Videos

AI glasses present a compelling platform for AI agents to serve as personalized memory assistants. To be genuinely useful, such systems must move beyond short-term video comprehension and address memory gaps that humans experience for practical, personal, or social purposes over longitudinal egocentric video streams. However, existing egocentric datasets predominantly focus on action recognition or generic QAs from short clips, measuring perceptual capabilities rather than realistic human memory needs. We introduce SuperMemory-VQA, an egocentric visual question answering (VQA) dataset for evaluating AI assistants on practical, long-horizon memory tasks. It contains 52.9 hours of everyday activities recorded with AI glasses, including synchronized RGB video, audio transcription, eye gaze, IMU, and SLAM trajectories. Through a human-verified annotation pipeline, we construct grounded 4,853 question-answer pairs that span object and location memory, intent recall, visual scene recall, timeline reconstruction, conversational memory, and in-context retrieval. Each question is posed as multiple-choice with an explicit "unanswerable" option to test hallucination robustness. Benchmarking leading agentic frameworks and LLM backbones reveals that existing systems remain far from reliable on real-world memory tasks, highlighting the need for new architectures for grounded AI memory that can answer only when evidence is sufficient. A participant survey further supports that our questions are realistic, useful, and aligned with everyday memory needs.

Recent advancements in video-based world models have demonstrated an unprecedented ability to synthesize high-fidelity visual sequences. However, a fundamental gap persists between visually plausible video generation and the functional requirements of a world model, particularly in maintaining a stable and reasonable internal state over extended temporal horizons. While existing benchmarks primarily emphasize visual quality, motion coherence, and text-video alignment, they largely overlook memory, the core capability of a world model to preserve consistency across long-term horizons and complex interactions. To address this gap, we present \textbf{MBench}, a comprehensive benchmark dedicated to quantifying and evaluating the memory capability of video world models. We systematically decompose the memory capability of video world models into three hierarchical and complementary core dimensions: entity consistency, environment consistency, and causal consistency, which are further refined into 12 quantifiable sub-dimensions for comprehensive characterization of long-term memory. Our benchmark is built upon rigorously curated real-captured long videos, and evaluated by rule-based quantitative matrices and VLM to enable objective and comprehensive consistency assessment. Extensive evaluations of mainstream state-of-the-art video world models reveal critical systemic limitations of existing methods in long-term state retention, providing a standardized benchmark and clear research direction to advance the field.

Visual Speech Recognition (VSR) aims to recognize speech from visual cues such as lip movements, but its performance is fundamentally limited by viseme ambiguity and pose-induced variations that introduce geometric distortions and occlusions. Existing approaches mainly rely on linguistic context or implicit invariance, leaving visual representations insufficiently robust under non-frontal views. In this work, we propose a pose-aware phoneme-level framework, termed HP-VSR-ResFiLM, that explicitly incorporates head-pose information into visual feature extraction. The proposed framework adopts a two-stage pipeline consisting of a pose-conditioned visual encoder in Stage 1 and a pretrained NLLB language model in Stage 2 for phoneme-to-text reconstruction. Specifically, Stage 1 incorporates a pose-conditioned residual Feature-wise Linear Modulation (FiLM) block after the 2D CNN frontend to adaptively refine visual representations using head-pose information. Experiments on LRS2 and LRS3 demonstrate that HP-VSR-ResFiLM achieves competitive performance under comparable training conditions, attaining word error rates (WER) of 25.0% and 33.2%, respectively, without relying on additional training data. Ablation studies further show that a single residual FiLM block consistently improves overall WER, while deeper modulation at Layers 3 and 4 provides larger gains for samples with yaw angles greater than 30° without degrading performance for smaller pose variations. These findings demonstrate that explicit pose-aware feature modulation offers an effective and computationally efficient solution for improving VSR robustness in unconstrained settings.

The CASTLE Challenge @ EgoVis 2026 evaluates long-form egocentric video question answering over 600+ hours of multi-perspective recordings. Each four-choice question requires evidence from videos, transcripts, auxiliary photos, people, days, rooms, and temporal context. We propose an evidence-aware multimodal reasoning pipeline based on Qwen. Our system parses question hints, retrieves ASR chunks, attaches auxiliary images, samples candidate video frames, and routes questions into static visual, speech/text, temporal, and mixed types with specialized prompts. Multiple inference passes are aggregated by confidence-weighted voting and converted into the official Codabench format. In ablation, LoRA improves the score from 0.21 to 0.50, and more sampled frames further raise it to 0.58. Our final system ranks first in the CASTLE Challenge @ EgoVis 2026.

Recent Large Multimodal Models (LMMs), primarily designed for offline settings, are ill-suited for the dynamic requirements of streaming video. While recent online adaptations improve real-time processing, they still face critical scalability challenges, with resource demands typically growing at least linearly with video duration. To overcome this bottleneck, we propose FlowNar, a novel framework for scalable streaming video narration. The core of FlowNar is a dynamic context management strategy for historical visual context removal, combined with our CLAM (Cross Linear Attentive Memory) module for streaming visual history retention, ensuring bounded visual memory usage and computational complexity, crucial for efficient streaming. We also introduce a realistic self-conditioned evaluation protocol and complementary evaluation metrics to assess streaming narration models under deployment-like conditions. Experiments on the Ego4D, EgoExo4D, and EpicKitchens100 datasets demonstrate that FlowNar substantially improves narration quality over strong baselines while being highly efficient, supporting processing of 10$\times$ longer videos and achieving 3$\times$ higher throughput (FPS). The code is available at https://github.com/zeyun-zhong/FlowNar.

As multimodal LLMs increasingly target video and audio, it is often assumed that such tasks require native omnimodal models. We show that this is not always the case: coding agents with only text+image access and a sandboxed tool-use interface can match, and in several settings outperform, SOTA native omnimodal models and predefined multimodal agent scaffolds across multiple audio-video benchmarks. Our trajectory analysis suggests that their strength comes from writing code and orchestrating tools to extract relevant evidence from transcripts, frames, and other modality signals, thereby converting omnimodal tasks into retrieval and information-processing problems rather than ingesting entire media streams. We further characterize their limitations through a failure taxonomy and process-level trace analysis, and show that simple skill injection, including human-written and self-distilled skills, substantially improves performance. To explore open-source elicitation, we introduce Code-X, a training recipe with the OmniCoding trajectory dataset and verifiable reward, and provide baselines on Qwen-3.5-9B and Qwen-3.6-27B. Finally, we argue that the next frontier is many-modality processing, and introduce TerminalBench-O, a process-level benchmark for real-world omnimodal processing tasks. Code will be available at https://github.com/Dongping-Chen/OmniCoding.

Video vision-language models (VLMs) are increasingly used in long-horizon and streaming settings, yet most video encoders still rely on spatiotemporal self-attention, causing compute and latency to grow quadratically with the number of frames. Existing efficiency methods improve scalability but often lose accuracy relative to full self-attention, for example through aggressive frame/token dropping or coarse attention approximations. We introduce StateKV, an inference-time method that adapts pretrained long-video VLMs to linear-time video prefill by carrying cross-frame context in a fixed-capacity, importance-based recurrent state, paired with a second full per-frame cache used for decoding. Across three long-video benchmarks and seven models spanning three families and multiple scales, StateKV remains close to full self-attention and consistently outperforms dominant sliding-window / recency-based streaming approximations, without fine-tuning or architectural changes. StateKV also reduces video-prefill cost measured FLOPs, enabling stronger accuracy at a fixed compute budget by running larger models. These results suggest a practical step toward scalable long-video understanding.

Text-to-video (T2V) generation faces challenging questions when generating videos with long horizons containing multiple events. Inspired by the intrinsics of the diffusion process, we probe video diffusion transformers (DiTs) and uncover intrinsic turning points in the DiT denoising trajectory where conditioning text affects generation from global layout to fine-grained details. Building on this finding, we present TunerDiT, a simple yet effective progressive steering method that requires no additional training for multi-event generation. TunerDiT comprises two steering handles: (1) Event-Partitioned Masking that enforces event boundaries while allowing cross-event transition bands; (2) Cross-Event Prompt Fusion that injects neighboring event semantics for late-stage refinement. We contribute a self-curated prompt suite for benchmarking multi-event generation, i.e., Meve. TunerDiT achieves state-of-the-art performance across 8 metrics and offers a tunable trade-off between video consistency and event separation, compared with other training-free methods. The improvement in text alignment increases with the event count, indicating a scaling possibility with increasing event count.

Self-supervised novel view synthesis (NVS) remains challenging to scale, despite the abundance of video data, largely due to the brittleness of training on realistic videos and the hard-to-predict scaling behavior of multi-network system designs. We introduce RayDer, a unified, feed-forward transformer that consolidates camera estimation, scene reconstruction, and rendering into a single backbone, turning self-supervised NVS into a well-posed single-model scaling problem. A minimal dynamic state, treated as a nuisance factor, absorbs time-varying content and enables stable training on unconstrained real-world video. Importantly, RayDer keeps static-scene NVS as its target task: dynamic content is leveraged purely as scalable supervision, not reconstructed as in dynamic-scene (4D) NVS. Across multiple model sizes and orders of magnitude in data, RayDer exhibits clean power-law scaling with data and compute, and outperforms static-scene data mixtures. On a large number of benchmarks, RayDer achieves strong zero-shot open-set performance competitive with state-of-the-art supervised approaches. Project Page: https://compvis.github.io/rayder

True video intelligence demands more than recognizing what is visible: it requires reasoning about why events unfold, predicting what would change under different conditions, and deciding what to do next. We refer to this progression, from perception through causal reasoning and simulation to strategic planning, as Strategic Video Intelligence (SVI). No existing benchmark evaluates this capability stack: in-the-wild videos lack verifiable ground truth for causal and strategic questions, while synthetic environments sacrifice the complexity of real multi-agent systems. To bridge this gap, we introduce SVI-Bench, a large-scale benchmark that leverages team sports as a dynamic microworld, combining the complexity of real-world multi-agent interaction (10-22 agents making coordinated decisions under adversarial pressure) with the verifiability of explicit rules and definitive outcomes. SVI-Bench comprises approximately 35K hours of broadcast video, 15M annotated actions, 15K hours of expert commentary, 23K game reports, and 103K structured statistical records across basketball, soccer, and hockey, all constructed via a data engine that transforms raw game data into a dense, cross-referenced corpus. We organize evaluation into 9 tasks spanning a progressive four-pillar hierarchy: Dynamic Scene Understanding, Causal Reasoning, Strategic Simulation, and Agentic Synthesis. Evaluating strong multimodal and agentic baselines, we find a capability cliff: models perform competently on perceptual tasks, achieving approximately 74% on fine-grained action QA, but degrade sharply at each successive cognitive level. Agentic tasks prove hardest: the strongest model achieves only 5% accuracy when required to autonomously gather and integrate evidence across a corpus of 1.8M clips.

Accurately predicting future events is fundamental to content understanding and decision-making across various domains. While prior research has primarily focused on text or short-video scenarios, long-video event prediction, characterized by vast multimodal context and more complex narratives, remains underexplored. Meanwhile, although recent Long-Video Language Models (LVLMs), built on Large Language Models (LLMs) and Vision-Language Models (VLMs), have shown promise in long-video question answering and summarization, they struggle to generalize to event prediction, as they can neither precisely extract event-related details nor perform fine-grained analysis of event development. To address this gap, we propose VISTA, a multi-level event semantics mining framework for long-video event prediction. Initially, VISTA applies a character-centric visual prompt to precisely extract event-related visual details, enhancing detail-level semantics; subsequently, it employs a knowledge-enhanced iterative retrieval strategy, guiding the LLM to progressively construct logically coherent event chains, thereby improving event-level narratives; ultimately, VISTA adopts a human-like propose-then-retrieve strategy to generate diverse future-oriented proposals and integrate multi-level clues, producing robust and accurate predictions. Extensive experiments on real-world datasets validate the effectiveness of VISTA for long-video event prediction.

Video-language models can process only a limited number of frames, making frame selection a key bottleneck for efficient video captioning. Most captioning pipelines still rely on uniform sampling, which is computationally cheap but agnostic to visual content. Adaptive frame sampling has recently emerged as a promising approach for selecting the most informative frames from a video; however, existing methods remain computationally expensive. We introduce PEEK, an efficient dynamic frame sampling method that distills caption-conditioned frame relevance rankings from a stronger teacher model into a lightweight temporal model that operates only on visual content. We find that, overall, on ActivityNet Captions and MSR-VTT, our method outperforms state-of-the-art methods across all evaluated downstream vision language models, especially when only one or two frames are selected for captioning, obtaining the best CIDEr for most frame budgets. On ActivityNet Captions, PEEK is particularly strong, winning 14 out of 16 configurations. Zero-shot evaluation on MSR-VTT shows that our model transfers best at low frame budgets, while results at four and eight frames are more mixed as temporal coverage and visual diversity become increasingly competitive. Compared with recent adaptive baselines, PEEK is both more accurate in the low-budget regime and more efficient: it adds only $5.2\%$ to the captioning time, compared with $65.4\%$ for CSTA and $211.9\%$ for MaxInfo. We release our code and pre-trained checkpoint at https://github.com/momentslab/peek.

In the era of Large Video-Language Models (LVLMs), the computational necessity of sparse frame sampling creates a fundamental ``temporal gap'', rendering models blind to critical causal transitions. Existing solutions relying on generative hallucination (e.g., latent diffusion) or autoregressive extrapolation often fail to maintain semantic consistency over long horizons, suffering from object vanishing and energetic instability. We propose a paradigm shift from probabilistic generation to variational mechanics with the \textbf{Semantic Least Action Principle (SLAP)}. Drawing a rigorous isomorphism between classical mechanics and semantic dynamics, we model the latent video trajectory as a path on a Riemannian manifold governed by a Semantic Lagrangian. By formulating the interpolation task as a Boundary Value Problem (BVP) solved via the discrete Euler-Lagrange equations, SLAP naturally enforces object persistence without pixel-level rendering. Extensive experiments show the effectiveness of our proposed SLAP.

This paper addresses the task of temporal sentence grounding (TSG). Although many respectable works have made decent achievements in this important topic, they severely rely on massive expensive video-query paired annotations, which require a tremendous amount of human effort to collect in real-world applications. To this end, in this paper, we target a more practical but challenging TSG setting: unsupervised temporal sentence grounding, where both paired video-query and segment boundary annotations are unavailable during the network training. Considering that some other cross-modal tasks provide many easily available yet cheap labels, we tend to collect and transfer their simple cross-modal alignment knowledge into our complex scenarios: 1) We first explore the entity-aware object-guided appearance knowledge from the paired Image-Noun task, and adapt them into each independent video frame; 2) Then, we extract the event-aware action representation from the paired Video-Verb task, and further refine the action representation into more practical but complicated real-world cases by a newly proposed copy-paste approach; 3) By modulating and transferring both appearance and action knowledge into our challenging unsupervised task, our model can directly utilize this general knowledge to correlate videos and queries, and accurately retrieve the relevant segment without training. Extensive experiments on two challenging datasets (ActivityNet Captions and Charades-STA) show our effectiveness, outperforming existing unsupervised methods and even competitively beating supervised works.

Studying scalable methods for visual generative modeling requires large, accessible, and stable datasets. We introduce GPIC, a Giant Permissive Image Corpus of approximately 28 trillion pixels. GPIC comprises diverse internet images captioned by a state-of-the-art vision-language model, including 100M training, 200K validation, and 1M test examples. Moreover, all GPIC images are permissively licensed for both research and commercial use. GPIC is safety-filtered, deduplicated, and centrally hosted on Hugging Face. We provide a benchmarking protocol for generative modeling on GPIC. Finally, we provide a reference baseline for pixel-space flow matching on GPIC. Our dataset, benchmark, and models are available at https://huggingface.co/datasets/stanford-vision-lab/gpic. Evaluation toolkit and code are available at https://gpic.stanford.edu

Natural human conversation is full-duplex and audio-visual: people simultaneously speak and listen while continuously interpreting and producing nonverbal cues, such as nods, smiles, and gestures. To support successful human-agent interaction, agents must model full-duplex audiovisual conversation; however, existing full-duplex benchmarks evaluate only speech. In this work, we present VideoFDB, the first benchmark to evaluate full-duplex audio-visual-to-audio-visual (AV2AV) conversational agents. VideoFDB contributes (i) 237 dyadic clips spanning 11 nonverbal conversational dynamics from real-world video calls, (ii) a taxonomy separating perception from generation behaviors, and (iii) a rubric-based LM-as-judge evaluation framework with interpretable axes for assessing conversational quality with respect to nonverbal conversational dynamics. Across open- and closed-source vision-speech agents, we find systematic failure modes: captioning collapse and visual-stream ignorance, and we show that current systems exploit vision for explicit visual question answering but not for the streaming joint audiovisual grounding required in natural conversation. We further evaluate cascaded speech-to-avatar systems and find that their architecture fundamentally precludes the production of full-duplex nonverbal cues. As the first benchmark for full-duplex AV2AV interaction, VideoFDB establishes a foundation for systematic evaluation and, we hope, will accelerate the advancement and development of next-generation multimodal conversational agents.

Video large language models (Video-LLMs) have demonstrated strong capabilities in video understanding tasks. However, their practical deployment is still hindered by the inefficiency introduced by processing massive amounts of visual tokens. Although recent approaches achieve extremely low token retention ratios while maintaining accuracy comparable to full-token baselines, most of them perform compression only at the late stage of prefilling, leaving the efficiency of the vision encoder unoptimized. In this paper, we first show that vision encoding contributes a large portion to the time-to-first-token (TTFT). Therefore, instead of compressing visual tokens only after the vision encoder, performing compression inside the encoder still leaves substantial room for exploration. Based on this insight, we propose EarlyTom, a training-free token compression framework that performs early-stage visual token compression inside the vision encoder, enabling significantly better TTFT reduction and higher throughput. In addition, we introduce a decoupled spatial token selection strategy that improves the overall compression effectiveness. EarlyTom reduces TTFT by up to 2.65x and FLOPs by up to 61% on a single NVIDIA A100 GPU for the LLaVA-OneVision-7B model, while maintaining accuracy comparable to the full-token baseline. These improvements substantially enhance the practicality of deploying Video-LLMs in real-world production scenarios.

Video Moment Retrieval (VMR) targets to retrieve the specific moment corresponding to a sentence query from an untrimmed video. Although recent works have made remarkable progress in this task, they implicitly are rooted in the closed-set assumption that all the given queries as video-relevant\footnote{In this paper, we treat video-relevant query'' as in-distribution (ID) query'' and video-irrelevant query'' as out-of-distribution (OOD) query''.}. Given an OOD query in open-set scenarios, they still utilize it for wrong retrieval, which might lead to irrecoverable losses in high-risk scenarios, \textit{e.g.}, criminal activity detection. To this end, we creatively explore a brand-new VMR setting termed Open-Set Video Moment Retrieval (OS-VMR), where we should not only retrieve the precise moments based on ID query, but also reject OOD queries. In this paper, we make the first attempt to step toward OS-VMR and propose a novel model \textbf{OpenVMR}, which first distinguishes ID and OOD queries based on the normalizing flow technology, and then conducts moment retrieval based on ID queries. Specifically, we first learn the ID distribution by constructing a normalizing flow, and assume the ID query distribution obeys the multi-variate Gaussian distribution. Then, we introduce an uncertainty score to search the ID-OOD separating boundary. After that, we refine the ID-OOD boundary by pulling together ID query features. Besides, video-query matching and frame-query matching are designed for coarse-grained and fine-grained cross-modal interaction, respectively. Finally, a positive-unlabeled learning module is introduced for moment retrieval. Experimental results on three VMR datasets show the effectiveness of our OpenVMR.

Given an untrimmed video and a sentence query, video moment retrieval using language (VMR) aims to locate a target query-relevant moment. Since the untrimmed video is overlong, almost all existing VMR methods first sparsely down-sample each untrimmed video into multiple fixed-length video clips and then conduct multi-modal interactions with the query feature and expensive clip features for reasoning, which is infeasible for long real-world videos that span hours. Since the video is downsampled into fixed-length clips, some query-related frames may be filtered out, which will blur the specific boundary of the target moment, take the adjacent irrelevant frames as new boundaries, easily leading to cross-modal misalignment and introducing both boundary-bias and reasoning-bias. To this end, in this paper, we propose an efficient approach, SpotVMR, to trim the query-relevant clip. Besides, our proposed SpotVMR can serve as plug-and-play module, which achieves efficiency for state-of-the-art VMR methods while maintaining good retrieval performance. Especially, we first design a novel clip search model that learns to identify promising video regions to search conditioned on the language query. Then, we introduce a set of low-cost semantic indexing features to capture the context of objects and interactions that suggest where to search the query-relevant moment. Also, the distillation loss is utilized to address the optimization issues arising from end-to-end joint training of the clip selector and VMR model. Extensive experiments on three challenging datasets demonstrate its effectiveness.

We introduce MMTM, a modular pipeline for topic discovery in long-form video that integrates speech recognition, audio and visual embeddings, and BERTopic clustering through a deterministic similarity-gated fusion. Evaluated cross-lingually on German (Tagesschau) and English (NBC) broadcast news, joint tri-modal modeling substantially improves topic quality: noise drops from 0.27 to 0.06, transition rate from 0.70 to 0.21, and normalized entropy rises from 0.84 to 0.92, indicating more coherent and temporally stable topics. Cluster validity (Calinski-Harabasz) improves by 5-12X across embedding spaces. Lexical coherence (NPMI) rises from 0.77 to 0.86 on German but is corpus-dependent and does not transfer to the shorter NBC broadcasts. We release the pipeline code and a human-validated 54-hour multimodal video topic corpus with dual-annotator visual evaluation and LLM-assisted labeling.

Video Motion Magnification (VMM) reveals imperceptible dynamics but often suffers from structural inconsistencies under complex geometric transformations. Existing learning-based methods generally face a trade-off between the limited global context of CNNs and the high computational cost of Transformers. In addition, current training protocols, largely dominated by simple linear motion, fail to capture the geometric and imaging complexities encountered in real-world videos. To address these issues, we propose GeoMag, a geometric-aware VMM framework built upon State Space Models to achieve globally consistent motion amplification with linear complexity. We further construct Geo-200K, a large-scale synthetic dataset that introduces rich geometric transformations together with sensor-realistic degradations, improving the diversity and realism of training signals. Extensive experiments on synthetic and real-world benchmarks show that GeoMag consistently outperforms prior methods in visual fidelity and computational efficiency, while producing fewer artifacts and better structural consistency.

Cross-Video Reasoning (CVR) has emerged as a critical frontier in multimodal intelligence, requiring models to retrieve, align, and aggregate evidence distributed across multiple videos. Current Multimodal Large Language Models (MLLMs) often struggle with CVR, as simple single-pass strategies encode multiple videos into a shared compressed context, potentially obscuring rare but critical evidence. In this paper, we propose AgentCVR, a multi-agent framework that treats CVR as an active evidence-acquisition task. AgentCVR employs a Master Agent to iteratively coordinate specialized Visual and Audio Agents for targeted evidence extraction. To ensure efficient training, we introduce Script-Simulated RL, which optimizes the agent's policy with LLM-generated semantic scripts and a lightweight text-based simulator, bypassing costly multimodal inference during online exploration. Experimental results on a comprehensive CVR benchmark show that AgentCVR outperforms single-pass baselines and achieves comparable performance to state-of-the-art closed-source systems, particularly in complex cross-video alignment and localization. To ensure reproducibility, our code is available at https://github.com/wang-jh24/AgentCVR.

Understanding long-form egocentric videos remains challenging for multimodal large language models (MLLMs) due to limited context length and insufficient grounding of fine-grained visual details. The recently proposed HD-EPIC benchmark highlights these limitations: even strong long-context models achieve relatively low performance across diverse video question answering tasks. In this paper, we propose a unified framework that decouples long-video reasoning into two complementary forms of evidence: semantic evidence and visual evidence. Semantic evidence captures global procedural structure through a coarse-to-fine extraction pipeline, while object-centric visual evidence preserves fine-grained grounding through bounding boxes and visual embeddings. During inference, we formulate reasoning as a query-conditioned evidence retrieval and integration process, dynamically selecting relevant information from both sources. Our approach achieves competitive performance in the HD-EPIC-VQA Challenge across multiple task categories. More broadly, our results demonstrate that explicitly structuring, retrieving, and integrating semantic and visual evidence is critical for effective long-video understanding with MLLMs.

We present the 1st-place solution to the ACCIDENT challenge at the CVPR 2026 AUTOPILOT Workshop, which asks for zero-shot prediction of accident timing, impact centroid, and collision type from CCTV footage. On a frozen Qwen3-VL-32B-Instruct checkpoint we build a three-stage pipeline (full-video joint prediction, time refinement, and single-frame grounding of the impact centroid), run the same pipeline a second time on a 235B Mixture-of-Experts sibling, blend the two outputs 9:1, and finally snap each predicted point onto the nearest vehicle detection. The final system reaches Public LB 0.55469 / Private LB 0.57080, roughly +0.21 over the strongest host baseline (Molmo-7B, 0.358) and wins the challenge. We ablate each component, report the negative results that shaped the final design, and release the code at https://github.com/fuumin621/cvpr2026-accident-1st-place-solution.

We present a method for harmonizing the lighting of a foreground video to match a target background scene, adjusting shadows, color tone, and illumination intensity (relightful harmonization). Unlike images, acquiring labeled data for videos, where identical motions are recorded under different lighting conditions, is practically infeasible and non-scalable. While one way to create such paired data is to apply existing image-based harmonization models frame by frame to a video, the resulting outputs often suffer from significant temporal jitters. We overcome this problem by introducing a novel lighting deflickering model that can stabilize the global and local lighting flickering artifacts. Our video diffusion model learns from these upgraded deflickered data with a volume of real and synthetic videos to generate high-quality video harmonization results. We further propose an asymmetric alpha mask conditioning technique to learn the clean boundaries from real videos. Experiments demonstrate that our model achieves strong temporal coherence, naturalness, cleaner boundaries, and physically meaningful lighting behavior, while maintaining strong relighting expressiveness compared to prior image-based and video-based harmonization methods.

As Text-to-Video (T2V) generation models continue to evolve, the complexity of video evaluation necessitates a fine-grained assessment across various axes. To address this, recent works have focused on developing Multidimensional Video Reward Models (MVRMs), which decompose the evaluation process to better align with the multifaceted nature of human visual perception. However, training effective MVRMs is fundamentally challenged by the complex nature of video data. In this work, we identify a critical phenomenon termed Dimensional Heterogeneity: the reliability of a training sample can vary substantially across evaluation dimensions, meaning that a sample may provide reliable supervision for one objective while inducing high supervision risk for another. Consequently, prevailing data-centric methods that filter based on global scalar metrics are ill-posed for T2V tasks. To address this, we propose a disentangled influence framework that that efficiently estimates dimension-specific supervision risk. Leveraging this framework, we introduce two dimension-disentangled refinement strategies: Dimension-Disentangled Pruning, which removes extreme high-risk samples, and Dimension-Disentangled Reweighting, which softly down-weights high-risk supervision. Extensive experiments demonstrate that our disentangled strategies significantly outperform global filtering baselines, yielding reward models with superior alignment to ground truth.

Colo-segment recognition in colonoscopy videos is a key requirement for many downstream tasks, but existing automatic recognition methods only use colonoscopy images without fully exploiting the use of temporal information, leading to poor performance. Additionally, relevant public video-based datasets are in scarcity. To tackle this problem, we curate and release a labeled dataset specifically for the task of colo-segment recognition. In addition, we propose a two-stage deep learning-based framework, Colo-Segment Recognition via SpatioTemporal Network (ST-ColoNet), for the task of colo-segment recognition from colonoscopy videos which includes the Colorlaus module that uses metric learning to optimize edge-mediated spatial feature extraction, as well as the Full-Temp module which combines three self-attention patterns to better approximate full self-attention on long colonoscopy sequences and optimize temporal feature aggregation. Through extensive ablation experiments, we show that our framework is capable of achieving state-of-the-art performance on the task of colo-segment recognition, achieving an accuracy of 81.0% and F1-score of 70.7%, which is a tremendous improvement over state-of-the-art methods.

Driven by the wave of large language models, Video-Language Models (VLMs) have become a significant yet challenging technology to bridge the gap between videos and texts. Although previous VLM works have made significant progress, almost all of them implicitly assume that all the texts are predefined by the specific template. In real-world applications, such a strict assumption is impossible to satisfy since 1) predefining all the texts is extremely time-consuming and labor-intensive. 2) these predefined text inputs are too restrictive and user-unfriendly, limiting their applications. It is observed that given a video input, texts with similar semantics but different templates lead to various performances. To this end, in this paper, we propose a novel plug-and-play framework for various VLM-based methods to fully bridge videos and texts. Specifically, we first generate positive and negative texts from the original ones to target specific text components. Then, we propose an attribute-based text reasoning strategy to mine fine-grained textual semantics of generated texts. Finally, we utilize videos as guidance to conduct cross-modal bridging by designing a self-weighted loss. Extensive experiments show that the proposed method can serve as the plug-and-play module to effectively improve the performance of state-of-the-art VLMs.

Video-Language Models (VLMs) have demonstrated impressive multi-modal reasoning capabilities across diverse computer vision applications. However, these VLMs are task-specific and assume that both video and language inputs are complete. However, real-world VLM applications might face challenges due to deactivated sensors (e.g., cameras are unavailable due to data privacy), yielding modality-incomplete data and leading to inconsistency between training and testing data. While straightforward incomplete input can boast training generalization-ability and lead to training failure, its potential risks to VLMs regarding safety and trustworthiness have been largely neglected. To this end, we make the first attempt to propose a unified incomplete video-language model to process the incomplete multi-modal inputs. Extensive experimental results show that our method can serve as a plug-and-play module for previous works to improve their performance in various multi-modal tasks.

Various approaches have been proposed to adapt Vision-Language Models (VLMs) to specialized domains for Video Question Answering, including fine-tuning and in-context learning. However, acquiring task-specific knowledge at the inference phase from only a small labeled support set without fine-tuning remains a challenge. In this paper, we propose a method that achieves adaptation solely through inference-time context injection. Our method first constructs a Reflective Dialogue (RD) -- a multi-turn conversation between two agents, in which Teacher poses each support question and delivers correctness feedback, and Solver answers and provides visual grounding explanations (or reflections) for both correct and incorrect answers. This dialogue history is then used as context at the inference phase. Experiments on the EgoCross benchmark demonstrate that our method outperforms both a baseline zero-shot setting and a standard in-context learning approach that passes support set examples directly, achieving 3rd place in the Open-source Track of the 1st Cross-Domain EgoCross Challenge at the CVPR 2026 EgoVis Workshop, for which this paper also serves as a technical report.

Video generation models (VGMs) are rapidly entering classrooms, yet existing benchmarks evaluate only perceptual quality, intrinsic faithfulness, generic safety, or video as a reasoning medium, and none assesses whether the outputs are educationally valid. In this work, we present EduVideoBench, the first balanced benchmark in the education domain, grounded in the Knowledge-Skills-Attitude (KSA) framework so that pedagogical adequacy and educational safety are evaluated jointly rather than as ad-hoc quality dimensions. Across five frontier VGMs, our results show substantial room for improvement across knowledge, skills, and attitude before they are classroom-ready. We complement this with a qualitative analysis of expert comments, finding that educational validity is multi-component, where a single misaligned element such as pacing, legibility, or notation can invalidate an otherwise correct video. We hope EduVideoBench will guide the development of VGMs that are pedagogically grounded and safe for the classroom.

This dataset provides high-resolution, annotated video sequences of shredded E40-grade steel and copper scrap on a conveyor belt. Captured in a controlled laboratory environment, the data reflects the industrial post-magnetic sorting stage, where manual intervention is typically required to remove copper contaminants. The dataset comprises 24,297 labeled frames across five subsets, featuring 396 steel and 101 copper objects categorized by size. It supports the development of machine learning models for material classification, object detection, and instance segmentation. Variations in object spacing and density are included to simulate realistic industrial sorting conditions. Ground truth annotations include pixel-wise segmentation masks and material classes. This dataset serves as a benchmark for evaluating automated sorting algorithms aiming to identify copper impurities within complex, heterogeneous steel scrap streams.

Recent video multimodal large language models (MLLMs) increasingly couple step-by-step reasoning with on-demand visual evidence retrieval, allowing models to revisit relevant video segments during inference. However, two structural gaps remain in existing thinking-with-video systems. (i) Sampling density is not a learnable decision: existing methods may let the model decide where to look, but the per-window frame rate is largely fixed. As a result, fine-grained evidence is often recovered through repeated retrieval calls, which increases inference context length and training difficulty. (ii) Retrieval and answer generation are usually optimized with a single trajectory-level advantage, so the "where to look" tokens and the "how to answer" tokens receive the same credit even when one is correct and the other is not. To address these gaps, we present DynFrame, a framework that emits the temporal window and the sampling density as native tokens within a single autoregressive pass. This learnable span-density retrieval enables acquiring multi-granularity evidence with a single retrieval step. Based on the above tokenized retrieval interface, we further introduce Segment-Decoupled GRPO (SD-GRPO), which splits each rollout at the retrieval boundary and assigns role-specific token-level advantages, separately crediting the sampling decision and the answer. Trained on the curated DM-CoT-74k and DM-RL-45k, DynFrame-4B is competitive with strong 7B-8B baselines across six benchmarks (NExT-GQA, Charades-STA, ActivityNet-MR, Video-MME, MLVU, LVBench), and DynFrame-8B sets new state-of-the-art on most metrics. Code is available at https://github.com/zhangguanghao523/DynFrame.

Omnimodal large language models enable unified audio video understanding, but long joint token sequences make inference costly, and existing benchmarks do not fully isolate audio visual association in noisy user generated videos. We introduce UGC-AVQA, a public UGC benchmark with 1,000 videos and 4,816 QA pairs, where an audio removal test ensures that benchmark questions require both acoustic and visual evidence. To reduce inference cost, we propose OMAC, a training free plug in compression method that preserves salient visual memory and temporally grounded audio anchors. To further make compact models robust to compressed inputs, we introduce O-MARC, a compression distillation framework for learning with memory compressed multimodal contexts. On Qwen2.5-Omni-3B, O-MARC improves the average score across four benchmarks to 45.8, outperforming full token inference at 44.1 and OmniZip at 41.0. OMAC also keeps inference efficient, reducing latency by 34.6% (1.53$\times$ speedup) and memory by 34.7% compared with full token inference.

With the rapid advancement of artificial intelligence generated content (AIGC) technologies, video forgery has become increasingly prevalent, posing new challenges to public discourse and societal security. Despite remarkable progress in existing deepfake detection methods, AIGC forgery detection remains challenging, as existing datasets mainly rely on open-source video generation models with quality far below that of commercial AIGC systems. Even datasets containing a few commercial samples often retain visible watermarks, compromising authenticity and hindering model generalization to high-fidelity AIGC videos. To address these issues, we introduce CoCoVideo-26K, a contrastive, commercial-model-based AIGC video dataset covering 13 mainstream commercial generators and providing semantically aligned real-fake video pairs. This dataset enables deeper exploration of the differences between authentic and high-quality synthetic videos and establishes a new benchmark for highly realistic video forgery detection. Building on this dataset, we propose CoCoDetect, a detection framework integrating contrastive learning with confidence-gated multimodal large language model (MLLM) inference. An R3D-18 backbone extracts spatio-temporal representations, while a confidence gate routes uncertain cases to an MLLM for reasoning about physical plausibility and scene consistency. Extensive experiments on CoCoVideo-26K and public benchmarks demonstrate state-of-the-art performance, validating the framework's robustness and generalizability. Our code and dataset are available at https://github.com/DonoToT/CoCoVideo.

Pre-trained video large language models excel at visual reasoning. However, they struggle when videos arrive with auxiliary streams, such as audio, depth map, or dense temporal evidence. In such a scenario, uniform fusion induces modality interference, allowing irrelevant channels to distract the model. To address this issue, we present a unified multimodal video understanding framework, named UniMVU, that performs instruction-aware fusion across video, audio, depth map, or any other modality inputs via two levels of dynamic gating: inner-modality gates emphasize salient regions within each modality, whereas modality-level gates re-weight whole streams; both are conditioned on the text instruction to adaptively balance modality importance. Our UniMVU combines cross-modal self-attention with instruction-driven inner-modality gating module and a modality-level gating module with control token; for time-aligned streams we further adopt a fast-to-slow fusion scheme that reduces redundancy. Across six benchmarks (AVQA, AVSD, Music-AVQA, ScanQA, SQA3D and MVBench), our UniMVU achieves consistent gains over static-fusion baselines achieving gains as high as 13.5 in terms of CIDEr metric. Further, our analysis shows that the gating mechanism aligns with the human-interpretable modality relevance, and ablations show the contributions of inner-modality and modality-level gating. Our UniMVU provides a simple, unified recipe for instruction-aware multimodal video understanding that scales to diverse modalities without hand-crafted fusion rules.

Many video reasoning tasks require tracking motion, temporal order, and evolving visual states across frames. Existing methods built on large vision-language models (LVLMs) often address this challenge by externalizing reasoning through textual chain-of-thought (CoT), keyframe selection, repeated frame reinsertion, or external tool use. While effective, such pipelines increase inference-time latency and engineering complexity, and they force temporal-visual evidence to be serialized into text or repeatedly re-encoded from frames. Inspired by the intuition that visual reasoning can occur implicitly before verbalization, we propose STORMS (Spatial-Temporal reasOning via inteRnalized Modeling), a two-stage framework that teaches LVLMs to reason through bounded continuous latent trajectories instead of explicit textual CoT. In Stage I, STORMS aligns latent tokens with thought-video representations derived from generated videos, grounding the latent states in dynamic visual evidence. In Stage II, the model is further trained with answer-only supervision, encouraging the reasoning process to be internalized without step-by-step annotations. Generated thought videos are used only during training; at inference, STORMS performs a bounded latent rollout without regenerating videos, reinserting frames, or invoking external visual tools. Experiments on VideoMME, MVBench, TempCompass, and MMVU show that STORMS improves video reasoning accuracy while substantially reducing inference overhead compared with tool or video-generation-based reasoning pipelines.

We introduce LLaVA-OneVision-2 (LLaVA-OV-2), the most capable vision-language model in the LLaVA-OneVision series to date, achieving superior performance across a broad range of multimodal benchmarks. The model builds on a native OneVision-Encoder and incorporates Windowed Attention for efficient local computation while maintaining native resolution. Its key advance is codec-stream tokenization: it treats compressed video as a continuous bit-cost stream, where bit-cost dynamics determine adaptive temporal groups, and motion-residual cues select salient spatial evidence into compact visual canvases. This allocation concentrates a limited token budget on event-bearing content, enabling more stable long-video token compression than fixed groups of pictures. A shared 3D RoPE further places codec canvases, sampled frames, and images in a unified spatiotemporal coordinate system. Furthermore, we build the LLaVA-OV-2 data and training stack around large-scale open supervision: approximately 8M re-captioned video samples for pretraining, a 4M-sample spatial corpus for fine-tuning. We also introduce JumpScore, a temporal-localization benchmark targeting fine-grained grounding in high-frequency, densely repeated motion, a regime underrepresented by existing video evaluations. A standout capability of LLaVA-OV-2 is its unified perception across video understanding, temporal grounding, spatial grounding, and manipulation-trace reasoning. On JumpScore, LLaVA-OneVision-2-8B reaches 74.9 JumpScore mAP, surpassing Qwen3-VL-8B (30.1) by +44.8 points; under matched visual-token budgets on the same benchmark, codec-stream inputs improve temporal grounding over frame sampling by +9.7 points. Across standard benchmarks, LLaVA-OneVision-2-8B further outperforms Qwen3-VL-8B by +4.3 average points on video tasks, +5.3 on spatial tasks, and +15.6 average J&F on tracking tasks.

Women's safety and security are paramount for a modern society. Crimes against women occur in daylight as well as in low-light conditions. Often, such events are captured through real-world surveillance cameras that operate at lower resolutions. Despite substantial progress in CV-related research, video anomaly detection (VAD) focused on women's safety has not yet been adequately addressed. Existing video anomaly datasets contain well-lit, high-resolution, close-shot videos, and fail to represent women-centric anomalies such as chain snatching, stalking, inappropriate touch, and other subtle forms of crime against women. To address these problems, we propose the ExtrAnom dataset, a new multi-modal benchmark containing 1001 videos with textual descriptions, 500 normal and 501 anomalous, classified into 5 different types of women-centric crimes. The dataset comprises low-light (8%), low-resolution videos (13%), long-shot (15%), along with daylight (64%) anomalous videos. And it covers anomalous events like stalking (3.9%), chain snatching (17.6%), kidnapping (7.3%), assassinations (2.3%), harassment (18.9%), and normal (50%). Each video is supplemented with 4 textual annotations, including one human-generated and three LLM-generated descriptions, enabling cross-modal and VLM-based validations. The aim of creating a women-centric dataset is to accurately detect the women-centric anomaly patterns, which are possible to observe visually. The dataset supplements the VLMs to accurately generate video-level descriptions. ExtrAnom has been benchmarked against popular unimodal and multi-modal VAD datasets (e.g., XD-Violence, UCF-Crime, and UCA) and SOTA methods. Experiments reveal that the existing datasets are insufficient to train models for detecting women-centric anomalies.

UAV action recognition faces a deployment shift that standard benchmarks often obscure: a model trained on UAV footage captured from low-depression viewpoints may be required to recognize the same action classes from high-depression viewpoints. While the action labels remain unchanged, this shift alters body visibility, motion projection, and scene context, encouraging models to rely on viewpoint-specific shortcuts. We introduce UAV-OVO, an Out-of-Viewpoint generalization benchmark for UAV action recognition. UAV-OVO derives view scores from uncalibrated videos, uses a view-isolation band to assign low-depression videos to the training and in-distribution test splits while reserving high-depression videos for out-of-distribution testing, and constructs ID/OOD test sets matched by class distribution so that performance differences reflect viewpoint shift rather than label imbalance. Across representative video recognizers, UAV-OVO reveals a substantial ID/OOD gap: models that fit the low-depression training distribution well often fail to transfer to held-out high-depression views, exposing viewpoint shortcuts hidden by aggregate accuracy. We further propose LATER, LoRA-Anchored Test-time Re-centering, which first adapts the recognizer with Low-Rank Adaptation (LoRA) and then uses the learned LoRA subspace as a semantic anchor for online feature re-centering. Specifically, LATER projects target-domain displacement onto the orthogonal complement of the LoRA subspace before re-centering features, reducing viewpoint-induced drift while preserving task-relevant semantics. Together, UAV-OVO and LATER provide a controlled testbed and a practical adaptation method for viewpoint-robust UAV video understanding.

Track materialization converts raw video into reusable object tracks that downstream queries can run against without rerunning tracking, but extracting those tracks efficiently and with high fidelity remains expensive. Prior systems reduce cost through temporal frame sampling, erasing the inter-frame motion that fine-grained tracking requires. In stationary video, however, large portions of each frame contain no objects of interest, and the remaining regions tolerate different sampling rates. We present Tetris, a track-extraction system that decomposes videos into a tile-based polyomino data model, enabling fine-grained spatiotemporal pruning that reduces detector calls with minimal fidelity loss. Tetris runs three operators upstream of the user-provided detector: a classifier identifies relevant tiles and groups them into polyominoes, an integer linear program (ILP) prunes redundant polyominoes under a user-specified accuracy constraint, and a packer assembles the survivors into canvases that minimize detector calls. Across 7 stationary-video datasets, Tetris stays within a 5% tracking accuracy loss of a full-frame, every-frame reference pipeline, whereas prior systems exceed this bound on 3 of the 7 datasets. At this 5% bound, Tetris achieves up to 17.4x higher throughput than prior systems and up to 68.8x higher than the reference pipeline. The project page is at https://tetris-db.github.io .

Metaphorical videos are prevalent across various real-world scenarios to convey complex ideas, and understanding them typically requires high-order cognitive capabilities. The lack of systematic studies on metaphorical video understanding not only constrains the real-world applicability of MLLMs but also impedes the thorough assessment of their high-order cognitive capabilities. To bridge this gap, we propose MetaphorVU-Bench, the first systematic and comprehensive benchmark dedicated to metaphorical video understanding. Through experiments, we find current MLLMs struggle with accurate metaphorical video understanding, lagging far behind human level, primarily due to defective cross-domain mapping. Motivated by this finding, we construct a metaphor knowledge graph as mapping augmentation and propose MetaphorBoost, an inference-time enhancement framework achieving consistent performance improvement. Our benchmark, analysis, and method provide useful insights and a foundation for future research on advancing MLLMs.

As AI-powered compliance monitoring becomes increasingly important in public governance and industrial safety, the ability to provide verifiable evidence and traceable accountability signals is essential. However, existing video anomaly detection datasets focus on event-level binary classification, lacking the rule-driven, explainable analysis required for real-world compliance scenarios. We introduce FoodMonitor, a benchmark for explainable compliance analysis in commercial kitchen surveillance. FoodMonitor comprises 477 video clips with 3,307 violation annotations across a dual-channel design covering both person-level and environment-level violations. Each annotation specifies which rule was violated, what non-compliant behavior occurred, and who committed it with frame-level bounding boxes. We establish a unified evaluation protocol with a two-stage matching mechanism that separately assesses spatial localization and semantic understanding, along with a composite metric ($C_{\text{score}}$) that balances environment and person detection performance. Systematic evaluation of several state-of-the-art multimodal large language models reveals that the best-performing model achieves only 0.360 $C_{\text{score}}$, with spatial localization and fine-grained rule understanding emerging as the primary bottlenecks. Our analysis identifies two distinct failure modes: localization-dominated errors and semantics-dominated errors, providing diagnostic insights for future model development.

The EPIC-KITCHENS-100 Action Detection challenge evaluates whether a model can localize the start and end of each action in long untrimmed egocentric videos and assign the corresponding verb--noun action label. In this report, we formulate our submission as EgoAction (Egocentric Action Composition with Reliability-Aware Temporal Fusion), a unified decoupled detection and fusion pipeline. The pipeline uses EPIC-finetuned VideoMAE-L features, trains separate noun and verb temporal detectors with causal temporal modeling, composes action hypotheses from top noun--verb pairs, and introduces a confidence-adaptive boundary fusion rule at post-processing time. The key observation is that verb and noun streams often fail differently: verb scores are sensitive to motion transitions, whereas noun scores are sensitive to hand-object visibility and object clutter. A fixed arithmetic mean of their predicted boundaries can therefore amplify localization errors when one stream degenerates. We replace this hard-coded mean with Dynamic Weighted Fusion (DWF), which normalizes the maximum noun and verb classification confidences into proposal-wise boundary weights and linearly combines the two intervals. This lightweight tensor-only operator shifts boundary authority toward the more reliable stream while preserving the decoupled action scoring mechanism. Together with sliding-window inference, top-K noun--verb action composition, and class-wise Soft-NMS, EgoAction provides a compact and reproducible system for egocentric temporal action detection.

Data-driven learning approaches for physics simulation, sometimes referred to as world models, have emerged as promising alternatives to traditional physics simulators due to their differentiable nature. Prior work has demonstrated impressive results in predicting the motions of rigid and non-rigid objects in complex scenes involving multiple interacting bodies. However, these models are typically trained in simulated environments because obtaining perfect state information such as complete scene point clouds and point correspondences over time is challenging in real-world settings. This reliance on synthetic data can limit their applicability when the sim-to-real gap is large. In this work, we aim to overcome these limitations by introducing a novel framework for training neural object dynamics models directly from unlabeled real-world videos. Specifically, we propose to learn a particle-based dynamics model compatible with a Gaussian splatting framework, which operates on dense particles derived from Gaussians (i.e., particles with scales and rotations) and predicts their position and rotation changes over time. The model is trained via rendering supervision, enabling learning from real-world videos without requiring particle-level labeled states. Our model operates directly on dense Gaussians without relying on heuristic subsampling anchor points. To enable this study, we also present a real-world dataset consisting of about 500 videos capturing diverse object interactions.

Keyframe selection is a direct way to provide verifiable visual evidence for long-video question answering (QA). Queries differ in what they require, and finding the right frames depends on knowing what to look for. Existing keyframe selectors either score every frame against a single query, or decompose the query into a fixed schema evaluated by a single visual tool. We propose ToolMerge, a keyframe retrieval method based on decomposition and merging: an Large Language Model (LLM) based planner decomposes the query into tool calls and specifies how their per-tool rankings are merged using boolean operators. To evaluate retrieval directly, we construct Molmo-2 Moments (M2M), a benchmark in which every question is anchored to a specific time interval by construction. Across QA, question retrieval, and caption retrieval, ToolMerge is competitive with prior keyframe selectors, most notably on caption retrieval, outperforming other methods by 5%. Code and data can be found at https://github.com/michalsr/ToolMerge .

Imitation Learning from monocular video demonstrations provides a scalable approach for teaching complex skills to humanoid robots. However, translating human motion to humanoids requires overcoming significant morphological mismatches. Standard approaches rely on Geometric Retargeting or Indirect Dynamic Retargeting pipelines. We identify that these intermediate kinematic projections introduce a geometric bias, restricting the search space and yielding suboptimal dynamic behaviors. In this paper, we propose Direct Dynamic Retargeting (DDR), a novel single-stage framework that generates high-fidelity, dynamically feasible trajectories directly from expert videos. By formulating the problem in the task space and leveraging a sampling-based Model Predictive Control solver within a physics simulator, DDR natively optimizes over complex contact sequences while mitigating input drift. Our experiments demonstrate that bypassing the geometric bias allows DDR to outperform state-of-the-art baselines in demonstration tracking accuracy. Furthermore, we establish that providing such physically viable references to RL agents accelerates training convergence and enhances the final execution of agile and balancing behaviors. Source code will be made publicly available.

Multi-shot video generation requires maintaining a consistent appearance of recurring entities across shots while remaining faithful to shot-specific text prompts. Recent autoregressive methods reuse previously generated frames as memory. However, full-frame storage entangles persistent entity information with transient scene context, leading to irrelevant information leakage and high computational cost. We propose an entity-centric memory in the form of an entity-indexed bank of latent patches. We introduce sparse token conditioning compatible with pretrained models, restricting self-attention to entity-relevant tokens and reducing computational cost. To support this, we introduce a structured multi-shot script format. We additionally propose a budgeted memory update strategy to maintain a compact, evolving memory. Finally, we equip the entity representation with a noise-injection mechanism that enables fine-grained appearance control, preventing leakage of irrelevant information. Our method improves prompt adherence and efficiency while preserving subject consistency.

Temporal Action Localization (TAL) has been extensively studied in generic video understanding, while fine-grained sports scenarios, such as professional badminton, remain underexplored due to their complex and subtle spatio-temporal dynamics. In this paper, we focus on fine-grained TAL in professional badminton videos and introduce a new benchmark dataset, Fine-Badminton, which consists of 31 matches with 29 fine-grained stroke categories, covering 2104 rallies and 27597 annotated actions. To effectively capture the intricate motion patterns in such scenarios, we propose a Decoupling Spatio-Temporal Adapter (DSTA), which enables efficient modeling of spatio-temporal features within a parameter-efficient framework. Specifically, DSTA decomposes motion representation into three parallel branches, capturing temporal dynamics as well as vertical and horizontal spatial variations. The design allows the model to better distinguish subtle differences among fine-grained actions. Extensive experiments on both the Fine-Badminton dataset and the ShuttleSet benchmark demonstrate that the proposed method achieves state-of-the-art performance while introducing only a marginal increase in computational and parameter cost. These results validate the effectiveness and efficiency of the proposed approach for fine-grained temporal action localization.

Recent Open-Vocabulary Action Recognition (OVAR) methods typically aggregate visual features into a global representation before computing text alignment, a process that obscures local patch information and fine-grained spatio-temporal cues. We propose Similarity Volume Aggregation (SimVA), a framework that constructs a dense 4D spatio-temporal similarity volume from patch-level visual-text similarities. SimVA constructs a spatio-temporal similarity volume over local video tokens and action classes, and employs class sampling to ensure similarity aggregation scalable to large vocabularies. The similarity volume is refined by spatial aggregation, which contextualizes local similarity patterns to improve intra-frame consistency. Motion-aware modulation further injects inter-frame variation cues, highlighting dynamically changing regions. Mamba-based temporal aggregation then models the evolution of class-conditioned similarity patterns across frames. By maintaining dense visual-text correspondence, SimVA effectively transfers CLIP to video action recognition, achieving competitive performance across zero-shot, few-shot, and base-to-novel benchmarks.

Retrieving events from large-scale video datasets is challenging due to complex temporal, spatial, and multimodal information. This paper presents U-CESE, our solution for the AI Challenge HCMC 2025, a Unified Clip-based Event Search Engine for multimodal event retrieval across diverse video sources. Building on CESE, U-CESE integrates its three modules into a single cohesive framework, ensuring consistent processing and retrieval across query types. A core component is the Unified Clipping Algorithm, which merges separate clipping algorithms into one efficient pipeline. To handle large-scale data, we propose DAKE, a lightweight, training-free keyframe extraction method using JPEG file size variations to identify significant scene changes. Finally, we introduce ReCap, a temporally consistent captioning framework inspired by Recurrent Neural Network, generating detailed and context-aware textual descriptions. Experiments show that U-CESE delivers robust, consistent, and efficient performance in large-scale multimodal event retrieval.

Video recordings of child-caregiver interactions enable investigation of attentional dynamics during naturalistic behavior. Such multimodal recording also allows researchers to examine how attention interacts with action and language use in real time. However, manual annotation of such data is time-consuming. Here, we introduce GazeBehavior Annotation Toolkit, a deep-learning-based toolkit designed to facilitate three key processes in data preprocessing and feature extraction: post-hoc synchronization across multiple videos, semi-automatic annotation of gaze target categories, and categorization of participants' poses and hand actions. This toolkit improves the efficiency and scalability of feature extraction from human egocentric eye-tracking and video data. Such improvement is critical in supporting large-scale and longitudinal investigations of attentional dynamics and naturalistic behavior in human early development.

Real-world long video understanding requires models to perform continuous tracking, information integration and memory retention over massive temporal spans within extreme video durations. Mastering this intense cognitive load constitutes the fundamental bottleneck in long video understanding. While existing benchmarks have driven progress by scaling up video duration, their evaluation tasks often require comprehending only short and isolated video segments, falling short of capturing the challenge of ultra-long-context reasoning. To measure this cognitive load, we emphasize continuous certificate length, defined as the video length a human must continuously watch to definitively answer a given question. Driven by this metric, we introduce VideoOdyssey, a benchmark specifically designed for ultra-long-context and omni-modal video understanding. VideoOdyssey is characterized by three key features: 1) Extreme video duration and diversity: spanning 11 domains and 54 subcategories with an average video duration of 109 minutes; 2) Comprehensive evaluation scenarios: offering two subsets to address different research focuses, i.e., VideoOdyssey-V for probing the limits of visual understanding in MLLMs, and VideoOdyssey-AV for evaluating synchronized audio-visual understanding for omni-modal models; 3) Ultra-long and multi-level continuous certificates: extending the average continuous certificate to 16 minutes for VideoOdyssey-V and 12.8 minutes for VideoOdyssey-AV. Crucially, we design 5 granular levels from seconds to hours, providing a comprehensive diagnostic tool to evaluate models across varying context lengths and cognitive loads. Extensive evaluations show that bottlenecks of current MLLMs extend beyond simple retrieval to include struggles with continuous reasoning across varying context lengths, fine-grained perception, and non-verbal omni-modal understanding.

Video Large Language Models (Video-LLMs) have made rapid progress on temporal video understanding, yet many fail at a basic perceptual primitive: signed image-plane motion direction. On simple videos of a single object moving left, right, up, or down, most Video-LLMs perform near chance, with above-chance cases largely attributable to prediction biases rather than genuine direction understanding. We call this failure directional motion blindness. We localize the failure by tracing motion direction information through the Video-LLM pipeline. Motion direction remains linearly accessible from the vision encoder, projector, and LLM hidden states, but the readout fails to bind this signal to the correct verbal answer option, revealing a direction binding gap. Although synthetic motion direction instruction tuning reduces this gap on the source domain, motion direction concept vector analysis shows that visual complexity weakens the signal magnitude and limits out-of-domain generalization. We introduce MoDirect, a dataset family for motion direction instruction tuning and evaluation, and DeltaDirect, a diagnosis-driven, projector-level objective that predicts normalized 2-D motion vectors from adjacent-frame feature deltas. On MoDirect-SynBench, instruction tuning with DeltaDirect improves motion direction accuracy from 25.9% to 85.4%. On MoDirect-RealBench, DeltaDirect improves real-world motion direction accuracy by 21.9 points over the vanilla baseline without real-world tuning data, while preserving standard video-understanding performance. Code: https://github.com/KHU-VLL/DeltaDirect

Camera pose matters. The position and orientation of each viewpoint define a shared spatial coordinate frame that relates observations across video frames. Yet this signal is largely absent from multimodal LLMs (MLLMs) for video understanding, which process frames as isolated 2D snapshots, instead of the persistent scene humans perceive. We revisit pose as a lightweight supervisory signal and introduce Cambrian-P, a video MLLM augmented with per-frame learnable camera tokens and a pose regression head. With a carefully designed sampling scheme, the model achieves substantial gains of 4.5-6.5% on spatial reasoning benchmarks such as VSI-Bench, generalizes across eight additional spatial and general video QA benchmarks, and, as a byproduct, achieves state of the art streaming pose estimation on ScanNet. Surprisingly, training on pseudo-annotated poses from in-the-wild video further improves general video QA benchmarks, showing pose helps beyond spatial reasoning. Together, these results position camera pose as a fundamental signal for video models that reason about the physical world.

Two-wheelers account for a disproportionately high share of road fatalities in the Global South. Research on two-wheeler rider behavior, however, lags far behind four-wheelers, where multimodal datasets have driven major advances in Advanced Driver Assistance Systems (ADAS). To address this gap, we present the MOtorized TwO-wheeler Rider (MOTOR) dataset, the first large-scale, multi-view, multimodal resource dedicated to two-wheelers in dense, unstructured traffic. MOTOR comprises 1,629 sequences (25+ hours of video data) collected from 16 riders and integrates synchronized front, rear, and helmet videos, rider eye-gaze from wearable trackers, on-road audio, and telemetry (GPS, accelerometer, gyroscope). Rich annotations capture traffic context, rider state, 12 riding maneuvers spanning conventional and unconventional behaviors, and legality labels (Legal, Illegal, Unspecified). We benchmark rider behavior recognition and maneuver legality classification using state-of-the-art video action recognition backbones (CNN and Transformer-based), extended with multimodal fusion, and find that combining RGB, gaze, and telemetry consistently yields the best performance. MOTOR thus provides a unique foundation for advancing safety-critical understanding of two-wheeler riding. It offers the research community a benchmark to develop and evaluate models for behavior analysis, legality-aware prediction, and intelligent transportation systems. Dataset and code is available at https: //varuniiith.github.io/MOTOR-Dataset/

Traditional visual object tracking (VOT) methods typically rely on task-specific supervised training, limiting their generalization to unseen objects and challenging scenarios with distractors, occlusion, and nonlinear motion. Recent vision foundation models, exemplified by SAM 2, learn strong video understanding priors from large-scale pretraining and offer a promising foundation for building more robust and generalizable trackers. However, directly applying SAM 2 to VOT remains suboptimal, as it does not explicitly model target motion dynamics or enforce geometric and semantic consistency across frames, both of which are essential for reliable tracking. To address this issue, we propose SAMOSA, a new tracking framework that adapts SAM 2 to complex VOT scenarios by explicitly leveraging motion, geometry, and semantic cues. Specifically, we introduce a lightweight nonlinear motion predictor to model target dynamics and guide mask selection as well as memory filtering. We further exploit semantic cues to detect target shifts and recover from tracking failures, while geometric cues are incorporated as structural constraints to improve tracking stability. In this way, SAMOSA bridges the gap between the implicit video understanding prior of SAM 2 and explicit tracking-oriented modeling. Extensive experiments show that SAMOSA consistently outperforms state-of-the-art SAM 2--based approaches on general benchmarks, demonstrates stronger generalization than supervised VOT methods, and achieves substantial gains on anti-UAV datasets, which typify complex nonlinear motion scenarios. Our code is available at https://github.com/DurYi/SAMOSA.

Long streaming video QA remains challenging due to growing visual tokens and limited reasoning length of large language models (LLMs). KV-caching stores the Key-Value (KV) of the historical tokens via LLM prefill and enables more efficient streaming QA. However, existing methods cache every one or two frames, causing redundant memory usage and losing fine-grained spatial details within frame or temporal contexts across frames. This paper proposes MuKV, a method that features a multi-grained KV cache compression module and a semi-hierarchical retrieval approach to improve both efficiency and accuracy for long streaming VideoQA. For the offline KV cache, MuKV extracts visual representations at patch-, frame-, and segment-levels. The multiple levels of granularity preserve both local cues and global temporal context, while maintaining efficiency with a dual signal token compression mechanism guided by self-attention and frequency. For online QA, MuKV designs a semi-hierarchical retrieval method to retrieve relevant KV caches for answer generation. Experiments on long-streaming VideoQA benchmarks show that MuKV significantly improves answer accuracy, without sacrificing memory and online QA efficiency. Moreover, our compression mechanism alone brings consistent benefits across answer accuracy, memory, and QA efficiency over baselines, showcasing highly effective contribution.

Multimodal Large Language Models (MLLMs) face significant computational overhead when processing long videos due to the massive number of visual tokens required. To improve efficiency, existing methods primarily reduce redundancy by pruning or merging tokens based on importance or similarity. However, these approaches largely overlook a critical dimension of video content, i.e., changes and turning points, and they lack a collaborative model for spatio-temporal relationships. To address this, we propose a new perspective: similarity is for identifying redundancy, while difference is for capturing key events. Based on this, we designed a training-free framework named ST-SimDiff. We first construct a spatio-temporal graph from the visual tokens to uniformly model their complex associations. Subsequently, we employ a parallel dual-selection strategy: 1) similarity-based selection uses community detection to retain representative tokens, compressing static information; 2) temporal difference-based selection precisely locates content-changing points to preserve tokens that capture key dynamic shifts. This allows it to preserve both static and dynamic content with a minimal number of tokens. Extensive experiments show our method significantly outperforms state-of-the-art approaches while substantially reducing computational costs. Our code is available in https://github.com/bingjunluo/ST-SimDiff.

Video large language models (Video LLMs) achieve strong benchmark accuracy, yet often answer video questions through shortcuts such as single-frame cues and language priors rather than by tracking spatiotemporal dynamics. This issue is exacerbated in RL post-training, where correctness-only rewards can further reinforce shortcut policies that obtain high reward without tracking video dynamics. We address this by asking a controlled counterfactual question: if the visual world changed while the question remained fixed, should the answer change or stay the same? Based on this view, we propose \textbf{Counterfactual Relational Policy Optimization (CRPO)}, a dual-branch RL framework for improving \emph{spatiotemporal sensitivity}. CRPO constructs counterfactual videos through horizontal flips and temporal reversals, trains on both original and counterfactual branches, and introduces a \textbf{Counterfactual Relation Reward (CRR)} between their answers. CRR encourages answers to change for dynamic questions and remain unchanged for static questions. This cross-branch constraint makes it difficult for shortcut policies to be consistently rewarded across both branches. To evaluate this property, we introduce \textbf{DyBench}, a paired counterfactual video benchmark with 3,014 videos covering reversible dynamics, moving direction, and event sequence, together with a strict pair-accuracy metric that prevents fixed-answer shortcuts from inflating scores. Experiments show that CRPO outperforms prior RL methods on spatiotemporal-sensitive evaluations while maintaining competitive general video performance. On Qwen3-VL-8B, CRPO improves DyBench P-Acc by +7.7 and TimeBlind I-Acc by +8.2 over the base model, indicating improved spatiotemporal sensitivity rather than stronger reliance on static shortcuts. The project website can be found at https://ddz16.github.io/crpo.github.io/ .