Caption
-
Mitigating Video Hallucinations via Attention-Guided Representation Perturbation
2026-10-14 07:50:08Zhixin Cai, Ruoqi Yuan, Xinchen Yang, Zhuofan Chen, Jun Bai, Chuantao Yin, Wenge RongAbstract
Video large language models (Video LLMs) often generate responses unsupported by the input video. Contrastive decoding (CD) can mitigate these hallucinations, but its effectiveness depends on how the negative branch weakens visual evidence. Existing methods rely on predefined perturbations or masking guided by question reconstruction, offering limited control over the model's visual reliance during answer generation. We introduce Attention-guided Representation Perturbation for Contrastive Decoding (ARP-CD), a training-free framework for constructing contrastive negatives directly in representation space. ARP-CD uses the attention mass assigned to visual tokens at the answer-prediction position to guide bounded gradient updates of the language model's input visual hidden states. These updates reduce visual attention and produce a negative input, whose predictions are contrasted with those of the original branch throughout generation. Experiments across four Video LLMs and four benchmarks show that ARP-CD outperforms representative training-free baselines in most model--benchmark settings, reducing hallucinations while improving general video understanding. Further analyses show that attention-guided perturbation directions improve negative quality and that the resulting negatives provide effective contrastive references for correcting the original predictions.
Abstract (translated)
URL
https://arxiv.org/abs/2610.19728
PDF
-
TVPara: Paralinguistic Captioning of Expressive TV-Series Speech in Complex Acoustic Environments
2026-10-14 06:59:02Ziyang Zhang, Wenyi Yu, Wen Wu, Xuenan Xu, Chao ZhangAbstract
Large audio-language models (LALMs) struggle to parse paralinguistics (e.g., emotion, pace, pitch) in speech that combines intense emotion with complex acoustic environments. Strengthening this ability calls for training data that reflects such conditions. However, existing corpora rely heavily on studio-recorded speech, offering limited emotional and acoustic diversity. To address this gap, we introduce TVPara, a paralinguistic captioning dataset built from TV series. Speech in TV series often carries diverse emotional expression and co-occurs with scene-specific background sound, providing expressive emotion and complex acoustic environments at scale. Constructed via LALM pre-annotation and human correction, TVPara includes two tasks: Single-Utterance (~492h) and Multi-Utterance (~213h) Paralinguistic Captioning. Fine-tuning open-source LALMs on TVPara yields an average relative improvement of 31.26% in paralinguistic understanding under complex acoustic environments. The gains further generalize to an unseen TV series and to cleaner, cross-domain speech.
Abstract (translated)
URL
https://arxiv.org/abs/2610.19678
PDF
-
HyrCap: Hybrid Rank-Calibration of Action Proposals for Temporal Event Understanding
2026-10-14 03:44:40Rui Chen, Xingyu Chen, Pengxin Xu, Kai Chen, Junzhi YuAbstract
Temporal action detection in long untrimmed videos suffers from inaccurate localization and low-quality proposals, especially with sparse annotations and ambiguous boundaries. For open-ended video understanding and LLM-driven event analysis, reliable proposals connect temporal localization, event annotation, and semantic understanding. However, conventional temporal action detection (TAD) pipelines lack explicit modeling of proposal localization quality, cross-representation support, and inter-proposal relations. We propose Hybrid Rank-Calibration of Action Proposals (HyrCap), a proposal-centric query-based TAD framework for temporal event understanding. Hybrid combines complementary video features with different localization cues, while Rank-Calibration adjusts proposal ranking confidence to reflect localization reliability and candidate competition. HyrCap treats dense query-based predictions as an over-complete temporal event hypothesis space. Cross-Representation Consensus Calibration, Localization Quality Estimation, and Proposal Redundancy Suppression improve proposal ranking reliability, localization quality, and candidate discrimination. Inspired by LLMs' capabilities in fine-grained semantic recognition, event description, and reasoning, we introduce HyrCap for Semantic Action Understanding, using calibrated proposals for standard TAD and as inputs to LLM-based semantic understanding. We define a staged evaluation protocol to separately measure event localization, segment-level semantic understanding, and combined event understanding performance. Experiments on public benchmarks show that HyrCap achieves state-of-the-art performance across multiple metrics and provides a finer-grained paradigm and evaluation protocol for query-based event segmentation and annotation.
Abstract (translated)
URL
https://arxiv.org/abs/2610.19526
PDF
-
VideoHarness-SRSI: Structured Recursive Self-Improvement for Video Understanding Harness
2026-10-14 02:17:07Zihao Lin, Haoyu Dong, Haobo Xu, Chi-Chih Chang, Xuying Ning, Yinglong Xia, Shuang Yang, Zhuokai Zhao, Xinying Song, Dongfang Liu, Lifu Huang, Yang Song, Qifan WangAbstract
LLM system capabilities depend not only on model parameters but also on the executable harnesses that organize information acquisition, tool use, reasoning, and memory. Recursive harness evolution can automate their design, but how it can reliably discover generalizable systems from limited, noisy, and repeatedly reused feedback remains unclear. We study this question in agentic video understanding, where rich multimodal evidence and coupled mechanisms make harness optimization challenging. We first introduce VideoHarness-BRSI, a recursive self-improvement framework that evolves video harnesses around a frozen vision-language model. Analyzing over 4,000 evolved candidates, we find it competitive with hand-engineered systems but identify three interconnected failure modes: validation overfitting, search-budget dilution, and the early-experience trap. We trace all three to a common root: the same imperfect validation evidence drives both harness revisions and the proposer's search strategy. We therefore propose VideoHarness-SRSI, an evidence-governed framework in which every search decision rests on evidence that is transferable, sufficient, and gathered before the search commits. It restricts the proposer to content-free feedback, informs exploration with instrumented warm-up experiments, and develops mechanisms in isolated probe-race branches before factorial composition and joint refinement. Across four video understanding benchmarks, VideoHarness-SRSI improves the equally weighted mean held-out score by 6.2% relative to VideoHarness-BRSI and reduces the mean validation-test gap by 37.9%. Reliable harness evolution thus depends on how search is organized and informed, not only on how well it optimizes validation performance.
Abstract (translated)
URL
https://arxiv.org/abs/2610.19424
PDF
-
Kandinsky S2V 1.0: Audio-Driven Avatar Video Generation
2026-10-13 23:11:39Nikolay Semenov, Artem Riabinin, Sergey Bassenko, Anna Averchenkova, Julia Agafonova, Anastasia Aliaskina, Kirill Kozlov, Lev Novitskiy, Nikita Samsonov, Nikolai Gerasimenko, Denis DimitrovAbstract
We present Kandinsky S2V 1.0, an open family of speech-to-video models for controllable audio-driven human animation, comprising Kandinsky S2V 1.0 Lite (2B parameters) and Kandinsky S2V 1.0 Pro (21B parameters). From a single reference image, a speech waveform, and a text prompt, the models generate long-form 25 FPS video in which lip motion, expression, pose, and gesture follow the speech. The architecture draws text and audio conditioning from one Qwen2.5-Omni model and adds a reference-aware three-scale visual memory. Training starts from the Kandinsky 5.0 Video foundation models [1] and combines mixed reference-memory layouts with an area-adaptive face loss. The training corpus contains 1.17M single-speaker clips selected by active-speaker detection, with captions generated directly in English and Russian. A side-by-side human evaluation against the strongest open-source baseline and two proprietary commercial models shows that Kandinsky S2V 1.0 Pro performs comparably to or better than these models on average and is also preferred for lip synchronization. On common speech to video benchmarks, Kandinsky S2V 1.0 Lite and Kandinsky S2V 1.0 Pro rank among the top performing models according to mean rank over five reported metrics. Beyond the base models, we provide {\pi}-Flow distilled variants of both Kandinsky S2V 1.0 Lite and Kandinsky S2V 1.0 Pro. The distilled Kandinsky S2V 1.0 Pro model generates video in 16 sampling steps without classifier-free guidance, achieving a 5.4 times end-to-end speedup over the base model, while matching or exceeding it on nearly all reported benchmark metrics. To accelerate open research and deployment in speech-to-video generation, we release the code, model checkpoints, and diffusers integration under the MIT license.
Abstract (translated)
URL
https://arxiv.org/abs/2610.19290
PDF
-
The Gap Is in the Geometry: A Convergent Meta-Analysis of Multi-Facet Aerial and Remote Sensing Video Understanding
2026-10-13 19:39:34Georgios Angelidis, Aggelos Psiris, Maria Vakalopoulou, Georgios Th. Papadopoulos, Giorgos Sfikas, Iraklis VarlamisAbstract
Aerial and remote sensing video is increasingly evaluated like web video, by posing multiple-choice questions to multimodal large language models (MLLMs) over sampled frames and reporting accuracy. This paper argues that this recipe measures the wrong aspect, and supports the claim with a convergent meta-analysis of a broad set of published results, organized in four evidence tiers, in which a claim becomes a position only when collected evidence agrees. Parameter scale exhibits a median of +3.4 accuracy points per tenfold increase on aerial benchmarks, far below a human gap of at least 19 points, whereas acquisition conditions control perception, as 315 of 318 evaluations in 19 within-paper contrasts lose accuracy when conditions such as altitude, view, object size or time horizon become harder. No aerial benchmark in the corpus combines video input with per-task human and random baselines, competence collapses from answering to acting in closed-loop navigation, and the same MLLMs score higher on the aerial benchmark than on ground spatial ones, but lower on studies that test cross-view geometry. From these findings the current study derives five falsifiable principles and proposes the AVU-Card, a reporting card for aerial video benchmarks
Abstract (translated)
URL
https://arxiv.org/abs/2610.19102
PDF
-
Rethinking Projector Training in Multimodal LLMs
2026-10-13 15:53:53Fabian Gr\"oger, Na\"el Ouerghemi, Shuo Wen, Maria Brbi\'cAbstract
Multimodal LLMs are typically trained by first aligning a vision encoder with a language model via projector training on large image-caption datasets, followed by joint training of the full multimodal model. Despite its widespread use, the role of the projector training stage remains poorly understood. We find that (i) using only $10$ to $20\%$ of the data for training the projector already recovers most downstream performance gains, and (ii) projectors trained on different datasets can be linearly interpolated while largely preserving performance. These results suggest that projector training mainly performs a coarse alignment, which many different projectors achieve equally well. Motivated by these findings, we propose PORTAL, a training-free projector initialization that requires neither paired image-caption data nor gradient-based optimization, but only per-modality summary statistics. PORTAL computes an optimal-transport-based map between Gaussian approximations of the vision and language feature distributions on a shared principal subspace, so that the projector training stage can be skipped entirely. Without any paired data, PORTAL matches the standard pipeline trained on 558k image-caption pairs across six LLM backbones, two vision encoders, and 16 benchmarks. It outperforms that pipeline in $3$ of $7$ (vision encoder, LLM) settings and reaches an average $98\%$ of its score across the other four.
Abstract (translated)
URL
https://arxiv.org/abs/2610.18598
PDF
-
PhysicsSliders: Learning Physical Controls for Audio Foley Generation
2026-10-13 15:53:45Peiwei Chang, Yongqiang Chen, Yuelan Cheng, Liting Gao, Yishun Lu, Li Zhang, Mark D. Plumbley, Philip J. B. Jackson, Wenwu WangAbstract
Physical properties provide intuitive controls for sounds made by interacting objects. The challenge is to make a generative model respond predictably to these controls, both individually and in combination. We present \method\ (PXS), trained on simulated contact sounds with 15 material, geometry and contact parameters. Two separately trained variants accept direct parameter inputs or numerical captions. PXS controls hardness, size, speed and damping, supports room descriptions and produces spectrally distinct outputs at fixed targets. When controlling damping and size together, specifying thickness improves frequency accuracy and joint target delivery. Measurements from real recordings define effective-damping targets, which simulator inversion maps to generation inputs without real-audio fine-tuning. Across 18 impact classes from five real-audio datasets, PXS-Physical follows these targets with mean grouped Spearman correlation $\rho=0.976$ and class-median error 0.094 dex (a factor of 1.24).
Abstract (translated)
URL
https://arxiv.org/abs/2610.18597
PDF
-
EgoOmniWorld: Unifying Egocentric Understanding via Ego-Action-World Interactions
2026-10-13 14:53:16Songruo Chen, Yuqian Fu, Yu Li, Mingkang Dong, Sifan Zhou, Bin Ren, Xu Zheng, Tianwen Qian, Mohamed Elhoseiny, Yanwei FuAbstract
Egocentric video understanding spans diverse capabilities, modalities, and evaluation formats, yet these are often studied in isolation despite their tight coupling in real-world interaction. We propose the Ego-Action-World interaction loop as a unifying principle: the wearer perceives the world, forms goals, acts, changes the world, and uses feedback to guide subsequent behavior. Based on this view, we introduce EgoOmniWorld, a benchmark organized around four complementary interaction structures: Node captures ego, action, and world states; Edge captures directed relations among them; Chain captures linked dependencies; and Loop captures feedback-closed interactions. We develop an interaction-first, evidence-grounded data pipeline that extracts structured interactions from egocentric videos, grounds them in visual and auditory evidence, and generates questions according to their interaction structure. Required modalities and single- or multi-turn formats follow naturally from the supporting evidence and dependencies. EgoOmniWorld contains 24,166 QA groups, including 20,284 training and 3,882 test samples. Experiments on representative open- and closed-source multimodal large language models reveal substantial and uneven limitations across interaction structures. Models trained on EgoOmniWorld improve in-domain interaction understanding and generalize to existing egocentric benchmarks. EgoOmniWorld provides a unified foundation for evaluating and advancing egocentric video understanding. Data, code, and models will be released.
Abstract (translated)
URL
https://arxiv.org/abs/2610.18499
PDF
-
Stabilizing the Modality Gap: Null-Space Projected Continual Learning for Vision-Language Models
2026-10-13 11:41:24Tiantian Peng, Yuyang Liu, Shuo Yang, Qiuhe Hong, YongHong TianAbstract
The modality gap is intrinsic to multimodal alignment, where embeddings from different modalities remain partially separated in a shared representation space. In continual learning, CLIP-based vision-language models (VLMs) not only suffer catastrophic forgetting but also experience modality gap drift that disrupts the shared embedding geometry. Such drift degrades cross-modal alignment and generalization. We propose FPCD, an efficient framework combining Fast Null-space Projection (FNP) and Contrastive Distillation (CD). FNP accumulates compact Gram matrices of previous tasks' activations and projects gradients onto layer-adaptive approximate null spaces to reduce interference with learned representations. By replacing singular value decomposition with symmetric eigendecomposition, FNP accelerates projection-basis construction by a factor of 5-9. CD uses a small reference set to align the bidirectional image-to-text and text-to-image similarity distributions of the fine-tuned model with CLIP, thereby preserving pretrained cross-modal relations. On the 11-task MTIL benchmark, FPCD achieves 87.6% Last, 76.4% Average, and 65.2% Transfer. It also maintains more stable modality gaps and stronger zero-shot retrieval performance on COCO Captions and Flickr30k than competing methods. These results highlight the importance of controlling modality gap drift during continual adaptation of VLMs.
Abstract (translated)
URL
https://arxiv.org/abs/2507.19839
PDF
-
MERIT: Multimodal Evidence Reasoning over Image Tables
2026-10-13 08:21:53Haoyang Jiang, Zhuotong Fan, Mingyang Yi, Ju Fan, Lanqing Hu, Xiao ZanAbstract
Table reasoning with language models typically assumes access to clean textual tables. In practice, tables are often serialized into text, which can corrupt cell contents, distort structural relationships, and discard visual cues, leaving the model to reason over incomplete evidence. In contrast, table images retain this information, but using it requires both accurate perception and reasoning grounded in the relevant cells. To address this, we propose MERIT (Multimodal Evidence Reasoning over Image Tables), a three-stage framework for learning to reason over table images, with serialized tables serving as optional auxiliary inputs (if provided). Our analysis identifies visual evidence acquisition as a bottleneck, motivating table-specific encoder adaptation through reconstruction, localization, and color-conditioned supervision. Building on the adapted representations, supervised fine-tuning teaches explicit evidence selection and reasoning using verified trajectories. Finally, reinforcement learning refines the policy with separate rewards for evidence correctness, calculation, final answers, and output format. By providing feedback on intermediate steps, the framework targets retrieval and computational errors that final-answer supervision alone does not distinguish. We evaluate the approach on table question answering and fact verification, analyze the contributions of visual inputs and training components, and assess generalization to additional datasets with different table structures and visual presentations.
Abstract (translated)
URL
https://arxiv.org/abs/2610.17937
PDF
-
From Token Correction to Claim Admission: Training-Free Hallucination Mitigation for Large Vision-Language Models
2026-10-13 04:35:53Hyeonggeun Han, Taehyun Cho, Sangwoo HongAbstract
Large vision-language models often exhibit object hallucination, generating claims that are not supported by the input image. To mitigate this issue, we propose Evidence-Guided Visual Claim Admission (EGVCA), a training-free framework that admits a semantic claim only when it receives sufficient image-dependent support. EGVCA measures this support by comparing the claim likelihood under the observed image and a visual-null input. For captioning, it controls semantic expansion without altering token generation. For binary visual question answering, it combines atomic visual support with the image-induced change in the model's native Yes--No margin. Experiments on CHAIR, POPE, and MME across three LVLM backbones show that EGVCA substantially reduces object hallucination while preserving caption quality. It achieves the highest POPE F1 in eight of nine backbone--sampling settings and improves the native MME total score on all three backbones. These results demonstrate the effectiveness of claim-level admission as a simple alternative to token-level distribution correction.
Abstract (translated)
URL
https://arxiv.org/abs/2610.17706
PDF
-
ATTEST: Evidence-Grounded Reinforcement Learning for Long-Audio Temporal Understanding
2026-10-12 17:29:06Ashutosh Chaubey, Jinxi Guo, Minh Tran, Michael Andersland, Ozlem KalinliAbstract
Temporal grounding is fundamental to audio understanding, yet large audio language models (LALMs) remain weak at it on long audio, where hallucinations compound and annotated data is scarce. Existing benchmarks cover only short clips or offer coarse, synthetic annotations. We introduce EgoSCAPE, a manually verified benchmark created from Ego4D with 700 real audios of durations up to 5 minutes, annotated with dense captions, sound sources, audio timelines, and temporal grounding queries. We further propose ATTEST, a GRPO-based reinforcement learning framework that improves temporal grounding without changing the base model's architecture or token vocabulary. ATTEST combines (i) a counterfactual grounding reward penalizing predictions that survive removal of their acoustic evidence, (ii) an attention-density reward encouraging generation to attend to its own claimed time spans, and (iii) token-wise advantages giving segment-level credit for long dense captions. ATTEST outperforms prior temporal grounding methods on existing dense captioning and grounding benchmarks, with fewer hallucinations. Together, EgoSCAPE and ATTEST establish a rigorous evaluation-training recipe that brings temporally grounded audio understanding closer to practical, long-form deployment.
Abstract (translated)
URL
https://arxiv.org/abs/2610.17149
PDF
-
The Value of Being Wrong: Self-Mined Visual In-Context Learning from Errors
2026-10-12 15:38:27Shu Zhao, Tan YuAbstract
Visual in-context learning (ICL) enables vision-language models (VLMs) to adapt at test time via demonstration examples, but its effectiveness depends critically on the examples selected. Recent counterfactual selection methods improve ICL by constructing demonstrations that expose how individual attribute changes affect the answer. However, they require multi-stage pipelines involving attribute extraction, caption engineering, and composed image matching. We propose a simpler alternative grounded in a key observation: a model's own prediction errors are natural counterfactual demonstrations that directly encode its failure modes. Our method, SMILE (Self-Mined Visual In-Context Learning from Errors), builds a model-specific pool of hard negatives offline by recording the target VLM's own mistakes. At test time, the prediction serves as a direction signal that surfaces, in the semantic embedding space, the past mistakes most likely to recur for the current query. This prediction-conditioned identification provides a unified mechanism for both closed-set classification and open-ended visual question answering without task-specific design. Across four benchmarks and four VLMs, SMILE delivers an average task-score gain of +3.09 points over the SOTA baseline at 5x lower latency. Project page: this https URL
Abstract (translated)
URL
https://arxiv.org/abs/2610.16977
PDF
-
Making Language the Low-Rate Code of Vision: Successive Refinement from Words to Pixels
Abstract
An ordered visual code lets a short prefix stand in for an image, and the training loss decides what that prefix contains. Under pixel reconstruction the first bits go to the directions of highest variance, and we prove that a caption can then be independent of the entire first half of a code that is rate-optimal for pixels at every length. In the rate-distortion sense a caption is therefore not the low-rate code of its image, unless the code is built to make it one. We study that construction and what it costs. We cast captioning and text-to-image generation as successive refinement under a distortion measure that changes along the code, from caption log-loss in the early stages to pixel distortion in the last, so that each stage is compared with its own optimum. Putting words first then costs nothing among the caption stages exactly when the captions form a summarization chain, and it costs the pixels at most the caption uncertainty left in a pixel-optimal reconstruction, which is large for short prefixes and small at full length. We build RUNG, an ordered code whose prefixes are supervised by a summarization ladder of captions and whose full length reconstructs the image; generation extends the code past the prompt and captioning truncates it. With data and capacity held fixed, ordering by language instead of reconstruction raises prefix-caption information 2.9 times and GenEval from 0.76 to 0.83, at a measured cost of 2.8 dB PSNR at a quarter of the code and 0.1 dB at full length. RUNG-XL (3.4B) reaches 0.88 GenEval and 87.6 DPG-Bench in 32 network evaluations and 0.91 GenEval when errors are redrawn at the rung where they appear; its coarse and fine captions agree 94.1% of the time, and it edits images without inversion.
Abstract (translated)
URL
https://arxiv.org/abs/2610.16552
PDF
-
VisExMEM: Visual Evidence-Based Explainable Multimodal Evaluation Metric
2026-10-12 09:07:37Subhadeep Roy, Christoph Leiter, Steffen EgerAbstract
Current image-text evaluation often relies on a global score, which gives limited insight into which semantic details are supported by the image. Explainable metrics provide rationales, but the visual evidence supporting those rationales can still be difficult to verify. We introduce VisExMEM, a reference-free evaluator that decomposes captions into atomic claims, localizes visual evidence, verifies each claim, and averages claim-level support into a global score. Explanations use the same evidence and decisions. We train Qwen3.5-based 4B and 9B models on 100K samples with multi-task SFT, counterfactual refinement, and evidence-aware GDPO. Across eight benchmarks, VisExMEM-9B achieves the highest descriptive composite among open-source evaluators in our comparison, exceeding VQAScore by 11.7 points while remaining competitive with GPT-5.6 Sol. We also introduce AGITA-Bench, containing 600 human-annotated image-text pairs and 8.7K atomic claims for global and claim-level evaluation. For untuned Qwen3.5 and GPT-5.6 Sol controls, decomposed prompting raises the same composite by 3.6-5.1 points relative to holistic prompting. Human evaluation shows that VisExMEM explanations outperform prior explainable metrics and approach GPT-5.6 Sol.
Abstract (translated)
URL
https://arxiv.org/abs/2610.16441
PDF
-
EventLens: Event-Structure Reinforcement Learning for Video Understanding
2026-10-12 04:46:04Guowen Zhang, Boshen Xu, Zihao Yue, Ziheng Wang, Xiaokun Liu, Xin Tao, Wenyu Qin, Pengfei Wan, Qin JinAbstract
Recent multimodal large language models (MLLMs) achieve strong results on video question answering, yet often fail to recover the temporal structure that makes a video coherent. They may merge neighboring events, miss single-event progression, or infer implausible relations across events. We study structured temporal understanding as the ability to recover event structure from visual evidence, including event boundaries, single-event progression, and multi-event relations. We argue that this gap arises from a mismatch between current post-training objectives and temporal reasoning, where caption-style supervision allows models to rely on language shortcuts rather than visual temporal grounding. To address this, we propose EventLens, a vision-centric reinforcement learning framework that learns video understanding through event-structure recovery. Instead of reconstructing captions, EventLens trains models to recover temporal structure under controlled perturbations, such as altered temporal granularity, reversed progression, and shuffled event order. We instantiate this framework with a three-level temporal ontology and derive three verifiable task families: event segmentation, progression discrimination, and multi-event relation reasoning. These tasks admit deterministic rewards, enabling task-specialized GRPO training without human preference labels or learned reward models. We further introduce multi-teacher on-policy distillation (MT-OPD) to consolidate specialized policies into a unified model. Experiments on Qwen3-VL backbones show consistent improvements on temporal grounding and event-sensitive reasoning benchmarks, while preserving general video-QA performance. Learning curves further demonstrate that EventLens achieves comparable downstream transfer with significantly reduced training cost compared to direct mixed RL optimization.
Abstract (translated)
URL
https://arxiv.org/abs/2610.16172
PDF
-
Across Viewpoints and Time: Trajectory-Generalizable Physical Attacks on World Models and Video Understanding Models
2026-10-11 04:49:53Song Xia, Meiwen Ding, Chenqi Kong, Wenhan Yang, Xudong JiangAbstract
World models and video understanding models are increasingly deployed in dynamic visual environments, yet their robustness to physical adversarial threats remains largely underexplored. We study whether a single physical adversarial texture can mislead the victim model across different real-world scenarios where observing viewpoints vary along uncontrolled object--camera trajectories. We identify that heterogeneous video semantics and trajectories induce poorly aligned gradients for optimizing the adversarial texture. To address that, we propose the Trajectory-aware adversarial Alignment and Refinement Attack (TARA), which incorporates trajectory-aware optimization to capture the evolving geometric variations of a printed texture under real-world scenarios. To improve cross-instance optimization compatibility, TARA incorporates trajectory-aware contrastive alignment to encourage attacked representations from diverse video--trajectory instances to approach a shared adversarial region and then performs trajectory-aware task refinement to translate the aligned representation into task-level prediction errors. For reproducible evaluation, we develop a geometry-grounded tracking pipeline that provides physically plausible adversarial regions with temporally consistent trajectories on EPIC-KITCHENS-100 and Something-Something V2. We further collect real-world print-and-recapture videos to validate attack effectiveness under physical deployment conditions. Together, the annotated benchmarks and real-world captured data form a reproducible testbed for future evaluation. Notably, our results reveal that a single printable adversarial texture can persistently disrupt world models and video understanding models under physical conditions, highlighting the fundamental security challenge.
Abstract (translated)
URL
https://arxiv.org/abs/2610.15343
PDF
-
Understanding Unified VLMs: From Attention Probing to Guidance Tuning for Multimodal Generation
2026-10-10 23:40:27Zhengyang Luo, Xiaoyu Qiu, Zhiheng Tan, Yuchen Liang, Yingbin Liang, Liyue ShenAbstract
Unified vision-language models (VLMs) perform both text-to-image (T2I) and image-to-text (I2T) generation with one shared Transformer backbone, yet how the backbone processes the condition in either direction remains poorly understood. We study this question from the perspective of conditional guidance: we probe every attention head to show the activation difference between two branches of classifier-free guidance (CFG) with different conditional guidance. On two unified VLMs of autoregressive and masked-diffusion paradigms, the probing reveals a subset of heads that are more sensitive to the change of condition while others are agnostic to it. This characteristic pattern holds in both T2I and I2T generation and is largely stable across data samples, demonstrating the internal mechanism of unified VLMs. Motivated by this, we propose a new inference-time intervention approach for unified multimodal generation, named \textit{HeadTuned-CFG}, which uses the probed condition-change awareness to select and tune the conditional guidance at the attention-head level. Experiments show consistent benefits in both multimodal generation directions of unified VLMs without additional training, and in particular substantial improvements over strong baselines in T2I generation. These results further support the key finding of our probing analysis: the observed pattern of functional specialization is an intrinsic property of unified VLMs.
Abstract (translated)
URL
https://arxiv.org/abs/2610.15227
PDF
-
EarthAlign: Learning Language-Aligned Representations of Earth System Data
2026-10-10 06:14:12Fadillah Adamsyah Maani, Numan Saeed, Jiaqi Ma, Dat Tien Nguyen, Youssef Wehbe, Duncan Watson-Parris, Salman KhanAbstract
Interpreting Earth system conditions is central to understanding environmental risks and informing decisions across weather- and climate-sensitive sectors. Existing multimodal weather systems primarily target task-specific prediction, generation, or question answering, while general-purpose alignment between Earth system data and language remains underexplored due to the limited availability of diverse language supervision grounded in meteorological conditions. To fill this gap, we introduce EarthAlign, a large-scale dataset with broad geographic coverage spanning 2000-2024 that combines 10 expert-curated natural event and disaster data sources with climatology-derived normal-weather samples. Each sample includes temporal and geospatial information and four grounded caption styles aligned with 63-variable ERA5 meteorological data. Building on EarthAlign, we introduce MeteoLIP, a dual-encoder framework for learning aligned meteorology-language representations. MeteoLIP incorporates MetTok, which uses a geolocation- and time-conditioned query to aggregate variable-specific patch embeddings into a compact sequence of meteorology tokens. To comprehensively assess Earth system multimodal representations, we introduce three downstream tasks: cross-modal retrieval, weather-condition classification, and physical variable estimation. Our experiments show that MeteoLIP improves upon the corresponding CLIP and SigLIP baselines across these tasks. Further analyses highlight the importance of meteorological input, temporal and geospatial conditioning, and multi-style supervision for meteorology-language alignment. Thus, EarthAlign and MeteoLIP provide a foundation for learning and evaluating language-aligned representations of Earth system data. Our code, dataset, and weights are publicly available at this https URL.
Abstract (translated)
URL
https://arxiv.org/abs/2610.14624
PDF
