation-action optimization from trajectory-level outcome supervision, OPO improves visual evidence acquisition and further enhances visual reasoning performance. Experiments show that OPO consistently improves Qwen3-VL models across scales and achieves competitive performance against larger models and tool-augmented baselines. Code is publicly available.

[73] PATTS: Parallel-Aware Test-Time Scaling for Thought-Image Generation in Multimodal Discrete Diffusion LLMs cs.CVPDF

Yasong Dai, Zeeshan Hayder, David Ahmedt-Aristizabal, Hongdong Li

TL;DR: 论文研究面向并行多模态离散扩散LLM的测试时扩展（TTS），针对推理token与图像token联合解码的thought-image生成轨迹，提出PATTS框架，在生成过程中搜索中间thought-image状态，并用跨模态验证信号和状态相关可靠性权重进行选择。实验表明该方法在GenEval上显著优于无TTS、Best-of-N和FKD。

Details

Motivation: 现有图像生成TTS多对完成图像或图像中心中间状态排序，而并行多模态离散扩散模型中推理与图像token共同解码，最终图像选择存在验证信号与组合正确性弱对齐的问题，因此需要面向中间thought-image状态的TTS方法。

Result: 在MMaDA-Parallel固定N=32评估中，PATTS取得GenEval 0.844，高于无TTS的0.628、Best-of-N的0.787和FKD的0.810，显示可靠性感知轨迹验证能提升测试时选择效果。

Insight: 创新点在于将TTS从最终图像选择转向并行thought-image中间状态搜索，并利用跨模态验证信号与模型内部轨迹诊断得到的可靠性权重调节外部语义验证影响；客观看，其价值在于为并行多模态离散扩散LLM的联合生成轨迹提供了更细粒度的验证与选择机制。

Abstract: Test-time scaling (TTS) improves generative models by allocating additional inference-time computation to sampling, search, and verification. Existing image generation approaches, however, typically rank completed images or image-centric intermediate states. We study TTS for parallel multimodal discrete diffusion language models, in which reasoning tokens and image tokens are decoded jointly within a single generation trajectory. We first diagnose a limitation of conventional final-image selection: on GenEval, a CLIP-style image-prompt score is weakly aligned with within-prompt compositional correctness, leaving substantial headroom between verifier selection and oracle selection. We then propose PATTS, a novel parallel-aware TTS framework that searches over intermediate thought-image states using cross-modal verifier signals. PATTS uses a state-dependent reliability weight, derived from model-internal trajectory diagnostics, to modulate the influence of external semantic verification during generation. In the fixed N=32 evaluation on MMaDA-Parallel, the PATTS reaches a GenEval score of 0.844, compared with 0.628 without TTS, 0.787 for Best-of-N, and 0.810 for FKD. These results show that reliability-aware trajectory verification can improve test-time selection for parallel thought-image co-generation under the evaluated search budget. Our code release is at https://github.com/Ringo-Star/PATTS.git

[74] An Agentic Framework for Quality-Aware and Scenario-Adaptive Image Compression cs.CVPDF

Zhenghao Chen, Lebin Zhou, Zicong Chen, Naifu Xue, Jiahao Li

TL;DR: 论文提出了一种名为PICA的多智能体框架，用于质量感知和场景自适应的图像压缩。该框架包含四个协作智能体（感知、改进、编码和编码后改进），能够根据用户意图和图像内容调整现有编解码器，适应退化输入。同时引入了与编解码器无关的ROI编码方法AROIC和场景引导的质量-率评估策略SGQR。实验表明PICA在多种场景和数据集上提升了传统和神经编解码器的压缩性能。

Details

Motivation: 现有图像编解码器难以适应多样化的实际压缩场景，因为它们通常针对特定目标优化，且假设输入干净，而真实传感器图像常包含退化。论文旨在解决编解码器适应不同压缩目标和退化输入的问题。

Result: 在干净、噪声和模糊图像上，针对标准、感知和机器视觉压缩场景的实验表明，PICA在Kodak、CLIC2020和HEVC数据集上一致提升了H.266/VVC和Cheng2020编解码器的压缩性能，并优于竞争方法。

Insight: 论文的创新点在于提出多智能体协作框架实现编解码器无关的自适应压缩，特别是AROIC将编解码器视为黑盒实现区域自适应，以及SGQR评估策略。客观来看，利用智能体分工和黑盒ROI编码来提升压缩灵活性和适应性是值得借鉴的思路。

Abstract: Despite decades of progress, image codecs remain difficult to adapt to diverse real-world compression scenarios for two main reasons. First, most image codecs are optimized for a specific objective and require substantial redevelopment when the compression goal changes. Second, they typically assume clean inputs, whereas sensor-captured images often contain degradations that alter rate-quality behavior. To address these challenges, we propose Practical Image Coding Agents (PICA), a multi-agent framework that adapts existing codecs to diverse user requirements and potentially degraded inputs. PICA consists of four collaborative agents: a Perception agent that interprets user intent and analyzes image content, an Improvement agent that jointly considers bitrate and reconstruction quality to guide pre-coding restoration, a Coding agent that refines quality-control settings from global to regional levels, and an After-coding improvement agent that addresses residual degradations and compression-induced artifacts through restoration guided solely by image quality. To enable codec-agnostic regional adaptation, we introduce Agentic ROI Coding (AROIC), which treats the codec as a black box and allows the agent to automatically schedule, execute, evaluate, and reflect ROI coding plans without architectural modifications or codec-specific retraining. We integrate these four agents with a Scenario-Guided Quality-Rate (SGQR) evaluation strategy to enable more effective scenario-adaptive compression across heterogeneous codecs. Comprehensive experiments on clean, noisy, and blurred images across standard, perceptual, and machine-vision compression scenarios demonstrate that PICA consistently enhances the compression performance of both the conventional H.266/VVC codec and the neural Cheng2020 codec, outperforming competing compression methods on the Kodak, CLIC2020, and HEVC datasets.

[75] ProAlign: Progressive Positional and Prototype-Guided Alignment for Aerial-Ground Person Re-Identification cs.CVPDF

Qiao Li, Jing Chen, Cong Wu, Kun He, Ruiying Du

TL;DR: 论文针对空中-地面行人重识别（AG-ReID）中严重的视角差异、遮挡和域间差异问题，提出ProAlign网络，通过逐层渐进位置嵌入和视图感知原型对比学习来缓解层级特征提取中的位置漂移与跨视图语义不一致。方法在CARGO及两个真实AG-ReID基准上验证有效，并在CARGO上取得整体检索性能SOTA。

Details

Motivation: AG-ReID需要匹配无人机与地面监控相机拍摄的行人图像，现有方法多关注可见的视角和遮挡问题，却忽略了层级特征提取中的累积位置漂移以及跨视图特征分布的语义不一致这两个结构性挑战。

Result: 在CARGO和两个真实世界AG-ReID基准上进行了大量实验，CARGO上整体检索性能达到SOTA，相比最强竞争方法Rank-1准确率提升1.60%，mAP提升5.01%。

Insight: 创新点在于LPPE在浅层使用条件位置编码缓解局部空间畸变、深层引入可学习静态位置嵌入强化全局语义先验，以及PCL为每个身份维护空中和地面双原型并通过双原型对比目标增强跨视图语义一致性；客观看，其价值在于从位置校准和原型对齐两个层面系统处理AG-ReID的结构性错配。

Abstract: Aerial-Ground Person Re-Identification (AG-ReID) aims to match pedestrian images captured by unmanned aerial vehicles (UAVs) and ground-based surveillance cameras. The task remains highly challenging due to severe viewpoint discrepancies, frequent occlusions, and substantial domain gaps between aerial and ground imagery. Beyond these observable factors, existing methods often overlook two fundamental structural issues: cumulative positional drift during hierarchical feature extraction and semantic inconsistency across cross-view feature distributions. To address these challenges, we propose ProAlign, a Progressive Positional and Prototype-Guided Alignment Network for AG-ReID. ProAlign comprises two key components: a Layer-wise Progressive Positional Embedding (LPPE) module and a View-Aware Prototype Contrastive Learning (PCL) module. LPPE performs adaptive positional calibration throughout transformer layers by employing conditional positional encoding in shallow layers to mitigate local spatial distortions, while introducing learnable static positional embeddings in deeper layers to reinforce global semantic priors. Meanwhile, PCL maintains view-specific aerial and ground prototypes for each identity and enforces cross-view semantic consistency via a dual-prototype contrastive objective. Extensive experiments on CARGO and two real-world AG-ReID benchmarks demonstrate the effectiveness of ProAlign, achieving state-of-the-art overall retrieval performance on CARGO with improvements of 1.60% in Rank-1 accuracy and 5.01% in mAP over the strongest competing methods.

[76] Can saliency maps flag chest radiograph classifier errors at a new hospital? An external validation study cs.CV | cs.LG | eess.IVPDF

Aashish Chapagai, Angad Singh Gurtatta, Erik Cambria

TL;DR: 论文在NIH ChestX-ray14上训练CNN和ViT胸片分类器，并在训练医院及美国、越南两家外部医院评估每个模型303,768个决策，检验saliency maps能否标记个体分类错误。研究用解剖富集度评分Grad-CAM解释，并测试其在模型置信度之外检测错误的能力，发现saliency定位弱于放射科医生，且对跨院错误检测帮助很小，建议部署监测应依赖模型置信度而非saliency。

Details

Motivation: 医学图像分类器部署时需要可靠监测个体错误，saliency maps常被视为安全保障，但其能否在新医院揭示分类器错误缺乏外部验证。

Result: Grad-CAM定位病灶的命中率最高仅0.575，低于放射科医生的0.888；将解剖富集度加入基于置信度的错误检测器后，AUC提升不超过0.011，且CNN中富集度与正确性的关系跨医院反转，导致迁移检测器退化。

Insight: 论文的创新点在于对saliency maps进行大规模跨医院外部验证，并提出解剖富集度指标检验其增量错误检测价值；客观来看，其重要发现是saliency解释的临床监测价值有限且跨站点不稳定，部署监控应优先锚定模型置信度。

Abstract: Saliency maps are offered as a safeguard for medical image classifiers, but whether they reveal individual errors at a new hospital remains largely untested. We trained a convolutional network and a vision transformer on NIH ChestX-ray14 and evaluated 303,768 decisions per model at the training hospital and at two external sites, in the USA and Vietnam, with radiologist-consensus labels. We scored each explanation by its anatomical enrichment, the area-normalized concentration of attribution in a segmented thoracic region, and tested whether it detects errors beyond the model’s confidence. Grad-CAM localized findings less often than radiologists (hit rate at most 0.575 versus 0.888). Adding enrichment to a confidence-based error detector calibrated at the training hospital never raised its area under the curve by more than 0.011, and for the convolutional network the relation between enrichment and correctness reversed between hospitals, degrading the transferred detector. Model confidence, not saliency, should anchor deployment monitoring.

[77] FloMo: 3D Scene Flow as a World Action Model Intermediate for Learning from Human Video cs.CV | cs.AIPDF

Jeremy Collins, Namra Patel, Ayush Agarwal, Shitij Govil, Animesh Garg

TL;DR: 论文提出 FloMo，一个 5B 视频生成模型，将其微调为预测三维场景流和动作 token，以三维运动作为视频预训练迁移到机器人动作的中间表示。方法将场景流渲染为 RGB 并编码到与视频相同的潜空间，使预训练视频骨干能高效适配场景流预测，并在无动作人类视频和有动作机器人遥操作数据上联合训练。实验表明该运动中间表示有利于泛化到未见物体、运动基元和任务，而直接预测视频或联合预测视频会带来负面影响。

Details

Motivation: 论文认为完整视频对于动作预测是冗余信号，视频预训练迁移到机器人动作的最佳媒介应是稠密三维运动，而非视频本身；预测运动比预测视频具有更好的样本效率。

Result: FloMo 在无动作人类视频与机器人遥操作数据联合训练后，可泛化到未见物体、运动基元和任务。消融显示：用视频预测目标替代运动预测会降低分布外泛化；在运动目标外加入视频预测会降低策略成功率；联合预测时动作 token 对运动的注意力远高于视频。

Insight: 创新点在于以三维场景流作为视频预训练与机器人动作之间的世界动作模型中间表示，并通过渲染为 RGB 后编码到视频潜空间来复用预训练视频生成骨干。客观看，其核心价值是用运动而非视频作为迁移媒介，并通过注意力与消融分析论证视频预测可能浪费模型容量。

Abstract: Pretrained video generation models distill internet-scale motion priors into robot policies, but full video is a redundant signal for action prediction. We argue that dense 3D motion is the medium through which video pretraining best transfers to robot actions, and that predicting motion rather than video affords better sample-efficiency for action prediction. FloMo is a 5B video-generation model fine-tuned to predict 3D scene-flow and action tokens. Rendering scene flow as RGB and encoding it into the same latent space as video enables a pretrained video generation backbone to efficiently adapt to predict 3D scene flow. Co-trained on action-free human video and action-labeled robot teleoperation data, our model benefits from human-video pretraining and generalizes to unseen objects, motion primitives, and tasks. Replacing motion prediction with a video prediction objective degrades out-of-distribution generalization, while adding a video prediction target alongside motion lowers policy success rather than improving it. When motion, video, and action are predicted jointly, action tokens attend far more to motion than to video, suggesting that predicting video wastes model capacity. Together, these results indicate that motion effectively mediates the transfer from video pretraining to action. More information can be found at flomo-wam.github.io.

[78] VIGOR: Visual Gain Ordering for Hallucination Mitigation in Multimodal Discrete Diffusion Language Models cs.CV | cs.AIPDF

Junzhe Chen, Tian Qin, Eugenie Shi, Tianshu Zhang, Qiang Ju

TL;DR: 论文针对多模态离散扩散语言模型（dLLM）中的幻觉问题，指出其根源不仅是token预测错误，更是”过早承诺”问题：在迭代去噪过程中，部分被掩码的位置因受语言先验驱动而非视觉证据支撑而过早被解掩码，进而带偏后续生成。为此作者提出VIGOR，一种无需训练、即插即用的推理策略，通过对比原始图像条件与视觉消融两次前向的置信度差异（即”视觉增益”），对掩码位置的解掩码顺序重新排序，优先解锁真正依赖图像证据的位置。

Details

Motivation: 多模态离散扩散模型作为自回归视觉语言模型的替代方案，仍存在输出看似合理却缺乏图像依据的幻觉。作者认为现有方法只把幻觉当作token预测问题，忽略了去噪过程中解掩码顺序导致的过早承诺偏差，因此需要从推理调度层面加以纠正。

Result: 在LLaDA-V和Lumina-DiMOO两个多模态dLLM上，VIGOR在多个benchmark上一致降低了幻觉并提升了多模态推理性能；作为无需训练的方法，可直接嵌入现有模型。

Insight: 创新点在于将幻觉归因于扩散去噪的”解码顺序”而非单纯的概率分布，并提出用视觉消融的反事实对比来度量每个位置对图像证据的依赖程度，从而指导解掩码优先级。该无训练、即插即用的推理期干预思路可推广到其他迭代式生成模型。

Abstract: Multimodal discrete diffusion language models (dLLMs) generate responses through iterative mask prediction and offer a promising alternative to autoregressive large vision-language models. However, they still suffer from hallucination, producing outputs that are plausible under language priors but insufficiently grounded in the image. We argue that hallucination in multimodal dLLMs is not only a token prediction problem, but also a premature commitment problem: during iterative denoising, some masked positions are unmasked too early because their confidence is driven more by language priors than by visual evidence, and these early commitments can bias later generation. To address this issue, we propose Visual Gain Ordering (VIGOR), a training-free and plug-and-play inference strategy that reorders masked positions according to counterfactual visual evidence. At each denoising step, VIGOR compares the confidence of each candidate token under the original image-conditioned pass and an additional visual-ablation pass, and prioritizes positions whose predictions are more strongly supported by the image. Experiments on LLaDA-V and Lumina-DiMOO show that VIGOR consistently reduces hallucination and improves multimodal reasoning performance across benchmarks. Code is available at https://github.com/THU-BPM/Vigor.

[79] Manifold Embedding of Deep Image Features for Image Matching via Neural Adjoint Maps cs.CVPDF

Nafie El Amrani, Moritz Heep, Florian Bernard

TL;DR: 论文提出一种基于神经伴随映射的图像匹配方法，利用功能映射框架对深度图像特征进行流形嵌入，并通过Galerkin方法推导离散Laplace-Beltrami算子以定义图像网格的几何结构，从而在零样本密集对应任务上实现更鲁棒和准确的匹配。

Details

Motivation: 现有基于基础模型的图像对应方法（如最近邻匹配）正则化不足，导致密集对应时噪声较大。论文旨在通过功能映射引入低维线性表示和谱正则化，但需为图像网格定义有意义的流形几何。

Result: 在多个真实世界图像的密集对应基准上，该方法在零样本图像对应方法中达到新的SOTA。

Insight: 创新点包括：利用深度特征流形嵌入构建图像的几何结构，采用有限元Galerkin方法推导离散LBO，并引入神经伴随映射表示非线性映射，从而将功能映射成功扩展到图像匹配任务。

Abstract: Estimating semantic correspondences between different object instances of similar categories in real-world images is a fundamental challenge in computer vision. While the recent progress in foundation models has significantly advanced solutions for such difficult image correspondence problems, respective methods are often insufficiently regularised. For example, common nearest-neighbour matching with foundation model features often results in noisy correspondences, which is particularly prominent when solving for dense correspondences. In this work, we tackle image matching via functional maps, which have been popularised for 3D shape matching due to their powerful spectral formalism utilising an efficient low-dimensional linear representation. However, functional maps require that domains (in our cases the rectangular image grids) are equipped with an informative and non-trivial geometry. To define a meaningful manifold structure for images, we use Galerkin’s method from finite elements to derive a discrete Laplace-Beltrami operator (LBO) based on a manifold embedding of deep image features. Further, we leverage a neural adjoint map that allows to represent non-linear mappings. We demonstrate that our method sets the new state of the art among zero-shot image correspondence methods on multiple dense correspondence benchmarks of real-world images. Our code is publicly available: https://github.com/NafieAmrani/NAMI

[80] High-fidelity training-free photovoltaic module mapping from multi-view UAV imagery cs.CVPDF

WenHao Duana, Qi Chena, Penglei Zhang, Ruizhe Deng, Zhiling Guo

TL;DR: 论文提出免训练的 MV-PVMapper 框架，将 SAM 3 等视觉基础模型与无人机摄影测量结合，把多视角图像中不完整、不一致的分割结果迭代融合为高保真光伏组件实例地图。方法通过文本提示生成候选、基于坡度校正的物方轮廓验证、跨视角实例选择，并将接受实例回投为框提示进行多轮分割与融合。

Details

Motivation: 解决多视角无人机影像中，视觉基础模型零样本分割结果存在不完整和跨视角不一致，难以恢复完整且几何准确的光伏组件实例并制图的问题。

Result: 在三个屋顶光伏电站的五方向航测数据（约1 cm GSD）上，两轮反馈在IoU=0.5下恢复7,710/7,711个参考组件且无误检；宏平均面积精确率和召回率达99.00%和98.52%，聚合Jaccard指数97.55%，显著高于正射影像分割和多视角像素投票基线的60.73%和41.24%，IoU=0.95下平均精度为95.67%。

Insight: 创新点在于无需任务特定标注和模型权重更新，耦合视觉基础模型与摄影测量：利用物方坡度校正和组件尺寸约束进行跨视角验证，并通过将已接受实例回投为框提示补充空间线索，实现迭代分割、验证与融合，提升实例完整性和边界精度。

Abstract: High-fidelity photovoltaic (PV) module mapping requires complete instance recovery and accurate delineation. Vision foundation models offer zero-shot segmentation, but converting incomplete and inconsistent predictions across views into a complete, geometrically accurate map of individual modules remains challenging. We propose the Multi-View PV Module Mapper (MV-PVMapper), a training-free framework coupling vision foundation models with unmanned aerial vehicle (UAV) photogrammetry. First, Segment Anything Model 3 (SAM 3) generates candidates from original images using text prompts. Their contours are then transferred into object space through per-module slope correction to limit distortion from surface-model noise, and verified against representative module dimensions to guide cross-view instance selection. Finally, accepted instances return to all covering images as box prompts for SAM~3, supplementing text prompts with spatial cues for iterative segmentation, verification and fusion. On five-direction surveys of three operational rooftop PV plants at approximately 1 cm nadir ground sampling distance, two feedback iterations recover 7,710 of 7,711 reference modules with no false positives at an intersection-over-union (IoU) matching threshold of 0.5. Macro-averaged area precision and recall reach 99.00 % and 98.52 %, respectively. Averaged across sites, the aggregated Jaccard index reaches 97.55 %, compared with 60.73 % and 41.24 % for orthomosaic segmentation and multi-view pixel-voting baselines, respectively; average precision at the stricter IoU threshold of 0.95 reaches 95.67 %. Ablation studies show that module-size constraints are critical for instance separation, while slope correction and feedback to all covering views improve boundary accuracy. These results are achieved without task-specific image annotation or segmentation-model weight updates.

[81] Perturb, Repair, Verify: Self-Play Vision-Language Verifiers for Compositional Understanding cs.CV | cs.LGPDF

Zijun Huang, Yingjun Du, Cees G. M. Snoek

TL;DR: 论文提出PROVE，一种通过“扰动-验证-修复-再验证”闭环进行自博弈的视觉语言验证器训练框架。它从正确图文对出发，由Mutator对物体、属性、数量和关系做类型化扰动，验证器判断扰动描述并给出文本反馈，Repairer据反馈修复出忠实描述，两轮验证产生结构化奖励。由于扰动、验证、修复共享同一多模态骨干网络，训练这些角色行为也能同步提升验证器本身。

Details

Motivation: 现有视觉语言验证器训练依赖外部强评判器或人工构造难负例的静态监督，受教师能力上限约束，且难随模型失败模式演化而自适应调整。论文希望用自博弈闭环减少外部标注/生成依赖，并让验证器在 informative mismatch 与 grounded repair 中持续学习。

Result: 实验表明PROVE在不同模型规模和模型族上提升了细粒度组合式验证能力，并在参考骨干上增强了视觉细节鲁棒性；在更广泛多模态任务上仅有有限且依赖benchmark的变化，未提及具体数值或SOTA结论。

Insight: 创新点在于将类型化扰动、文本反馈修复和双轮验证组织成自博弈闭环，用结构化奖励同时监督失配识别与视觉 grounding 修复；且将Mutator、Verifier、Repairer视为同一演化多模态骨干的角色条件行为，使共享参数更新反哺验证器，避免静态教师监督并实现随失败模式演化的训练信号。

Abstract: Vision-language verifiers judge whether an image faithfully matches a textual description, yet existing training pipelines typically rely on static supervision from stronger external judges or curated hard-negative data. Such supervision requires external generation or annotation, is teacher-bounded, and is difficult to adapt as model failures evolve. We introduce PROVE (Perturb-and-Repair Optimization for Vision-Language Evaluation), a self-play framework that trains verifiers through a closed perturb-verify-repair-verify loop. Starting from a known-correct image-caption pair, a Mutator introduces a typed perturbation over objects, attributes, counts, or relations. The verifier then judges the perturbed caption and provides textual feedback, which a Repairer uses to recover a faithful caption for re-verification. These two verification rounds produce structured rewards for informative mismatches and visually grounded repairs. Since perturbation, verification, and repair are role-conditioned behaviors of the same evolving multimodal backbone, updating the shared parameters also improves the verifier itself. Experiments show that PROVE improves fine-grained compositional verification across model scales and families and strengthens visual-detail robustness on the reference backbone, with limited, benchmark-dependent changes on broader multimodal tasks.

[82] What Does the Model See? Representation Validity in Multimodal LLM-Based Video Analysis cs.CV | cs.CLPDF

Christian Pipal, Eva-Maria Vogel, Morgan Wack

TL;DR: 论文研究基于多模态大语言模型的视频分析中，输入表示选择（文本、音频、视频及组合）如何影响测量与后续分析。作者在6个开源MLLM和3个人工标注视频数据集上比较“表示有效性”，发现没有一种输入表示始终最优，且分类准确率提升不一定带来与人工编码更一致的统计估计。结论是输入表示需针对所测构念和后续估计量进行验证。

Details

Motivation: 社会科学研究者用MLLM大规模分析视频时需决定给模型提供哪些视频信息，但不同输入表示对测量和下游分析的影响尚不明确。论文旨在提出并用“表示有效性”检验文本、音频、视频及组合输入对结果的影响。

Result: 在七个新闻相关构念上，原生视频macro-F1约为0.70-0.77，提取文本约为0.52-0.80；三个模型文本更好、三个视频更好，组合输入很少提升。最高F1表示仅在17/42次比较中最好复现人工编码回归系数；即便各模型采用最高F1可部署表示，45%的样本比例估计与人工基准相差超过10个百分点。

Insight: 创新点在于提出“表示有效性”框架，系统比较多模态输入表示对测量和统计推断的影响，而非只看分类指标。客观看，其价值在于揭示高F1不等于下游估计有效，强调面向构念和估计量的输入表示验证。

Abstract: Multimodal large language models (MLLMs) allow social scientists to analyze video at scale, but researchers must decide which parts of a video to provide as input. We use the term representation validity to examine how text, audio, video, and combined inputs affect measurement and subsequent analysis. We compare input representations across six open-weight MLLMs and three human-annotated video datasets. No representation consistently performs best. On seven news-related constructs, native-video performance varies little across models (macro-F1 = .70-.77), while extracted-text performance ranges from .52 to .80. Three models perform better with text and three with video. Combining inputs rarely improves performance. Higher classification accuracy also does not reliably produce estimates closer to those based on human codes. The highest-F1 representation best reproduces human-coded regression coefficients in 17 of 42 comparisons. Even when each model uses its highest-F1 deployable representation, 45% of sample-prevalence estimates differ from the human-coded benchmark by more than ten percentage points. Input representation therefore requires validation for both the construct being measured and the quantity estimated from the resulting labels.

[83] Factorized MLLM Embeddings for Composed Image Retrieval cs.CV | cs.IRPDF

WonJun Moon, Seongchan Kim, Dahun Kim, Seungryong Kim

TL;DR: 论文针对组合图像检索（CIR）提出FactEmbed，将多模态大语言模型（MLLM）对组合查询的理解与检索判别嵌入的构建解耦。方法用一组因子化潜在表示模拟参考图到目标图的预期视觉修改，并通过中间目标想象目标、因子级对比损失和深层注意力蒸馏来学习可检索表示。推理时同时利用[EOS]全局嵌入和因子化潜在表示的均值池化结果进行高效重排。

Details

Motivation: 现有基于MLLM的CIR方法通常直接用对比监督将组合查询映射为单一检索嵌入，未显式解析修改文本所意图的视觉变化，导致MLLM的组合理解能力未被充分利用。论文希望先理解并分解组合查询中的视觉修改语义，再构建判别性检索嵌入。

Result: 摘要称在多个CIR基准上进行了广泛评估，FactEmbed优于先前方法；但未给出具体数据集名称、定量指标或明确的SOTA数值。

Insight: 创新点在于将组合查询理解与检索嵌入构建解耦，引入因子化潜在表示和中间目标想象目标来显式建模视觉修改；结合因子级对比损失与深层注意力蒸馏传播分解表示；推理时利用同一前向产生的[EOS]全局嵌入和均值池化因子表示实现低成本候选重排。

Abstract: Composed Image Retrieval (CIR) aims to retrieve a target image given a reference image and a modification text. Although recent approaches leveraging Multimodal Large Language Models (MLLMs) have shown significant progress, we argue that the MLLM’s compositional understanding remains underutilized. This stems from directly mapping composed queries into a retrieval embedding via contrastive supervision alone, without explicitly resolving the intended visual modifications. In this work, we introduce Factorized MLLM Embeddings (FactEmbed), which decouples compositional query comprehension from building discriminative embeddings. Rather than directly projecting the composed query into a single embedding without internal reasoning, FactEmbed employs a set of factorized latents that simulate the intended visual modifications via an intermediate target imagination objective. This process encourages the latents to ground into target semantics before forming discriminative representations for retrieval. Subsequently, to propagate these factorized representations throughout the model for retrieval, we optimize a factor-level contrastive loss while distilling the decomposed representations into a deeper attention layer of the MLLM. Finally, for efficient inference, we retain the [EOS] token to establish a global-level embedding on top of these factorized latents, while we exploit the mean-pooled factorized latents to rerank a short candidate list at negligible cost, as both are produced by the same forward pass. Extensive evaluations demonstrate that FactEmbed outperforms prior works across CIR benchmarks.

[84] TAQ-MOTR: Tracking-Aware Detection Query Formation for End-to-End Multiple Object Tracking cs.CVPDF

Hanyang Liu, Hong Zhang, Ding Yuan, Yang Han, Yifan Yang

TL;DR: 论文提出TAQ-MOTR，针对端到端多目标跟踪中检测查询形成缺乏轨迹状态感知的问题，设计解码器前的混合查询生成器（HQG）融合当前帧图像特征、轨迹记忆和可学习记忆，生成跟踪感知的检测查询，并引入仅训练用的辅助检测分支提供额外监督，在DanceTrack等数据集上取得性能提升。

Details

Motivation: 现有TBP方法大多沿用单帧检测的查询形成方式，在解码器前构建检测查询时对轨迹状态感知不足，导致当前未跟踪对象被遗漏，且遮挡下可见度降低的已跟踪对象缺乏互补证据，引发关联失败或重复响应。

Result: 在DanceTrack测试集上达到72.0 HOTA、63.1 AssA、76.6 IDF1，并在DanceTrack、SportsMOT和BFT数据集上均提升了性能。

Insight: 创新点在于提出跟踪感知的检测查询形成框架，利用HQG融合多源信息并以轨迹状态为条件生成查询；同时设计仅训练用的辅助检测分支，通过解耦的一对一标签分配提供额外监督。客观来看，将检测查询与轨迹状态条件化、并引入辅助分支增强未跟踪对象发现，为端到端跟踪的查询设计提供了新思路。

Abstract: Tracking-by-propagation (TBP) methods perform end-to-end multi-object tracking by propagating track queries across frames, jointly handling detection and identity association without heuristic post-processing while naturally modeling temporal information. However, as detection and association are tightly coupled, detection queries in TBP are expected to discover currently untracked objects and provide complementary current-frame evidence for challenging tracks whose queries and identities are retained. Existing methods largely inherit detection-query formation from single-frame detection and construct such queries before the decoder with little awareness of trajectory states. Consequently, currently untracked objects may be missed, while challenging tracked objects with reduced visibility under occlusion may lack complementary evidence, causing association failures or duplicate responses. To address these issues, we propose TAQ-MOTR, a tracking-aware detection query formation framework that employs a pre-decoder Hybrid Query Generator (HQG) to fuse current-frame image features, track memory, and learnable memory. By conditioning detection queries on trajectory states, TAQ-MOTR adapts them to \rev{the discovery of currently untracked objects and tracked-object complementation We further introduce a training-only auxiliary detection branch whose one-to-one label assignment is decoupled from that of the main tracking branch, providing additional detection supervision to strengthen the shared HQG and decoder for the discovery of currently untracked objects. TAQ-MOTR improves performance on DanceTrack, SportsMOT, and BFT, achieving 72.0 HOTA, 63.1 AssA, and 76.6 IDF1 on the DanceTrack test set. Code is available at https://github.com/Magicmiya/TAQ-MOTR

[85] Diffusion Plug-in Priors for Self-Supervised Ultrasound Video Restoration cs.CVPDF

Chu Chen, Mingduo He, Yining Li, Yinhao Wu, Wei Tang

TL;DR: 论文提出 UltraDiff，用于无配对数据、无需重新训练网络的超声视频联合超分辨与去斑。方法将冻结的预训练扩散模型作为即插即用先验，通过优化扩散采样器的潜变量种子，在范围约束 MAP 框架下逐帧恢复视频。作者引入基于 Rayleigh 尺度族模型的斑噪感知数据保真项、TV 正则和时序种子热启动。实验在模拟斑噪的 EchoNet-LVH 和真实临床超声心动图视频上验证了恢复质量及下游 EF 预测收益。

Details

Motivation: 超声视频受硬件分辨率限制和信号相关斑噪影响，而监督式恢复通常需要配对的低/高分辨率数据，现实中难以获得。论文希望在不使用配对训练数据、不重新训练网络的情况下，同时实现超声视频超分辨和去斑。

Result: 在 EchoNet-LVH 模拟斑噪与加性噪声视频上，使用自然图像先验的 UltraDiff 比最强基线 PSNR 提升 1.24 dB，使用超声相关先验还能进一步提升。在真实临床超声心动图视频上，零样本 EF 预测的平均绝对误差相对最强基线 DUP 从 7.31 降至 6.37 EF 点。

Insight: 创新点在于把冻结扩散模型作为通用即插即用先验，并结合斑噪感知的 Rayleigh 数据保真项、TV 正则和跨帧潜变量种子热启动，实现无真值超声视频恢复。客观看，其价值更多来自超声噪声建模、扩散先验与视频时序一致性的系统集成，而非扩散模型本身，且方法不绑定特定扩散检查点、无需配对数据和重训练，便于下游测量。

Abstract: \textit{Objective:} Ultrasound videos are limited by hardware-constrained spatial resolution and signal-dependent speckle, while paired low- and high-resolution acquisitions for supervised restoration are rarely available. We aim to jointly super-resolve and despeckle ultrasound videos without paired training data or network retraining. \textit{Methods:} We propose UltraDiff, which uses a frozen pre-trained diffusion model (DM) as a plug-in prior and restores each frame by optimizing the latent seed of the DM sampler under a range-constrained maximum a posteriori formulation. We derive a speckle-aware data-fidelity term from a Rayleigh scale-family model and characterize its relative-residual and variance-adaptive weighting properties. We further introduce a stabilized fidelity that is bounded below and strictly convex in log intensity for positive observations in its unclamped form. A total-variation regularizer complements the DM prior, and a temporal seed update warm-starts each frame from the optimized seed of its predecessor. \textit{Results:} On EchoNet-LVH videos with simulated speckle and additive noise, UltraDiff with a natural-image prior improves PSNR by 1.24,dB over the strongest baseline, and ultrasound-related priors improve it further. On real clinical echocardiography videos, UltraDiff reduces the mean absolute error of zero-shot ejection fraction (EF) prediction from 7.31 to 6.37 EF points relative to DUP, the strongest baseline. \textit{Conclusion:} Combining a speckle-aware fidelity with frozen diffusion priors enables ground-truth-free ultrasound video restoration that benefits downstream measurement. \textit{Significance:} UltraDiff needs neither paired data nor retraining and is not tied to a specific diffusion checkpoint. Code associated with this work will be made publicly available upon acceptance.

[86] Route Anything: A Study of Model and Policy Routing for General-Purpose AI Systems cs.CVPDF

Pengfei Zhou, Yixing Ma, Hexin Wang, Weidong Tang, Zhenglin Wan

TL;DR: 论文提出 Route Anything 统一框架，研究面向通用人工智能系统的模型与策略路由，即如何为不同任务动态选择专用模型、工具或策略以平衡性能与计算成本。作者建立了质量—成本权衡下的路由理论基础，并构建覆盖25个基准、452个候选配置、374,401次评分运行的全模态轨迹矩阵，用于训练和评测路由器。实验表明，逐输入路由能显著优于最佳固定模型，并在多类任务上提升效率。

Details

Motivation: 通用人工智能需要跨领域、跨模态地理解、生成、推理和行动，单一模型难以掌握所有能力，因此需要编排异构专用模型、工具和策略。论文关注的核心问题是：系统如何动态选择并协调最有能力的候选，同时在性能与计算成本之间取得平衡。

Result: 逐输入 oracle 路由比最佳固定候选平均高5.4分；学习路由器在16个基准上高0.3分、视频生成高2.1分、机器人操作高1.1分，且策略时间减少24%。相比参考模型，路由平均带来8.82倍效率提升。

Insight: 论文宣称的创新包括统一的质量—成本路由理论、全模态轨迹矩阵基准，以及系统研究学习路由器的能力和路由提升效率的条件。客观来看，其价值在于将模型路由从单一模态扩展到多模态与智能体/机器人策略，并通过大规模评测验证了能力编排相对于固定模型的潜力。

Abstract: Artificial general intelligence (AGI) requires a method that can understand, generate, reason, and act across diverse domains and modalities. Rather than relying on a single model to master every capability, a promising path toward AGI now is to orchestrate a heterogeneous ecosystem of specialized models, tools, and policies (candidates) into a unified AI system. At the heart of this vision lies a fundamental question: how can an AI system dynamically select and coordinate the most capable candidate for each task while balancing performance and computational cost? In this work, we present Route Anything, a unified framework for studying model and policy routing across general-purpose AI systems. We develop a unified theoretical foundation that characterizes optimal routing under quality–cost trade-offs and establishes how the information available to a router determines its attainable performance. To systematically investigate routing across diverse capabilities, we first construct an omni-modal trajectory matrix spanning 25 benchmarks across language understanding, image, video, and speech generation, coding and workplace agents, and robotic manipulation, comprising 374,401 scored runs from 452 candidate configurations. Building on this framework, we train and test different routers to investigate the potential of omni-modal model routing, the capabilities of learned routers, and the conditions under which routing improves efficiency. Our results reveal substantial opportunities for capability orchestration: per-input oracle routing outperforms the best fixed candidate by 5.4 points, and learned routers outperform it by 0.3 points across 16 benchmarks, by 2.1 points on video generation, and by 1.1 points on robot manipulation while using 24% less policy time. More importantly, routing raises efficiency by 8.82x on average compared with the reference model.

[87] Möbius: A Unified Motion-Language Model Bridging Understanding and Generation cs.CVPDF

Jiazhi Shu, Chengfeng Zhao, Jinyi Wang, Yubo Zhao, Zhiyang Dou

TL;DR: 论文提出 Möbius，一个统一运动-语言模型，在单一框架内同时支持人体运动理解、生成和编辑。它将扩散过程引入与运动理解共享的多模态表示空间，让干净运动、噪声运动和文本共同参与注意力计算，并直接在连续帧级原始运动上操作，无需离散 token 或预训练运动 VAE。

Details

Motivation: 现有统一运动模型只是部分统一：理解依赖语言模型，生成仍依赖独立的扩散模块和运动潜变量，框架割裂且流程复杂。论文希望构建真正统一的模型，在同一表示空间中完成运动理解、生成与编辑。

Result: 在 HumanML3D 上，Möbius 在统一运动模型中达到 state-of-the-art 性能；扩展到 1.7B 参数和 1700 小时运动数据后，仍验证了框架的有效性和可扩展性。

Insight: 创新点在于将扩散生成过程统一到运动理解的多模态表示空间中，并直接用连续原始运动建模，避免离散 tokenization 和预训练运动 VAE；通过共享注意力与角色特定参数统一多任务。客观来看，这种设计有望减少理解与生成之间的表示鸿沟，并提升统一模型的多任务扩展性。

Abstract: We aim to build a truly unified motion model that can understand, generate, and edit human motion within a single framework. Existing motion UMMs remain only partially unified: motion understanding is handled by a language model, while generation is often delegated to a separate diffusion module operating over learned motion latents. We introduce Möbius, which brings the entire diffusion process into the same multimodal representation space used for motion understanding. Clean motion, noisy motion, and text jointly participate in shared attention, while role-specific parameters process understanding and generation streams. Importantly, Möbius operates directly on continuous frame-level raw motion, requiring neither discrete tokenization nor a pretrained motion VAE. This simple formulation naturally supports motion captioning, text-to-motion generation, temporal completion, and instruction-guided editing within a single model. On HumanML3D, Möbius achieves state-of-the-art performance among unified motion models. Scaling to 1.7B parameters and 1,700 hours of motion data further demonstrates that our framework is both effective and scalable.

[88] YANAMI: A Large-Scale Multimodal Dataset and Benchmark for Multiturn Stateful Interaction cs.CVPDF

Shengyu Wang, Tengyue Nong, Junming Liu, Haosong Peng, Siyu Li

TL;DR: 论文提出了YANAMI，首个大规模多模态对话数据集和基准，用于多轮状态交互。包含YANAMI-400K（40万对话、200万图像、260万轮次，覆盖多图像融合、图像编辑、世界建模、视觉推理四个场景）和YANAMI-Bench（400个多轮对话）。实验表明在YANAMI-400K上微调BAGEL可提升图像编辑、生成和多模态状态交互性能，但现有统一多模态模型在状态维护与视觉实现上仍有困难。

Details

Motivation: 现有统一多模态模型的数据集和基准以单轮任务为主，即使多轮数据集也常可独立解决各轮，缺乏对跨轮对话状态跟踪、维护和更新的监督与评估，因此需要专门针对多模态状态交互的大规模数据集和基准。

Result: 在YANAMI-400K上微调BAGEL后，图像编辑提升10.3%，图像生成提升2.0%，多模态状态交互提升13.23%，同时保持通用视觉理解能力；YANAMI-Bench评估显示现有统一多模态模型仍难以维护、更新和视觉实现演变的对话状态。

Insight: 论文的创新点在于构建首个大规模多模态状态交互数据集和基准，涵盖四个代表性场景，并联合评估对话状态理解与视觉实现，揭示模型理解状态与忠实生成视觉输出之间的差距，为后续研究提供重要基准和训练资源。

Abstract: Unified Multimodal Models (UMMs) integrate visual understanding, generation, and editing within a single unified framework. However, existing unified multimodal datasets and benchmarks remain dominated by single-turn tasks. Even in recent multi-turn datasets, many turns can be solved independently without tracking information across turns. Consequently, they provide limited supervision and evaluation for multimodal stateful interaction, which requires models to understand, maintain, update, and faithfully realize evolving dialogue states across turns. To address this gap, we introduce YANAMI, the first large-scale multimodal dialogue dataset and benchmark for stateful interaction, comprising YANAMI-400K and YANAMI-Bench. YANAMI-400K contains 400K multi-turn multimodal dialogues, with more than 2M images and 2.6M dialogue turns. The dialogues span four representative scenarios: Multi-Image Fusion, Image Editing, World Modeling, and Visual Reasoning, and require models to continuously track, preserve, and revise evolving multimodal dialogue states across turns. YANAMI-Bench comprises 400 multi-turn dialogues for systematically evaluating multimodal stateful interaction. It jointly evaluates Dialogue State Understanding and Visual Realization, characterizing the gap between models’ ability to understand evolving dialogue states and their ability to faithfully realize them in visual outputs. Experiments show that fine-tuning BAGEL on YANAMI-400K improves image editing, generation, and multimodal stateful interaction by 10.3%, 2.0%, and 13.23%, respectively, while preserving its general visual understanding capabilities. Evaluations on YANAMI-Bench further reveal that existing UMMs still struggle to maintain, update, and visually realize evolving multimodal dialogue states.

[89] Low-Density 3D Point Cloud Classification cs.CVPDF

Ahmed Baha Ben Jmaa, Faten Chaieb

TL;DR: 论文关注真实场景中3D点云密度变化导致的分类难题，指出现有方法在稠密点云上表现良好但在低密度下性能下降。作者首先系统评测了多种先进点云分类方法在不同密度下的表现，并提出面向低密度点云分类的LD-PointNet++架构。

Details

Motivation: 真实传感技术限制和环境变化会使点云密度降低，现有分类方法难以获得完整几何表示，在低密度场景下性能明显下降，因此需要提升低密度点云分类的鲁棒性。

Result: 在ModelNet基准数据集上，LD-PointNet++在低密度场景下优于现有先进方法，并在不同密度水平下保持超过90%的稳定准确率。

Insight: 论文的创新点在于首次系统基准评测不同密度下的点云分类方法，揭示低密度性能退化问题，并设计针对低密度点云优化的LD-PointNet++，强调跨密度分类一致性。

Abstract: 3D point cloud classification, a task in computer vision, has recently gained substantial interest due to its extensive applications across various domains such as augmented and virtual reality, robotics, and autonomous driving. However, classifying 3D point clouds becomes challenging in real-world scenarios where the density levels vary. These variations are due to limitations inherent to current sensing technologies and variable environmental conditions, which often prevent the capture of detailed and comprehensive geometric representations. Existing classification methods, which have demonstrated impressive results on dense point clouds, may be affected and could experience reduced performance when dealing with low-density point clouds. Addressing this challenge, this paper introduces LD-PointNet++, an architecture optimized for the classification of low-density point clouds. To pave the way for LD-PointNet++, we conduct the first benchmarking study that systematically evaluates state-of-the-art point cloud classification methods across varying densities, revealing significant performance declines in low-density environments. Experimental results on the ModelNet benchmark dataset demonstrate that LD-PointNet++ not only outperforms these methods in low-density scenarios but also demonstrates consistent accuracy above 90% across different density levels.

[90] GSCV++: Lightweight 3D Gaussian Splatting Sequence Compression via Standard Video Codecs cs.CVPDF

Qi Yang, Yingzhan Xu, Kai Zhang, Li Zhang

TL;DR: 论文提出 GSCV++，一种基于标准视频编解码器的轻量级 3D Gaussian Splatting（GS）序列压缩流水线。方法联合优化预处理、3D 到 2D 映射和实际视频压缩配置，分别引入 EA-SHDR、P-MiniPLAS 和 CA-GC 模块，以降低复杂度并提升与硬件视频编解码器的兼容性。实验表明其在率失真性能和编解码效率上优于现有 SOTA 方法。

Details

Motivation: 面向工业应用对低复杂度和硬件兼容性的需求，解决 3D GS 序列压缩中数据量大、2D 映射不適合视频编码器以及实际压缩配置未充分优化的问题。

Result: 实验显示 GSCV++ 相比 SOTA 方法在帧内预测取得超过 20%、帧间预测取得超过 60% 的率失真增益，并在同类视频编解码器方案中实现更快的 2D 图生成、编码和解码。

Insight: 创新点在于联合优化 GS 压缩全流程：EA-SHDR 自适应降低帧间 SH AC 维度以构建紧凑稳定的时序特征空间；P-MiniPLAS 通过先验加权和小尺度重映射生成更适合视频编码的平滑 2D 特征图；CA-GC 采用非均匀位深拆分和编码单元尺寸选择，在不改变视频编解码规范的前提下提升压缩性能并控制复杂度。

Abstract: This paper presents GSCV++, a lightweight pipeline for 3D {G}aussian {S}platting (GS) sequences {C}ompression with standard {V}ideo codecs. To meet the requirements of low-complexity and hardware compatibility from the industry, GSCV++ jointly improves preprocessing, 3D-to-2D mapping, and practical video compression configuration of GS compression. First, for preprocessing, energy-adaptive inter-frame spherical harmonic (SH) alternative current (AC) dimensionality reduction (EA-SHDR) is proposed to produces a compact and stable temporal SH AC feature space. It can reduce the data volume of following mapping and practical compression, as well as less complexity while eliminating redundant information. Second, for 3D-to-2D mapping, prior-based Inter-MiniPLAS (P-MiniPLAS) is designed to generate smooth 2D feature maps through prior-based weighting factor inference and small scale remapping. Compared with vanilla Parallel Linear Assignment Sorting (PLAS) and Morton scan, P-MiniPLAS ensures the generated 2D maps are more suitable for the video codec while maintaining a low level of computational complexity. Third, for practical compression, content-aware generalized compression (CA-GC) is developed to use a non-uniform bit-depth split for coordinate map generation, and a coding unit size selection is introduced for efficient compression complexity control, which fully exploits the performance of video codecs without involving normative changes. Experiments demonstrate that GSCV++ reports over 20% in intra- and 60% in inter-prediction rate-distortion gains compared with state-of-the-art (SOTA) methods. GSCV++ also achieves faster 2D map generation, encoding, and decoding than other SOTA methods using the same type of video codecs. Ablation studies verify the effectiveness and robustness of the proposed modules and provide insights for subsequent research.

[91] DreamX: Video Diffusion Model is a Universal Latent Action Learner cs.CV | cs.AI | cs.ROPDF

Jisoo Kim, Chaehyun Kim, Junsu Kim, Jiho Park, Seongchan Kim

TL;DR: DreamX 将预训练视频扩散模型改造为通用潜在动作学习器，从视频模型时空特征中提取紧凑且与视觉外观解耦的潜在动作，并将每帧潜在视频从1200个token压缩到4个token（300倍）。该潜在动作空间可作为跨人类/机器人、环境和动作参数化的共享动作接口，并支持视频迁移、零样本动作迁移及紧凑潜空间策略学习。

Details

Motivation: 现有潜在动作模型从视觉转移中推断动作时，表示容易将真实动作与无关视觉变化纠缠在一起，导致跨具身迁移能力受限；论文希望学习对同一动作动态一致、对具身和外观变化不变的通用动作表示。

Result: 在LARY-Bench上，DreamX在潜在动作模型中达到SOTA；其潜在空间支持跨具身、跨环境的any-to-any视频迁移和零样本动作迁移，并能降低基于视频扩散模型的策略学习计算成本。

Insight: 创新点在于复用预训练视频扩散模型的丰富时空特征与生成先验来学习紧凑、解耦的潜在动作，并将其统一为跨具身共享动作接口；客观看，将视频生成模型转化为动作表示学习器、实现高压缩率并兼顾迁移与策略学习，是其可借鉴之处。

Abstract: Learning a universal action representation from video requires modeling dynamics that are consistent across different executions of the same action yet invariant to changes in embodiment and appearance. Existing latent action models infer actions from visual transitions, but their representations often entangle the underlying action with irrelevant visual variations, limiting transfer across embodiments. We introduce DreamX, which repurposes a pretrained video diffusion model as a universal latent action learner. Our approach extracts compact latent actions from the video model’s rich spatiotemporal features, compressing each latent video frame by $\boldsymbol{300\times}$ ($\boldsymbol{1{,}200 \rightarrow 4}$ tokens). In addition, by leveraging the video model’s generative prior, these latent actions capture action-relevant dynamics disentangled from visual appearance. The resulting latent space serves as a shared action interface across embodiments (human and robot), environments, and action parameterizations. On LARY-Bench, DreamX achieves state-of-the-art performance among latent action models. Moreover, actions encoded in one setting remain effective in another: the latent space supports any-to-any video transfer and zero-shot action transfer across diverse embodiments and environments. Furthermore, this shared interface also enables policy learning in a compact latent action space, reducing the computational cost of video-diffusion-based policies.

cs.AI [Back]

[92] Alignment Faking Depends on the Value, Not Just the Model cs.AI | cs.CLPDF

Manodeep Ray, Vivek Srivastava, Shirish Karande, Murari Mandal

TL;DR: 论文研究大语言模型中的“Alignment Faking”现象，即模型在被监控时遵从指令、在认为无人监控时放弃指令。作者认为该行为不仅取决于模型本身，还取决于具体价值维度，因此提出按“模型—价值”配对进行画像：在32个ValueBench维度、10个开源/闭源模型（8B~675B）上，于无系统提示、无监控、有监控三种条件下评估合规性差距，并分析推理轨迹。

Details

Motivation: 已有工作主要在有害内容和少量道德基础价值上证实Alignment Faking，尚不清楚它在更广价值空间中的分布规律。论文旨在说明合规行为是否在感知监控变化后保持，是模型与价值配对的属性，而非模型单独属性。

Result: 在32个ValueBench维度和10个模型上，合规性差距范围为0.05~0.50；一个蒸馏70B模型可超过约675B模型。结构方差主要来自“哪个模型被问及哪个价值”的交互，而非模型或价值单独因素；推理轨迹分类显示监控下训练信号推理占主导，无监控下价值冲突推理占主导。

Insight: 创新点在于将Alignment Faking从模型层面细化为逐价值的模型—价值配对评估，并提出三条件评估框架、六类推理轨迹分类和value cards，将行为倾向补充到模型卡中。客观看，其可借鉴之处是强调模型行为评估需考虑价值维度交互，并用推理轨迹解释监控感知下的行为变化。

Abstract: Language models can follow a directive when they believe they are monitored and abandon it when they believe they are not, a pattern termed Alignment Faking. Prior work establishes this for harmful-content objectives and a handful of moral-foundations values, leaving open how it distributes across a wider value space. We argue it is value-dependent: whether compliance survives a change in perceived oversight is a property of the model–value pairing, not of the model alone. Our framework profiles Alignment Faking per value, scoring each model-value pair under three conditions: bare (no system prompt), unmonitored, and monitored. On 32 ValueBench dimensions across ten open- and closed-weight models (8B-~675B), compliance gaps span 0.05-0.50, with a distilled 70B above a ~675B model. We observe that majority of the structural variance depends on which model was asked about which value and not on the model or value alone. A six-category taxonomy over reasoning traces tracks the behavioural shift: training-signal reasoning dominates under monitoring and value-conflict reasoning dominates when unmonitored. We package per-model results as value cards, extending model cards with behavioral disposition. Code and data are available at https://github.com/respailab/VALET.

[93] Functional-Group Knowledge Is Structured in Language Models Before It Is Spoken cs.AI | cs.CLPDF

Tran Nhat Khang Nguyen, Anh Nguyen, Obed Issakah, Janghoon Ock

TL;DR: 论文研究语言模型在回答分子官能团问题前，内部是否已结构化编码官能团知识。作者分析五个开源模型在覆盖92个分子和20类官能团的提示上的内部表示，发现即使未经化学训练，模型也会按化学家族组织官能团，并将化学差异编码为一致方向；这种结构在中层突然出现。基础模型正确命名率仅58.5%，但其错误多为未能作答而非答错，说明输出评估会低估模型内部化学知识。

Details

Motivation: 仅凭语言模型输出无法区分模型是缺乏化学知识，还是识别了官能团却未能正确表达；论文希望从内部表示层面判断化学语言模型真正编码了什么知识。

Result: 在5个开源模型、92个分子和20类官能团上，基础模型正确命名率最低，仅58.5%，但其内部表示仍能按化学家族组织官能团并保留有意义的化学模式，且错误主要是失败回答而非错误答案。

Insight: 创新点在于用内部表示分析区分“不知道”和“说不出”，并发现官能团知识在模型中层以结构化、方向一致的方式涌现；化学微调会重组表示，使官能团而非提示措辞主导，这为化学agent的模型选择与适配提供了表示层面的评估视角。

Abstract: Language models serve as the reasoning core of agents for chemistry and materials science, where they must interpret molecular structure before planning syntheses or proposing materials. Yet their outputs show only what they say, not what they know. Given a prompt such as The primary functional group in 1-Fluorobutane (CCCCF) is'', a small general-purpose model may answer the alkyne group’’ instead of an alkyl fluoride. The output alone cannot tell whether the model failed to recognize the fluorine or recognized it and failed to say so. We address this by reading the internal representations of five open-source models on prompts covering 92 molecules and 20 functional groups. Even without chemistry training, the models organize functional groups by chemical family, and they encode chemical differences as consistent directions. The shift from thioether ($\mathrm{R{-}S{-}R’}$) to thiol ($\mathrm{R{-}SH}$), added to ether ($\mathrm{R{-}O{-}R’}$), lands nearest alcohol ($\mathrm{R{-}OH}$). This structure emerges abruptly in the middle layers, where representations shift from tracking the wording of the prompt to tracking the functional group. The base model names the correct functional group in only 58.5% of prompts, the lowest of the five, yet it encodes functional-group families and preserves chemically meaningful patterns. Its errors are mostly failures to answer rather than wrong answers. Functional-group knowledge is structured in the model before it is spoken. Chemistry fine-tuning reorganizes it, making functional groups rather than prompt wording dominate the representation. These results suggest that evaluating chemical language models through outputs alone can underestimate the chemical information they encode. Representation analysis can distinguish a lack of knowledge from a failure to express it, which matters when choosing or adapting models for chemical agents.

[94] LOLPlay: Benchmarking Real-Time Visual Agents in League of Legends cs.AI | cs.CVPDF

Zhaojian Yu

TL;DR: 论文提出 LOLPlay，一个基于《英雄联盟》原生环境的实时视觉智能体基准，包含 36 个英雄特定任务，覆盖资源管理、战斗、合作和目标等类别。智能体仅通过截图观察并通过鼠标键盘动作交互，基准提供自动初始化和 82 个可执行子目标检查，用于区分部分进展与任务完成。论文评估了三个模型在三种推理强度下的表现，发现当前视觉智能体在实时控制中仍面临推理与及时行动协调的挑战。

Details

Motivation: 实时视觉智能体需要在环境持续变化时协调推理与行动，但现有评测难以系统衡量其在复杂、动态、长时程游戏任务中的部分进展和实时控制能力。论文旨在构建原生《英雄联盟》基准，推动视觉推理与及时行动结合的智能体研究。

Result: 在 324 次任务评估中，GPT-6 Astra 在低/中推理强度下取得最高任务成功率 25.0%，在中推理强度下取得最高子目标成功率 40.3%；但对所有模型而言，高推理强度均未优于低推理强度。对 Astra 而言，更长请求导致交互轮数减少，表明额外推理在实时控制中可能带来成本。

Insight: 论文的创新点在于构建原生《英雄联盟》实时视觉智能体基准，设计英雄特定任务与可执行子目标检查，并系统评估推理强度对实时控制的影响。其发现揭示了子目标组合不完整、动作在场景变化后执行等问题，强调需要将视觉推理与及时行动更紧密结合的智能体设计。

Abstract: Visual agents need to coordinate reasoning and action while their environment evolves during inference. We introduce LOLPlay, a benchmark in native \emph{League of Legends} with 36 champion-specific tasks spanning resource management, combat, cooperation, and objectives. Agents interact through screenshots and mouse–keyboard actions; automated initialization and 82 executable subgoal checks distinguish partial progress from completion. Across 324 task evaluations of three models at three reasoning-effort settings, GPT-6 Astra achieves the highest task success rate of 25.0% at low and medium effort, and the highest subgoal success rate of 40.3% at medium effort. However, high effort improves neither metric over low effort for any model. For Astra, longer requests leave fewer interaction rounds, suggesting a cost to additional reasoning in real-time control. Trajectories reveal incomplete subgoal composition and actions executing after the observed scene changes, motivating agents that combine visual reasoning with timely action.

[95] WASP-Mem: Let Agent Organize Procedural Memory. Human Spend Effort on Diagnosis cs.AI | cs.CL | cs.LGPDF

Peizhi Niu, Ge Shi, Vaibhav Gorde, Sarthak Ghosh, Christos Faloutsos

TL;DR: 论文提出WASP-Mem方法，让LLM agent将执行轨迹和反馈提炼为相互链接的wiki页面以组织程序性记忆，实现无需训练的持续学习。在TheAgentCompany基准上，相比最强基线提升通过率4.6-6.1%且存储减少57-61%。研究发现人类诊断比自我判断更有效，主张让agent组织记忆、人类专注诊断的分工。

Details

Motivation: LLM agent难以自动化复杂企业流程，因为静态预构建框架无法捕捉专有策略、未记录行为或状态依赖的恢复路径。程序性记忆虽可从经验中学习，但需将重复成功和诊断失败整合为可重用动作与恢复策略。

Result: 在TheAgentCompany基准上，WASP-Mem相比最强基线（三个模型）将检查点通过率提升4.6-6.1%，同时存储需求减少57-61%。单轮人类诊断比四轮自我判断高出7.0-10.9%。

Insight: 创新点在于用相互链接的wiki页面组织程序性记忆，并发现规定性知识组织（如人工定义schema和显式去重规则）反而损害性能，而人类诊断能有效将失败轨迹转化为纠正性知识。可借鉴之处：让agent自主组织记忆，人类精力应集中在诊断环节。

Abstract: LLM agents struggle to automate complex enterprise procedures because static, pre-built harness cannot capture proprietary policies, undocumented behaviors, or state-dependent recovery paths that surface only during execution. Procedural memory accrued from agent experience offers a training-free path to closing this gap – yet unlike episodic records or semantic facts, it must consolidate repeated successes and diagnosed failures into reusable actions and recovery strategies. We introduce WASP-Mem (Wiki-based Agent-Synthesized Procedural Memory), a nimble but effective method that distills execution traces and feedback into persistent, interlinked wiki pages, enabling continual agent learning. On TheAgentCompany benchmark, WASP-Mem raises the checkpoint pass rate by 4.6-6.1% over the strongest baseline we compared across three models while requiring 57-61% less storage. We further present a systematic controlled study spanning consolidation method, feedback type, trace ingestion scope, and refinement rounds. Our results reveal that prescriptive knowledge organization such as human defined schemas and explicit deduplication rules hurts performance rather than closing the procedural-knowledge gap. In contrast, human diagnosis converts failed traces into corrective knowledge and drives sustained improvement: a single round of human diagnosis outperforms four rounds of self-judgment by 7.0-10.9%. These findings point to a simple division of labor: let agents organize procedural memory, and invest human effort where it matters most – diagnosis.

[96] AlgoPilot: Cross-Paradigm Reasoning in Language Models via Strategy Selection and Guidance cs.AI | cs.CL | cs.LGPDF

Yilun Hao, Ruixiao Yang, Hudson Hilal, Yongchao Chen, Kaizhi Qian

TL;DR: 论文提出 AlgoPilot 跨范式推理框架，使语言模型在求解前自适应选择解题策略并生成结构化问题形式化。框架包含策略选择模型 SteerLM、结构化指导模型 GuideLM 和执行模型 ExecutionLM，基于 Qwen3-8B 在多类推理任务上验证。

Details

Motivation: 现实问题涵盖程序合成、符号推理、规划和优化等不同领域，往往需要不同求解范式；LLM 的核心挑战在于既要选对解题方法，又要正确形式化问题。

Result: 在 5 类问题、13 个 benchmark、66 个领域上，AlgoPilot 将同模型平均准确率从 38.4% 提升至 72.9%，并比最佳大 LLM prompting 基线 GLM-5 高 11.7%；消融验证各组件有效，且策略选择与形式化模块可迁移到其他执行模型。

Insight: 创新点在于执行前进行跨范式策略选择与结构化问题形式化，并通过 SFT/RL 学习自适应选择而非依赖固定推理模板；客观看，将策略选择、形式化指导和执行解耦并验证模块可迁移性，是其较有价值的设计。

Abstract: Real-world problems span diverse domains, including program synthesis, symbolic reasoning, planning, and optimization, and often require fundamentally different solution paradigms. A central challenge for LLM-based reasoning is twofold: identifying an appropriate problem-solving approach and formulating it correctly. In this paper, we propose AlgoPilot, a cross-paradigm reasoning framework that enables adaptive selection of the solving approaches and structured problem formulation prior to execution. The AlgoPilot framework contains three components: 1) a SteerLM, trained via supervised fine-tuning (SFT) and reinforcement learning (RL), that adaptively selects the solving strategy (paradigm and modes) based on input problems, 2) the corresponding expert GuideLM, trained via SFT, that generates structured formulation guidance tailored to the selected strategy, and 3) an ExecutionLM that generates solutions conditioned on both the selected strategy and its formulation. Unlike prompting-based methods that rely on fixed reasoning templates, our approach learns to adaptively choose and structure problem-solving strategies based on the input. We build AlgoPilot on Qwen3-8B and evaluate across 5 problem categories and 13 benchmarks with 66 unique domains, comparing against both the latest similarly scaled small models and large LLMs. AlgoPilot significantly improves average accuracy for the same model from 38.4% to 72.9%, outperforming the best large LLM prompting baseline GLM-5 by 11.7%. Ablation studies confirm the effectiveness of each component. Notably, we show that the learned steering and formulation modules can transfer across execution models, suggesting their potential to improve other LLMs. Project page: https://sites.google.com/view/algopilot.

[97] Think Less, Locally: Local Preference Learning for Reasoning Compression cs.AI | cs.CLPDF

Muyao Yuan, Liyan Xu, Weizhan Zhang, Muyan Jiao, Jiangnan Li

TL;DR: 论文提出 LoCo，一种面向大推理模型推理压缩的局部偏好学习方法。该方法从模型自身推理轨迹中检测候选“过度思考”状态，固定对应推理前缀并从同一状态重新生成更短且经验证的后续推理，构造状态条件偏好对并用 DPO 训练，使模型学会在当前状态下判断应继续推理还是结束。

Details

Motivation: 大模型长思维链在获得足够信息后仍继续推理，导致推理成本高；现有压缩方法多从整体轨迹入手，较少关注给定当前状态是否需要继续推理的局部决策。

Result: 在 5 个数学和 3 个代码生成 benchmark 上，LoCo 将数学任务完成长度降低 41.7%–45.4%、代码任务降低 23.4%–36.3%，同时保持或提升准确率，并在所有代码 benchmark 上取得最佳 AUC；仅用数学数据训练也能迁移到代码生成。

Insight: 创新点在于将推理压缩建模为状态级“继续/结束”的局部偏好学习，而非简单鼓励短轨迹；通过冻结前缀、重生成短且验证的后续并构造 DPO 偏好对，实现更细粒度的推理效率优化。

Abstract: Large reasoning models achieve strong performance on challenging tasks, but their long chain-of-thought often incurs substantial inference costs by continuing to deliberate after sufficient information has been obtained. Existing reasoning compression methods primarily operate on reasoning trajectories, paying less attention to the local decision of whether further reasoning is necessary given the current state. We propose LoCo, a local preference learning method for reasoning compression. LoCo detects candidate overthinking states from the model’s own traces, freezes the corresponding reasoning prefixes, regenerates shorter verified continuations from the same states, and trains on the resulting state-conditioned preference pairs with DPO. This formulation teaches the model when to continue reasoning and when to conclude, rather than simply encouraging shorter trajectories overall. We construct training data entirely from self-generated mathematical traces and evaluate LoCo on five mathematical and three code generation benchmarks. LoCo reduces completion length by 41.7%–45.4% on mathematics and 23.4%–36.3% on code, while maintaining or improving accuracy and achieving the best AUC across all code benchmarks. LoCo consistently improves the accuracy-efficiency trade-off on mathematics and transfers effectively to code generation despite math-only training, demonstrating the cross-domain generality of local preference learning for reasoning compression. The code and model will be publicly released at https://github.com/MuyaoYuan/LoCo.

cs.CR [Back]

[98] Voluntary Servitude: Weaponized Safety Warning Mechanisms for Controlling AI Agents cs.CR | cs.AI | cs.CL | cs.LGPDF

Ilan Zendel, Matan Levy, Stav Cohen, Avi Mendelson, Amit LeVi

TL;DR: 论文提出一种针对LLM智能体的“自愿奴役”攻击：不注入显式指令，而是伪造普通评论中的安全/危害警告，利用智能体自身的避害机制使其排除合法选项并选择攻击者目标。作者在六个模型、十个消费服务领域和三类攻击上评测，显示其效果与指令注入相当且可绕过检测器，并可迁移到浏览器智能体和检索流水线。还提出在智能体指令中加入“需佐证”要求，可在非自适应攻击下减少约95%的暴露。

Details

Motivation: 现有安全威胁多被建模为指令注入，防御依赖识别命令语法；论文关注攻击者能否利用智能体自身的安全/避害推理而非注入命令来操控决策。

Result: 攻击成功率60.4%，接近指令注入的66.0%；经检测器过滤后有效性强1.7–4.4倍，并可迁移至生产浏览器智能体与检索管线，端到端成功率为1.0–57.0%。加入佐证要求后在非自适应攻击下减少约95%的暴露。

Insight: 创新在于揭示“安全警告”本身可作为攻击面，攻击无需显式指令即可劫持智能体推理；防御上通过要求多源佐证而非仅检测命令语法来缓解。

Abstract: LLM agents increasingly act on information retrieved from public sources they treat as credible, which turns a manipulated recommendation into a transaction, a dependency, or a decision with physical consequences. The security literature models this threat as instruction injection, in which an adversary writes a command into retrieved content and deployed defenses learn to recognize command syntax. We show that an attacker can instead steer an agent through the agent’s own harm-avoidance. A single fabricated hazard, written as an ordinary review, leads the agent to disqualify the legitimate options and select the attacker’s target by its own reasoning, with no command for a detector to recognize. We call this voluntary servitude. These payloads steer agents about as often as instruction injection (60.4% vs. 66.0%), and deployed injection detectors score them as benign, so behind a filter they are 1.7-4.4x more effective. We measure this at the generation stage across six models and ten consumer-service domains, for three attack families, and show that it transfers to production browser agents, outperforming instruction injection, and to a retrieval pipeline (1.0-57.0% end-to-end success, depending on the retriever). We then show that a corroboration requirement added to the agent’s instructions removes roughly 95% of the exposure against a non-adaptive attacker without retraining. With the requirement in place, the agent still reads the planted hazard and no longer acts on it.

cs.DC [Back]

[99] EpiLoRA: Fusing Many LoRA Adapters into Low-Precision GEMMs for Serving and RL cs.DC | cs.CL | cs.LGPDF

Hieu Tran, Hong Yu

TL;DR: 论文提出 EpiLoRA，一个面向 vLLM 引擎的插件，用于将 LoRA 适配器的降维投影（down-projection）提前执行，并让低精度 GEMM 作为依赖内核在收尾阶段（epilogue）为每个租户加上升维投影（up-projection），从而把多租户 LoRA 的小内核开销融入低精度 GEMM。该方法同时适用于推理服务与 LoRA 强化学习训练，可在 FP8/4-bit 基座上高效服务多租户，并支持在单个 48GB GPU 上训练 27B 模型。

Details

Motivation: 基座模型转向 FP8/4-bit 后 GEMM 本身变便宜，但每租户 LoRA 适配器仍以独立小 GPU 内核运行，其延迟在解码每步中占比越来越大，成为多租户服务与 LoRA RL 训练的瓶颈。

Result: 在 NVIDIA Hopper 上，单适配器解码每 token 时间比 vLLM 最快配置低最多 24%，并快于 SGLang 和 TensorRT-LLM；多租户下适配器内核在多数 GPU-模型组合上吞吐超过 vLLM。LoRA RL 训练可直接在引擎 FP8 权重上计算，无需 16-bit 副本，27B 模型可在单张 48GB GPU 上训练。

Insight: 核心创新是观察到 GEMM 仅在收尾阶段需要降维投影结果，因而先启动降维投影，再让低精度 GEMM 作为依赖内核边流式加载权重边在 epilogue 中融合各租户升维投影，消除独立小内核；同时打通推理与 RL 训练的低精度权重复用，减少生成与训练模型间的不匹配。

Abstract: Low-rank adapters (LoRA) let one base model serve many fine-tuned variants: multi-tenant engines keep a single copy of the base and batch the requests of many tenants, each with its own adapter. As base models move to 8-bit (FP8) and 4-bit formats, their general matrix multiplications (GEMMs) become cheaper, but the adapters do not: engines run them as separate small GPU kernels beside the GEMM, whose latency takes a growing share of every decoding step. An adapter projects its input down to a low rank and back up, and our key observation is that the GEMM needs the down-projection only in its epilogue, the final stage in which it writes its output. EpiLoRA, a plugin for the vLLM engine, therefore launches the down-projection first and the low-precision GEMM as a dependent kernel that streams its weights meanwhile and adds every tenant’s up-projection in its epilogue. On NVIDIA Hopper GPUs, EpiLoRA decodes for a single adapter with up to 24% lower time per token than vLLM’s fastest configuration, and faster than SGLang and TensorRT-LLM; with many tenants, its adapter kernels deliver more throughput than vLLM’s on most GPU-model pairs we tested. The same low-precision base also trains: a LoRA reinforcement-learning trainer that computes on the engine’s FP8 weights needs no 16-bit copy, so a 27B model trains on one 48 GB GPU, and an engine that then keeps activations in 16 bits removes most of the remaining mismatch between the model generating training data and the one trained.

cs.LG [Back]

[100] Zero-Gated Cross-Attention cs.LG | cs.AI | cs.CLPDF

Basavaraj R M, Manav Verma

TL;DR: 论文提出归纳门控交叉注意力 IGCA，一种零初始化交叉注意力桥接模块，使非文本模态注入冻结语言模型时在初始化阶段保持原模型行为不变，并通过门控逐步引入多模态条件。方法被集成到 Falcon-40B、open_llama_30b 和 Bloom-7b1 的六个解码层中，在 Falcon-40B 上训练 59,500 步后训练损失显著下降，并在留出集上相比关闭门控的基线大幅降低损失。

Details

Motivation: 将非文本模态注入冻结语言模型常导致早期优化不稳定，使架构收益与短暂训练动态难以区分；论文希望在不大幅破坏预训练模型行为的前提下，稳定地实现逐步多模态条件注入。

Result: IGCA 在初始化时是精确的代数 no-op，最大输出差异为 0.00e+00。Falcon-40B 上训练损失从 3.9403 降至 0.0007；留出评估中最佳 gate-on checkpoint 损失为 2.2201，相比 gate-off 基线 4.3469 降低 48.9%（2.1268 nats），平均留出损失从 4.3423 降至 2.2498，平均门控幅度 |tanh(α)|=0.0713。

Insight: 创新点在于用零初始化交叉注意力桥接和归纳门控实现初始化时严格保持预训练模型行为，再渐进式开启多模态条件，从而将架构增益与早期训练动态解耦。客观来看，其核心贡献更偏向稳定的多模态适配与训练策略，而非全新的注意力计算机制，且跨多个冻结语言模型骨干集成增强了方法通用性。

Abstract: Injecting non-text modalities into frozen language models often introduces early optimization instability, making it difficult to separate architectural gains from transient training dynamics. We introduce Inductively Gated Cross-Attention (IGCA), a zero-initialized cross-attention bridge that preserves pretrained model behavior at initialization while enabling gradual multimodal conditioning. Integrated across six decoder layers of frozen Falcon-40B, open_llama_30b, and Bloom-7b1 backbones (with 40,404,961,288 trainable adapter parameters used during training), IGCA acts as an exact algebraic no-op at initialization with a maximum output discrepancy of 0.00e+00. Over 59,500 training steps on Falcon-40B, the training loss decreases from 3.9403 to 0.0007 (with final ten step losses of [0.0001, 0.0003, 0.0005, 0.0002, 0.0005, 0.0011, 0.0005, 0.0006, 0.0005, 0.0007]). On held-out evaluation, the best gate-on checkpoint achieves a loss of 2.2201 compared to 4.3469 for the gate-off baseline – a 48.9% reduction (2.1268 nats) – while average held-out loss improves from 4.3423 to 2.2498 (1.9075 nats) with a mean gate magnitude of $\vert{}\tanh(α)\vert{} = 0.0713$.

[101] When Strong Reasoners Join the Pool: Small Union Premiums and Costly Runaway Generations in LLM Routing cs.LG | cs.CLPDF

Teng-Ruei Chen

TL;DR: 论文在固定协议下用MATH-500（500题×30个种子样本）评测LLM路由中“调用多个模型并保留任一正确答案”相对单模型oracle上限的额外收益。强推理模型加入候选池后，两个推理模型准确率均达96.1%，但多模型并集收益仅0.215–0.588个百分点，远低于此前3.5pp；同时少量超长生成显著拉高成本，甚至使额外模型无法完成评测。作者主张oracle gap应连同模型池、协议、每格样本数和解码预算一起报告。

Details

Motivation: 现有路由基准常用“任一模型答对即算解决”的事后oracle，在随机解码下混淆了单查询选一个模型的预言机上限与多模型并集收益，可能高估路由收益；论文要量化强推理模型加入池后这一收益有多大，以及超长生成带来的成本。

Result: 在MATH-500固定协议、45,000条评分响应上，Nemotron Nano 9B v2与Super 49B v1.5均达96.1%，Llama-3.3-70B为76.2%；两/三模型池的经验并集收益仅0.215–0.588pp，相对总调Nano的选择提升1.27pp，Nano重采样三次与三模型各调一次几乎等效。仅31条响应触及65,536 token上限，但封顶响应最多占某模型输出长度平方代理的60.2%，长生成还使计划中的Llama-3.1-8B过慢未完成。

Insight: 论文宣称的创新点是重新分解并实测随机解码下的路由oracle，发现强推理模型池的并集溢价很小且超长生成成本不可忽视。客观看，其可借鉴之处在于固定协议、预提交生成和可审计ledger的严谨测量，以及提出报告oracle gap时应披露模型池、协议、样本数和解码预算。

Abstract: Routing benchmarks often judge a router against a hindsight oracle that counts a query as solved whenever any model’s recorded answer is correct, but under stochastic decoding that oracle mixes two different objects: the clairvoyant ceiling of committing to one model per query and the premium of calling several models and keeping any correct answer. A companion study measured this premium at 3.5 percentage points (pp) for 11 small open models on MATH-500; how large it is once current reasoning models enter the pool was unknown. We measure it with a fixed-protocol, pre-committed generation on a single NVIDIA B300: all 500 MATH-500 problems, 30 seeded samples per problem, Llama-3.3-70B-Instruct and NVIDIA’s Nemotron Nano 9B v2 and Nemotron Super 49B v1.5 (a NAS derivative of the same 70B) under vendor sampling, a 65,536-token cap, a pinned Math-Verify scorer, and an append-only attempt ledger, for 45,000 scored responses. The two reasoning models both reach 96.1% accuracy against 76.2% for the 70B baseline. The empirical product-coupling premium is 0.215-0.588 pp across all two- and three-model pools; the empirical plug-in selection headroom over always calling Nano is 1.27 pp; and resampling Nano three times gives practically the same empirical any-correct rate as calling all three models once. Only 31 responses reached the cap, 25 of them with exactly periodic tails, yet the capped responses account for up to 60.2% of a model’s output-length-squared proxy, and long generations made a fourth planned model, Llama-3.1-8B, too slow to finish. The premium is far below the companion’s 3.5 pp, but models, pool size and protocol changed together; an oracle gap should therefore be reported with the pool, the protocol, the samples per cell and the decoding budget that produced it.

[102] Say the Same, Act Differently: Diagnosing and Steering Text-Action Divergence in Reasoning VLAs cs.LG | cs.CR | cs.CV | cs.ROPDF

Zihao Feng, Qingzhao Zhang, Chunyu Xia, Bo Yu, Z. Morley Mao

TL;DR: 论文研究推理型视觉-语言-动作（VLA）模型中自然语言理由与物理动作可能解耦的安全问题，揭示视觉扰动可在生成理由逐字不变的情况下改变动作。作者提出文本中性动作子空间（TNAS）诊断方法，在自动驾驶模型Alpamayo和操作模型InstructVLA上定位对动作敏感但几乎不影响文本的方向，并利用该方向引导攻击以验证解耦。结论是不变的理由文本不能证明动作安全，推理VLA的监控必须直接验证动作通道。

Details

Motivation: 现有对VLA推理的攻击通常会篡改理由文本并留下可见证据，论文关注更隐蔽的安全风险：模型说出相同理由却执行不同动作。其出发点是文本监控可能无法保证具身安全，因此需要诊断并度量文本与动作通道之间的分歧。

Result: 在Alpamayo的154个障碍物标注驾驶场景中，TNAS引导攻击在理由逐字不变时于52.6%全图扰动和35.1%局部前摄像头补丁场景下引发新碰撞，而未引导动作攻击仅4.55%。相比保文本基线，TNAS将精确理由保持率从83.0%提升到89.5%，并在相当碰撞率下减少68.4%的teacher-forced文本漂移；在InstructVLA中减少40.5%文本漂移并在85.0%样本保持精确响应，同时产生1.73单位动作块偏移。

Insight: 创新点在于提出TNAS诊断，利用动作敏感方向与文本损失梯度的近正交性找到文本中性动作子空间，并跨驾驶与操作任务定量验证文本和动作通道可被显著解耦。客观来看，其可借鉴之处是把“理由不变”视为不可靠安全证书，并提供一种诊断和引导框架来暴露VLA中隐藏的动作风险，强调监控应直接检查预测动作通道。

Abstract: Reasoning vision-language-action (VLA) models pair physical control with natural-language rationales that supervisors can read to monitor embodied safety. Prior attacks on VLA reasoning corrupt the rationale itself and leave visible evidence in the text. We show that visual perturbations can instead alter the physical action while keeping the generated rationale word-for-word identical. At the shared text/action interface of Alpamayo for driving and InstructVLA for manipulation, our text-neutral action-subspace (TNAS) diagnostic shows that the locally most action-sensitive direction is nearly orthogonal to the active text-loss gradients, with 1.87% and 5.28% norm overlap. At equal hidden-state perturbation norms across both models, interventions along this text-neutral residual shift actions 4.64x to 47.8x more than text-gradient perturbations, whereas text-gradient perturbations induce 21.2x to 277x more text drift. Steering visual perturbations along this direction reaches the same decoupling from the image. Across 154 obstacle-annotated driving scenes in Alpamayo, TNAS-guided attacks induce a new collision under a word-for-word identical rationale in 52.6% of scenes under full-image perturbations and 35.1% under a localized forward-camera patch, compared with 4.55% for unguided action attacks. Relative to a text-preserving baseline, TNAS guidance raises exact rationale preservation from 83.0% to 89.5% and cuts teacher-forced text drift by 68.4% at a comparable collision rate. In InstructVLA, TNAS guidance cuts text drift by 40.5% and preserves the exact response on 85.0% of examples alongside a 1.73-unit action-chunk shift. An unchanged rationale therefore cannot certify a safe action, so monitors for reasoning VLAs must verify the predicted action channel directly.

[103] Sketchpad: certified exact arithmetic with a small vision-language model cs.LG | cs.CL | cs.LOPDF

Mostafa ‘Neo’ Mohsenvand

TL;DR: 论文提出 Sketchpad，让小型视觉语言模型 Qwen3.5-0.8B 配合低秩适配器在渲染算术纸上逐步执行算术过程；每步只读取三格视图并输出数字、进位/借位和下一操作，由固定运行时移动视图且不写入结果数字。作者用 Lean 证明：若每步仅依赖有界视图，则全长度输入上的精确性可归约为有限视图集合上的可检查性质，并在浏览器笔记本 GPU 上通过全部 2,446 个视图验证。

Details

Motivation: 现有语言模型做长算术只在样本上估计准确率，无法保证未见输入；工具调用虽精确但把计算移出网络。论文希望让网络自身执行精确算术，并把任意长度输入上的精确性转化为有限、可验证的性质。

Result: 在部署设备浏览器中全部 2,446 个视图通过验证；模型完成任意长度有符号加减、长乘除、平方根和最大公约数，预注册 200/200 全对，最长 1,000 位加法和 20 位乘法共 220/220 全对，而十个对比模型中最强的 GPT-6.1 Sol 为 217/220。

Insight: 创新点在于用有界视图+固定运行时将无限输入精确性归约为有限视图检查，并用 Lean 形式化证明，同时给出数值扰动容忍和过程组合扩展；运行时只移动视图不写结果数字，适配器乘法门可在不用纸面时精确恢复基座模型。

Abstract: Language models are accurate on most long-arithmetic problems, but their accuracy is estimated on samples and carries no guarantee for an unseen input; tool use is exact but moves the computation outside the network. We show that exactness at all input lengths becomes a finite, checkable property of a network when each of its steps reads only a bounded view of an external working memory. In Sketchpad, Qwen3.5-0.8B with a low-rank adapter executes arithmetic procedures on a rendered sheet. Each forward pass reads a three-cell view and emits one step: a digit, a carry or borrow, and the next operation. A fixed runtime moves the view and writes no digit of a result. We prove in Lean that the model is then exact on all inputs of any length if and only if it is exact on the finite set of views the runtime presents, that no finite test suffices when steps also read earlier steps, and that the guarantee tolerates bounded numerical perturbation and extends to compositions of procedures. All 2,446 views pass on the deployment device, a laptop GPU running the model in a web browser. The certified model performs signed addition and subtraction, long multiplication and division, square roots and greatest common divisors at arbitrary length. It solved 200 of 200 pre-registered problems in the browser, and 220 of 220 sums of up to 1,000 digits and products of up to 20 digits, of which GPT-6.1 Sol, the strongest of ten other models, solved 217. A symbolic runtime built on these procedures solves, expands and factors polynomials with every number computed by the model, and a multiplicative gate on the adapter recovers the base model exactly when the sheet is not in use.

[104] Self-Privileged Fine-Tuning cs.LG | cs.CLPDF

Yangrong Hu, Xueyu Zhou, Yue Wu, Jian Huang

TL;DR: 论文研究监督微调（SFT）中 token 级加权方法的目标分布性质，指出动态加权会诱导过度尖锐的分布，而静态加权更稳定但依赖先验质量。为此提出 Self-Privileged Fine-Tuning（SPFT），用带特权信息（如数学题标准答案）的初始模型冻结副本作为先验，对 token 加权进行指导，并在数学推理和代码生成任务上验证有效性。

Details

Motivation: 现有 SFT 的 token 级加权方法缺乏对优化目标和诱导分布的理论理解；动态加权可能过度尖锐，静态加权虽稳定却受先验分布质量限制。论文希望设计更稳定且有效的 token 级加权微调方法。

Result: 在数学推理和代码生成实验的主要设置上，SPFT 相比标准 SFT 和其他基线取得提升；摘要未给出具体定量数值或明确 SOTA 表述。

Insight: 创新点在于理论刻画动态/静态 token 加权的目标分布，并提出用带特权信息的冻结初始模型作为先验，通过贝叶斯分解按目标 token 对预测真实结果的信息量重新加权。客观看，将特权信息先验引入 SFT token 加权是较有借鉴意义的思路。

Abstract: Supervised fine-tuning (SFT) is a key component of language model post-training, enabling effective task adaptation and serving as an important foundation for subsequent reinforcement learning. Recent methods seek to improve SFT by introducing token-level weights into the training objective, where these weights are typically derived from a model’s output distribution at each token position. To better understand how such weighting schemes affect SFT, we analyze their optimization objectives and theoretically characterize the target distributions they induce. Our analysis shows that dynamic weighting methods, such as Dynamic Fine-Tuning (DFT), induce a heavily sharpened distribution, whereas static weighting methods fit a target distribution proportional to the product of the data distribution and a frozen prior. These findings suggest that static weighting provides a more stable training target, while its effectiveness critically depends on the quality of the prior distribution. Building on this observation, we propose Self-Privileged Fine-Tuning (SPFT), which uses a frozen copy of the initial model conditioned on privileged information (such as ground truth for mathematical questions) as a prior to provide guidance for token-level weighting during fine-tuning. Through a Bayesian decomposition, we further show that SPFT reweights the standard static prior according to how informative each target token is for predicting the ground-truth outcome, thereby assigning larger weights to more predictive tokens. Experiments on mathematical reasoning and code generation show that SPFT improves over standard SFT and other baselines across the main settings.

[105] LoopVQ: Sub 1-bit KV Cache Quantization for Looped Transformers cs.LG | cs.AI | cs.CLPDF

Minsoo Cheong, Sungjoo Yoo

TL;DR: 论文提出 LoopVQ，一种面向循环 Transformer（looped transformers）的 KV Cache 向量量化方法，利用循环步骤间 KV 缓存的高度冗余以及不同循环步骤对量化敏感性的差异，联合编码多个循环步骤的通道，并微调码本以匹配全精度模型输出。该方法在 0.5 bits/element 下优于现有面向标准 LLM 的 KV Cache 向量量化方法，并可与 TaSQ 结合在 0.25 bits/element 下保留较多 BF16 精度；同时在 SGLang 中实现后显著降低解码延迟并提升服务吞吐。

Details

Motivation: 循环 Transformer 通过重复应用相同层减少参数内存，但每个循环步骤仍需存储自己的 keys 和 values，导致 KV Cache 随循环步数线性增长，成为主要内存瓶颈。论文关注如何在循环模型上对 KV Cache 进行极低比特量化。

Result: 在 0.5 bits/element 下，LoopVQ 在 Ouro 和 Huginn 模型上于通用任务、长链式思维推理和长上下文检索任务中优于现有标准 LLM 的 KV Cache 向量量化方法；与 TaSQ 结合后，在 0.25 bits/element 下仍保留较多 BF16 精度。在 SGLang 实现中，32K token 上下文下解码延迟几乎减半，1.4B 和 2.6B Ouro 模型相对 BF16 基线峰值服务吞吐分别最高提升 4.3 倍和 6.1 倍。

Insight: 创新点在于观察到循环模型 KV Cache 沿循环轴高度冗余且不同循环步骤量化敏感性不同，进而提出跨多个循环步骤联合编码通道的向量量化，并微调码本以对齐全精度输出。该方法与现有 KV Cache 量化方法正交，可组合使用，并具备实际推理系统部署优化价值。

Abstract: Looped transformers are a parameter-efficient alternative to standard transformers for language modeling. By reapplying the same layers across multiple loop steps, recent looped language models match the performance of non-looped models several times their size. While this architecture effectively reduces parameter memory, each loop step still stores its own keys and values, so the key-value (KV) cache grows linearly with the number of loop steps and becomes the main memory bottleneck. We observe that the KV cache of looped models is highly redundant along the loop axis and that the loop steps differ substantially in their sensitivity to quantization. Based on these observations, we propose $\textbf{LoopVQ}$, a KV cache vector quantization method for looped transformers that codes channels of multiple loop steps jointly, and fine-tunes the codebooks to match the output of the full-precision model. At 0.5 bits per element, LoopVQ outperforms existing KV cache vector quantization methods for standard LLMs on general, long chain-of-thought reasoning, and long-context retrieval tasks with Ouro and Huginn models. Its design is orthogonal to these methods, and combined with TaSQ, it retains much of the BF16 accuracy even at 0.25 bits per element. We also implement LoopVQ in SGLang, where it nearly halves the decode latency at a 32K-token context and raises the peak serving throughput by up to $4.3\times$ and $6.1\times$ on the 1.4B and 2.6B Ouro models compared to the BF16 baseline.

cs.RO [Back]

[106] RAFT: Reasoning-Aligned Fine-Tuning Improves Robustness of VLA Post-Training cs.RO | cs.CVPDF

Shresth Grover, Andrey Kolobov, Dhruv Shah

TL;DR: 论文提出RAFT（推理对齐微调），一种两阶段VLA后训练方法：先在任务相关推理标注上训练低秩适配器，再冻结适配器并在动作演示上微调策略，以提升有限示范下VLA策略对分布外变化的鲁棒性。

Details

Motivation: 通用VLA策略在少量示范下适配目标任务时，虽能提升常规成功率，但对 appearance、物体和指令变化鲁棒性下降；RAFT旨在解决这一后训练中的鲁棒性退化问题。

Result: 在物理DROID机器人上，RAFT取得75.0%分布外成功率，普通微调为27.9%；在多阶段双臂YAM任务上平均成功率70.0%，最强基线为36.0%；模拟pick-and-place和open-drawer任务在视觉、物体和指令偏移下也有提升。

Insight: 创新点在于将任务相关推理标注用于低秩适配器预训练并冻结，再微调动作策略，使策略更关注任务区域且后期层对象身份线性探针准确率更高；更广泛的推理数据混合还能在不同示范预算下提升下游成功率。

Abstract: Adapting a generalist vision-language-action (VLA) policy to a target task with limited demonstrations can improve nominal task success while reducing robustness to variations in appearance, objects, and instructions. We introduce reasoning-aligned fine-tuning (RAFT), a two-stage post-training procedure: RAFT first trains low-rank adapters on task-relevant reasoning annotations, then freezes them while fine-tuning the policy on action demonstrations. On a physical DROID robot, RAFT achieves $75.0%$ out-of-distribution success, compared with $27.9%$ for plain fine-tuning; on a multi-stage bimanual YAM task, it reaches $70.0%$ mean success, compared with $36.0%$ for the strongest baseline. RAFT also improves success under visual, object, and instruction shifts in simulated pick-and-place and open-drawer tasks. Compared with plain fine-tuning, RAFT policies place more attention on annotated task regions and yield higher linear-probe accuracy for object identity in later layers. A broader reasoning-data mixture further improves downstream success across demonstration budgets.

[107] Hybrid Particle-Voxel Dynamics for Multimodal 3D World Simulation cs.RO | cs.CV | cs.LGPDF

Jimmy Wang, Yifan Chou, Kaifeng Zhang, Yixuan Wang, Binghao Huang

TL;DR: 论文提出 3D World Simulator，一个动作条件多模态世界建模框架，用统一的 3D 表示联合预测外观、几何和触觉。方法结合混合粒子-体素自编码器，将点云压缩为紧凑 3D 体素隐变量，并用动作条件隐动力学模型预测其演化。该框架支持任意视角的几何一致渲染和空间对齐触觉信号集成，可实现 50 Hz 的交互式长时多模态预测，并用于策略训练评估的数据生成及基于模型的规划。

Details

Motivation: 现有动作条件世界模型多在 2D 图像或视频空间预测未来观测，未显式建模底层 3D 场景，导致跨视角空间一致性和长时交互预测困难。论文希望从统一 3D 表示出发，联合建模视觉外观、几何与触觉，以提升多模态世界模拟的空间一致性和长时预测能力。

Result: 实验展示了交互式、长时、多模态预测可达到 50 Hz；模型能够利用触觉和多视角视觉观测进行策略训练与评估的数据生成，并扩展到包含多样物体的大规模数据集。论文还展示了在未见物体和机器人配置下的基于模型规划能力，但摘要未给出具体 benchmark 数值或 SOTA 对比。

Insight: 主要创新在于提出混合粒子-体素表示与动作条件隐动力学结合的多模态 3D 世界模型，从统一 3D 隐空间联合预测外观、几何和触觉，并支持任意视角一致渲染与空间对齐触觉融合。客观看，其可借鉴点是将点云压缩为体素隐变量后进行动作条件动力学建模，从而兼顾 3D 几何一致性、多模态观测和长时交互预测，并服务于策略学习与规划。

Abstract: Predicting how the world evolves in response to robot actions is central to policy learning, evaluation, and planning. However, many action-conditioned world models predict future observations in 2D image or video space without explicitly modeling the underlying 3D scene, making spatial consistency across viewpoints and over long-horizon interactions challenging. We introduce 3D World Simulator, an action-conditioned multimodal world modeling framework that jointly predicts appearance, geometry, and tactile observations from a unified 3D representation. Our approach combines a hybrid particle-voxel autoencoder that compresses point clouds into compact 3D voxel latents with an action-conditioned latent dynamics model that predicts their evolution. The shared 3D representation supports geometrically consistent rendering from arbitrary viewpoints and flexible integration of spatially aligned tactile signals. Our experiments demonstrate interactive, long-horizon, multimodal prediction at 50 Hz. We then show how our model enables scalable data generation for policy training and evaluation using tactile and multi-view visual observations, including wrist-mounted and novel camera views. Furthermore, we scale our method to large datasets containing diverse objects and demonstrate model-based planning with unseen objects and robot configurations. Together, these results highlight the potential of hybrid particle-voxel dynamics as a foundation for multimodal 3D world simulation.

[108] Trusting Imagined Futures: Auditing Imagined Rewards for VLA Post-Training cs.RO | cs.CVPDF

Zicheng Kong, Zhenbo Xu, Haoyan Zhang, Jinghan Yang, Zhenzhen Li

TL;DR: 论文提出Trusting Imagined Futures（TIF），一种面向视觉-语言-动作（VLA）模型后训练的离线监督选择方法。该方法利用视频世界模型想象候选动作结果，并通过执行反馈审计预测奖励，筛选高质量真实轨迹用于行为克隆。TIF在多个LIBERO任务和真实机器人任务上提升了选择轨迹质量与最终成功率。

Details

Motivation: 视频世界模型可用想象结果评估候选动作，但高预测奖励可能偏向执行质量差的轨迹，导致后训练监督不可靠。论文旨在解决如何可信地利用想象奖励为VLA后训练选择监督轨迹的问题。

Result: 在四个LIBERO套件和三个训练种子上，TIF平均成功率达76.25%，较现有轨迹训练提升3.00个百分点；加入审计门控在固定奖励模型和优势边际下再提升1.47个百分点。在三个真实机器人任务的60个候选池上，选择轨迹的干净完成率从58.33%提升至70.00%。

Insight: 创新点在于用执行反馈审计想象奖励，结合多维执行质量预测与不确定性识别奖励高估，并通过优势边际和审计风险阈值决定是否替换监督轨迹。客观来看，其可借鉴之处是将世界模型奖励筛选与真实执行质量审计结合，降低高奖励但低执行质量轨迹对行为克隆的负面影响。

Abstract: Video world models enable candidate actions to be evaluated through imagined outcomes, but high predicted rewards can favor trajectories with poor execution quality. We introduce Trusting Imagined Futures (TIF), an offline supervision selection method for vision-language-action (VLA) post-training that audits predicted rewards using execution feedback. A multidimensional reward model predicts a utility score for candidate ranking and five dimensions of execution quality. An auditor uses reward predictions and uncertainty to identify overestimation patterns learned from paired imagined videos and real executions. TIF accepts the candidate with the highest predicted utility when its advantage over the existing trajectory exceeds a margin and its audit risk passes a threshold; otherwise, it retains the existing supervision. The retained real trajectories then supervise behavior cloning. Across four LIBERO suites and three training seeds, TIF achieves a mean success rate of 76.25%, improving over training on the existing trajectories by 3.00 percentage points with the same trajectory count and training budget. Adding audit gating yields 1.47 percentage points with the reward model and advantage margin held fixed. The gains persist against advantage ranking and logistic regression risk selection at matched replacement counts. On recorded videos from three real robot tasks, TIF increases clean completion among selected trajectories from 58.33% to 70.00% across 60 candidate pools.

[109] Do Agentic Policies Need to Look at Every Step? Geometric Abstraction for Reasoning and Action cs.RO | cs.AI | cs.CVPDF

Yihang Liu, Yiming Chen, Mingzhe Zheng

TL;DR: 论文研究多模态语言模型在机器人交互中是否需要在每一步都重新解析原始视觉观测。作者提出将常规空间推理从图像转移到文本和符号，通过可从局部观测刷新的3D几何抽象作为语言模型直接读取的接口；该接口用类型化、证据感知的符号引用表示物体部件，从标定RGB-D中接地，并在动作后返回带度量残差的紧凑文本检查点。符号任务结构可跨决策复用，度量值则重绑定到当前场景，仅在需要时调用直接视觉推理。

Details

Motivation: 多模态语言模型虽能直接推理机器人场景，但每一步重复解释原始视觉可能不必要且代价高。论文旨在减少重复多模态查询，同时保持多步操作所需的时空推理能力，将例行空间推理从图像转移到可复用的文本与符号表示。

Result: 在离线策略合成和连续在线执行中，基于接地文本与符号的推理提升了多步操作任务完成率，并显著减少重复多模态查询；摘要未给出具体benchmark数值或SOTA对比。

Insight: 创新点在于将显式3D几何状态作为智能体机器人策略的有效基底：几何抽象按局部观测刷新，符号任务结构跨决策复用，度量值每步重绑定，并仅在需要时进行直接视觉推理。客观上，这种设计把“每步重读图像”解耦为可复用符号结构与即时度量接地，有望降低多模态查询开销并提升多步操作稳定性。

Abstract: Recent multimodal language models can reason directly about robot scenes, but it is unclear whether they need to repeatedly reinterpret raw visual observations at every step of an interaction. We propose shifting routine spatial reasoning from images to text and symbols through a 3D geometric abstraction that is refreshed locally from observations and read directly by the language models. Specifically, our interface represents scene geometry via typed, evidence-aware symbolic references to object parts, grounds them from calibrated RGB-D, and returns compact textual checkpoints with metric residuals after each action. Symbolic task structure can thus be reused across decisions while metric values are rebound to the current scene, with direct visual reasoning invoked only on demand. Across offline policy synthesis and continuous online execution, reasoning over grounded text and symbols improves multi-step manipulation completion while substantially reducing repeated multimodal queries, positioning explicit geometric state as an effective substrate for agentic robot policies.

[110] Trajectory Evaluation via Rollout-Free World Model for End-to-end Autonomous Driving cs.RO | cs.AI | cs.CVPDF

Junzhe Wang, Jinqing Zhang, Qingjie Liu

TL;DR: 论文提出RFWorld，一种无需rollout的世界模型，用于端到端自动驾驶中的候选轨迹评估。它构建共享场景记忆，并通过训练时轻量时序读出与语义监督形成未来可读表示；推理时候选轨迹直接以稀疏空间读出查询共享记忆进行评分，从而避免逐候选未来状态递归预测带来的高计算和误差累积。

Details

Motivation: 现有基于模型的轨迹评估器通过显式预测候选条件未来状态来实现未来感知评分，但rollout推理成本高且会累积预测误差；作者认为规划无需为每个候选重建完整未来场景，只需提取未来相关且可用于可靠比较的信息。

Result: 在NAVSIM和Bench2Drive上，RFWorld学习到更规划对齐的世界表示并达到SOTA；推理计算约减少为规划时域乘以候选轨迹数量的量级，同时避免递归rollout误差。

Insight: 创新点在于用共享场景记忆替代逐候选rollout，通过训练时轻量时序读出和语义监督塑造未来可读表示，推理时用稀疏空间读出直接评分；客观看，其将轨迹评估从“重建未来场景”转为“查询共享记忆”，在保持未来感知的同时兼顾效率与误差控制。

Abstract: Although existing candidate trajectory evaluators have substantially improved end-to-end planning accuracy, their underlying mechanism remains suboptimal. Model-based evaluators typically make scoring future-aware by explicitly predicting candidate-conditioned future states over the planning horizon, which we term rollout. However, rollout incurs high inference cost and accumulates prediction errors. We argue that planning does not require reconstructing a full future scene for each candidate, but only future-relevant information for reliable comparison. We therefore propose RFWorld, a Rollout-Free World model for end-to-end trajectory evaluation. RFWorld constructs a shared scene memory and shapes it into a future-readable representation via lightweight training-time temporal readout with semantic supervision, keeping training overhead low. At inference time, candidate trajectories directly query the shared memory via sparse spatial readout for scoring, reducing computation by roughly the planning horizon times the number of candidate trajectories, while avoiding recursive rollout errors. Experiments on NAVSIM and Bench2Drive show that RFWorld learns a more planning-aligned world representation and achieves state-of-the-art performance.

eess.IV [Back]

[111] A Novel Conditional Adversarial Diffusion Model for High-Fidelity OCT-to-OCT Image Translation and Precise Anti-VEGF Therapy Response Prediction in Diabetic Macular Edema eess.IV | cs.CVPDF

Reza Rasti, Armin Biglari, Sobhan Mohammadi Hafshejani, Alireza Dehghani, Michael J. Allingham

TL;DR: 论文提出一种条件对抗扩散框架，从DME患者基线OCT图像合成抗VEGF治疗后的OCT图像，以非侵入方式预测患者对anti-VEGF治疗的初始反应。模型在美国队列（n=127）进行受试者级5折交叉验证，并在独立伊朗队列（n=44）外部验证。与14种基线及先进图像翻译模型相比，该方法在两个队列的治疗反应预测任务上均取得最佳性能，可作为辅助眼科医生治疗前决策的工具。

Details

Motivation: DME是工作年龄人群严重视力丧失的重要原因，anti-VEGF注射虽为一线治疗，但部分患者反应不佳且治疗成本和负担较高。论文旨在治疗前自动、非侵入地评估患者初始反应，避免无效患者陷入试错治疗范式，并支持个性化治疗选择。

Result: 在美国队列上，所提方法在10%和20%中央子视野厚度减少阈值下的平衡准确率分别为85.7%和93.0%；在外部伊朗队列上分别为85.2%和95.3%。在与14种基线及SOTA图像翻译模型比较中，该方法在两个队列的治疗反应预测任务上均达到最佳表现。

Insight: 创新点在于将条件对抗扩散模型用于OCT-to-OCT图像翻译，由基线OCT合成治疗后OCT，并据此推导患者初始anti-VEGF反应。客观来看，其价值在于把治疗反应预测转化为可控的图像翻译与外部验证问题，同时强调非侵入、治疗前决策支持和跨队列泛化评估。

Abstract: Diabetic macular edema (DME) is a leading cause of severe vision loss among working-age adults. Although intravitreal anti-vascular endothelial growth factor (anti-VEGF) injections currently serve as the first-line therapy for DME, a notable subset of patients fails to respond despite the substantial cost and clinical burden, suggesting they may benefit more from alternative interventions. Automated, noninvasive pre-treatment response assessment could therefore spare non-responders from the current trial-and-error treatment paradigm. In this study, we propose a novel conditional adversarial diffusion framework that synthesizes post-treatment optical coherence tomography (OCT) images from baseline scans, enabling the derivation of each patient’s initial response to anti-VEGF therapy. The model was developed with subject-level 5-fold cross-validation on a US cohort (n = 127) and externally validated on an independent Iranian cohort (n = 44). Benchmarked against fourteen baseline and state-of-the-art translation models, the proposed model achieved the best treatment-response prediction performance among the methods evaluated for that task on both cohorts, with balanced accuracies of 85.7% and 93.0% on the US cohort and 85.2% and 95.3% on the external Iranian cohort at the 10% and 20% central subfield thickness reduction thresholds, respectively. Ultimately, this framework offers a decision-support tool to help ophthalmologists anticipate initial anti-VEGF outcomes in DME prior to treatment initiation, toward personalized therapy selection. The source code will be made publicly available upon acceptance.

[112] DUNEDepth: Bronchoscopic Depth and Pose Estimation via Hybrid Adaptation of a 3D-Vision Foundation Model eess.IV | cs.CVPDF

Qi Chang, Austin Kao, Ange Lou, William E. Higgins

TL;DR: 论文提出 DUNEDepth，用于支气管镜视频中的深度估计与相机位姿估计，以辅助将支气管镜术中视频与术前 3D 胸部 CT 对齐。方法基于 3D 视觉基础模型进行混合适配，结合共享 DUNE ViT 编码器、双 MASt3R 交叉注意力解码器以及融合 MASt3R 和 Reloc3r 元素的预测头，并采用合成监督与自监督相结合的混合训练策略，利用支气管镜视频和 CT 生成的虚拟支气管镜视图进行微调。

Details

Motivation: 支气管镜导航和计算机辅助诊断对早期肺癌检测很重要，但其有效使用依赖于准确建立术中支气管镜视频与术前 CT 之间的对应关系；从检查视频中估计支气管镜深度和相机运动仍是未解决问题。

Result: 在支气管镜视频序列上的实验表明，DUNEDepth 能比先前自监督方法更准确地估计深度信息和相机运动轨迹。

Insight: 创新点在于将 3D 视觉基础模型混合适配到支气管镜场景：设计共享 ViT 编码器、双交叉注意力解码器和融合式预测头，并提出联合合成监督与自监督的训练策略，利用 CT 虚拟支气管镜视图为真实支气管镜视频学习提供额外监督。

Abstract: State-of-the-art bronchoscopy guidance systems and computer-aided diagnosis methods have emerged as important tools for early lung cancer detection. Effective use of these tools depends on being able to accurately link a patient’s bronchoscopic airway exam video to their pre-operative 3D chest CT scan. Unfortunately, the derivation of such links, which include bronchoscope depth and camera motion, from the exam video remains an open problem. Building on recently proposed 3D-vision foundation models, we propose DUNEDepth, a hybrid architecture that combines a shared DUNE ViT encoder, a twin MASt3R cross-attention decoder structure, and a prediction head that includes elements of the MASt3R and Reloc3r models. To fine-tune the architecture to our application, we utilize an integrated hybrid synthetic-supervised/self-supervised training strategy that draws on both bronchoscopic video and CT-derived virtual bronchoscopic views. Experiments drawing on bronchoscopic video sequences demonstrate that DUNEDepth more accurately derives depth information and camera motion trajectories than previously proposed self-supervised methods.

Table of Contents

cs.CL [Back]

[1] Why Pay for Long Reasoning When Rewrite Do Trick? Causal Effects of Reasoning History Compaction on Future Generation and Agent Cost cs.CL | cs.AIPDF

[2] Balancing Chain-of-Thought and Answer Supervision for Small Language Models on Medical Question Answering cs.CL | cs.AIPDF

[3] Dory1.0-4B: A Looped Normalized-GPT cs.CLPDF

[4] Language Model Augmented Decision Transformer for Multi-Task Offline Reinforcement Learning cs.CL | cs.LGPDF

[5] Evaluation Contracts and Evaluation Leverage: Sliced-Gap Readings in Structured Prediction Can Move Without Model Change cs.CLPDF

[6] Where Do Distilled Students Store Thinking? cs.CL | cs.LGPDF

[7] Constructing a Regional Slang Lexicon for Electricity Demands via Soft-Label Bidirectional Anchoring and Active Learning cs.CL | cs.AIPDF

[8] Do Multi-Turn Conversational Capabilities Carry Over from Text to Speech? A Parallel Multilingual Benchmark cs.CLPDF

[9] Typological Feature Induction as a Benchmark for Inductive Metalinguistic Reasoning cs.CLPDF

[10] When Ground Truth Is Not Ground Truth: Automatic Labeling for Hallucination Detection cs.CL | cs.AIPDF

[11] ReSI-Math: Exploring Recoverable Capability Frontiers for Efficient Reasoning Improvement cs.CL | cs.AIPDF

[12] One Assistant, Many Memories: Benchmarking Real-World Group Assistance When Recall Is Not Enough cs.CL | cs.AIPDF

[13] TraceRouter++: Causal Semantic Routing Realignment via Optimal Transport cs.CLPDF

[14] Shuffled OPSD: When Semantic Consistency Hurts On-Policy Self-Distillation? cs.CL | cs.AIPDF

[15] Who is the Pronoun Pro? Comparing LLMs and Humans at Number-Disambiguation of ‘they’ cs.CL | cs.AI | cs.HCPDF

[16] When Self-Verification Hurts: Calibrated Answer Repair for Financial Question Answering cs.CLPDF

[17] Beyond Accuracy: What Language-Subspace Interventions Capture, Remove, and Improve cs.CL | cs.AIPDF

[18] MDSKILL: Learning to Internalize Skill Memory through Multi-View Distillation for Efficient Dynamic Agent Self-Evolution cs.CLPDF

[19] Suppressing Misaligned Behaviors in Language Models, Durably cs.CL | cs.AI | cs.LGPDF

cs.CV [Back]

[20] Physics-Informed Functional Tucker Method with RKHS Factors for Sparse Spatiotemporal Reconstruction cs.CV | cs.AIPDF

[21] Beyond Proposals: Region-Consistent Open-Set Object Detection cs.CVPDF

[22] FlowBench: Counterfactual Imagination Training for Spatial Intelligence cs.CVPDF

[23] SurgEye: Instruction Tuning-Free Perception-Reasoning Framework for Surgical Scene Understanding cs.CVPDF

[24] Compression Forcing: Hierarchical History Tokenization Enables Flexible World Models cs.CVPDF

[25] Do Pretrained Backbones Transfer to Physical Sensors? A Systematic Study of Multimodal Initialization Across the Sensor Stack cs.CVPDF

[26] From Visual Matching to Textual Conditioning: Cross-Domain Few-Shot Segmentation via Reference Inversion cs.CVPDF

[27] ScienceVision: Can AI Observe and Reason Like a Scientist? cs.CV | cs.AIPDF

[28] Generalizable AI-Generated Image Detection with Embedding-Space Explanations cs.CV | cs.AI | cs.MMPDF

[29] Co-Occurrence Priors Override Visual Spatial Evidence: Understanding False Alarms in Vision-Language Models for Construction Struck-By Hazard Detection cs.CV | cs.LGPDF

[30] What Remains in Sight? Autoregressive Video Decoding as Representation-Guided Context Rewriting cs.CV | cs.AI | cs.LGPDF

[31] ReconDreamer#: Self-Evolving Closed-Loop Synergy of Reconstruction and Generation for Dynamic Driving Scenes cs.CVPDF

[32] Reliability-Role Collaborative Learning for Semi-Supervised Cardiac Image Segmentation cs.CVPDF

[33] CounterSketch: Physics-Grounded Counterfactual Reasoning for Sketch-to-Video Generation cs.CVPDF

[34] ProGeo-RC: Prototype-guided Semantic Reliability and Geometric Error Calibration for Semi-supervised Cardiac Image Segmentation cs.CVPDF

[35] Sketch-Omni: Instance-Level Sketch-Based Visual Prompting for Perception MLLMs cs.CVPDF

[36] Weight-Space Programming for Improving Image-Restoration Diffusion Models: Technical Conjectures and a Recursive Improvement Loop cs.CVPDF

[37] GroundSeek-CXR: Retrospective Evaluation-Guided Active Observation Framework for Multi-Label Chest Radiography Diagnosis cs.CVPDF

[38] Mitigating Spatial Blindness in Multiple Instance Learning for Whole-Slide Image Analysis cs.CVPDF

[39] ReSMap: Recasting Satellite Priors for Robust and Accurate Online HD Map Construction cs.CV | cs.ROPDF

[40] IGSW: Multi-Weather Urban Scene Reconstruction with Intrinsic-Decoupled Gaussian Splatting cs.CVPDF

[41] SphereFlow: Missing Modality Imputation via Geometric Transport on Hypersphere cs.CVPDF

[42] VidUEU-Agent: A Data-Curation Agent for Multimodal Understanding, Editing, and Unified Tasks cs.CVPDF

[43] GVT3D: Gesture-Vision-Text Token Enrichment for 3D Embodied Reference Understanding cs.CVPDF

[44] Entropy-Gated Dynamic Rectification for Object Hallucination in Large Vision-Language Models cs.CVPDF

[45] DynOSeg: Dynamic Object Segmentation via Motion-Geometry Fusion cs.CVPDF

[46] Is Forgetting Unified? Exploring Concept Unlearning in Unified Multimodal Models cs.CVPDF

[47] Beyond Token Representations: Explicit Visual Object Grounding for Video Reasoning Segmentation cs.CVPDF

[48] KeyWise: Efficient Video-Only Piano Transcription with a Lightweight Per-Key Architecture cs.CV | cs.LGPDF

[49] Bridging CLIP with DINO: Cross-Modal Information Maximization for Online Test-Time Adaptation cs.CV | cs.LGPDF

[50] 3DGS-VBench+ and 3DGSV-Assessor: A Multi-Dimensional Subjective Benchmark and a Reference-Flexible Objective Model for 3D Gaussian Splatting Video Quality Assessment cs.CVPDF

[51] E2I-Flow: EEG-to-Image Reconstruction via Wavelet-Temporal Cross-Attention and Conditional Flow Matching cs.CV | cs.LGPDF

[52] BtsPruner: Role-Aware Visual Token Pruning for Large Vision-Language Models cs.CV | cs.AIPDF

[53] SEEK-VAU: Towards Evidence-Faithful Video Anomaly Understanding via Agentic Search cs.CVPDF

[54] Gradient-Guided Orientation Learning for Weakly Supervised Oriented Object Detection cs.CVPDF

[55] ReXMo: End-to-End Video Character Replacement with Expression and Motion Preservation cs.CVPDF

[56] Therapy-FIS: A Dataset for Assessing Facilitative Interpersonal Skills in Psychotherapeutic Setting cs.CV | cs.AI | cs.CYPDF

[57] Phase-DGS: Phase-guided Dynamic Gaussian Splatting from Unsynchronized Multi-view Video cs.CV | cs.GRPDF

[58] Learning from Unlabeled Photographs for General Oral Image Analysis cs.CVPDF

[59] DenseTwins: Dense Modality Decoupling for Multimodal Self-Supervision cs.CV | cs.AI | cs.LGPDF

[60] Cycle WAM: Coupling Action and Video through Cycle Consistency cs.CV | cs.AI | cs.ROPDF

[61] SPOD: Recovering Suppressed Temporal Evidence in Video-LLMs via Single-Pass Orthogonal Decomposition cs.CVPDF

[62] iMED: A Multi-Endoscope Dataset for Surgical 3D Perception cs.CV | cs.ROPDF

[63] Beyond Local Refinement: Wide-Area Precise Cross-View Localization cs.CVPDF

[64] Pixel-space video generation with U-JiT cs.CVPDF

[65] NightIQA: A Comprehensive Dataset and Benchmark for MLLM-Based Nighttime Image Quality Assessment cs.CVPDF

[66] OCR in the Foundation Model Era: From Specialist Models to Unified Models cs.CVPDF

[67] VisSearch-Bench: Evaluating Long Horizon Visual Search in Multimodal Agents cs.CV | cs.AIPDF

[68] Heads That Write, Not Just Point: Image Retrieval Heads in Vision-Language Models cs.CV | cs.CLPDF

[69] EP-LVMQA: Evidence-Preserving Coarse-to-Fine Localization for Low-Quality Long Video Moment Question Answering cs.CVPDF

[70] Behind Video Generation: Observations and Insights on Multimodal Understanding from Kling cs.CVPDF

[71] VISTA: Vision-Language Integrated Semantic Triage Architecture for Online Face Clustering in Long Videos cs.CV | cs.MMPDF

[72] Learning Where to Look: Observation Policy Optimization for Thinking with Images cs.CVPDF

[73] PATTS: Parallel-Aware Test-Time Scaling for Thought-Image Generation in Multimodal Discrete Diffusion LLMs cs.CVPDF

[74] An Agentic Framework for Quality-Aware and Scenario-Adaptive Image Compression cs.CVPDF

[75] ProAlign: Progressive Positional and Prototype-Guided Alignment for Aerial-Ground Person Re-Identification cs.CVPDF

[76] Can saliency maps flag chest radiograph classifier errors at a new hospital? An external validation study cs.CV | cs.LG | eess.IVPDF

[77] FloMo: 3D Scene Flow as a World Action Model Intermediate for Learning from Human Video cs.CV | cs.AIPDF