In the context of long-term video understanding with large multimodal models, many frameworks have been proposed. Although transformer-based visual compressors and memory-augmented approaches are often used to process long videos, they usually compress each frame independently and therefore fail to achieve strong performance on tasks that require understanding complete events, such as temporal ordering tasks in MLVU and VNBench. This motivates us to rethink the conventional one-way scheme from perception to memory, and instead establish a feedbackdriven process in which past visual contexts stored in the context memory can benefit ongoing perception. To this end, we propose Question-guided Visual Compression with Memory Feedback (QViC-MF), a framework for long-term video understanding. At its core is a Question-guided Multimodal Selective Attention (QMSA), which learns to preserve visual information related to the given question from both the current clip and the past related frames from the memory. The compressor and memory feedback work iteratively for each clip of the entire video. This simple yet effective design yields large performance gains on longterm video understanding tasks. Extensive experiments show that our method achieves significant improvement over current state-of-the-art methods by 6.1% on MLVU test, 8.3% on LVBench, 18.3% on VNBench Long, and 3.7% on VideoMME Long. The code will be available at https://github.com/FujitsuResearch/QViC-MF/.
论文检索
输入标题、作者或关键词，从 2,006 篇学术成果中精准定位
REVISOR: Beyond Textual Reflection, Towards Multimodal Introspective Reasoning in Long-Form Video Understanding
PDF ↗Self-reflection mechanisms that rely on purely text-based rethinking processes perform well in most multimodal tasks. However, when directly applied to long-form video understanding scenarios, they exhibit clear limitations. The fundamental reasons for this lie in two points: (1) long-form video understanding involves richer and more dynamic visual input, meaning rethinking only the text information is insufficient and necessitates a further rethinking process specifically targeting visual information; (2) purely text-based reflection mechanisms lack cross-modal interaction capabilities, preventing them from fully integrating visual information during reflection. Motivated by these insights, we propose REVISOR (REflective VIsual Segment Oriented Reasoning), a novel framework for tool-augmented multimodal reflection. REVISOR enables MLLMs to collaboratively construct introspective reflection processes across textual and visual modalities, significantly enhancing their reasoning capability for long-form video understanding. To ensure that REVISOR can learn to accurately review video segments highly relevant to the question during reinforcement learning, we designed the Dual Attribution Decoupled Reward (DADR) mechanism. Integrated into the GRPO training strategy, this mechanism enforces causal alignment between the model's reasoning and the selected video evidence. Notably, the REVISOR framework significantly enhances long-form video understanding capability of MLLMs without requiring supplementary supervised fine-tuning or external models, achieving impressive results on four benchmarks including VideoMME, LongVideoBench, MLVU, and LVBench.
Images and videos are discrete 2D projections of the 4D world (3D space + time). Most visual understanding, prediction, and generation operate directly on 2D observations, leading to suboptimal performance. We propose SeeU, a novel approach that learns the continuous 4D dynamics and generate the unseen visual contents. The principle behind SeeU is a new 2D\to4D\to2D learning framework. SeeU first reconstructs the 4D world from sparse and monocular 2D frames (2D\to4D). It then learns the continuous 4D dynamics on a low-rank representation and physical constraints (discrete 4D\tocontinuous 4D). Finally, SeeU rolls the world forward in time, re-projects it back to 2D at sampled times and viewpoints, and generates unseen regions based on spatial-temporal context awareness (4D\to2D). By modeling dynamics in 4D, SeeU achieves continuous and physically-consistent novel visual generation, demonstrating strong potentials in multiple tasks including unseen temporal generation, unseen spatial generation, and video editing. All data and code will be public at \href https://yuyuanspace.com/SeeU/ here .
Understanding physical transformation processes is crucial for both human cognition and artificial intelligence systems, particularly from an egocentric perspective, which serves as a key bridge between humans and machines in action modeling. We define this modeling process as Egocentric Instructed Visual State Transition (EIVST), which involves generating intermediate frames that depict object transformations between initial and target states under a brief action instruction. EIVST poses two challenges for current generative models: (1) understanding the visual scenes of the initial and target states and reasoning about transformation steps from an egocentric view, and (2) generating a consistent intermediate transition that follows the given instruction while preserving object appearance across the two visual states. To address these challenges, we propose the EgoIn framework. It first infers the multi-step transition process between two given states using TransitionVLM, fine-tuned on our curated dataset to better adapt to this task and reduce hallucinated information. It then generates a sequence of frames based on transition conditions produced by the proposed Transition Conditioning module. Additionally, we introduce Object-aware Auxiliary Supervision to preserve consistent object appearance throughout the transition. Extensive experiments on human-object and robot-object interaction datasets demonstrate EgoIn's superior performance in generating semantically meaningful and visually coherent transformation sequences.
Video panoptic segmentation (VPS) aims to jointly detect, segment, and track all objects while partitioning the video into semantically consistent regions. We introduce the task setting of unsupervised VPS, omitting any human supervision. Existing unsupervised scene understanding works mainly focused on image segmentation tasks; the video domain remains underexplored. We propose VideoCUPS, the first unsupervised VPS approach. VideoCUPS generates temporally consistent panoptic video pseudo-labels from scene-centric videos by exploiting unsupervised depth, motion, and visual cues. Training on these pseudo-labels using a novel Video DropLoss yields an accurate, unsupervised VPS model. To benchmark progress, we introduce a comprehensive evaluation protocol and four competitive baselines, extending state-of-the-art unsupervised panoptic image and instance video segmentation models to VPS. VideoCUPS outperforms all baselines and demonstrates strong label-efficient learning. With VideoCUPS, our evaluation protocol, and baselines, we provide a strong foundation for future research on unsupervised VPS.
As large language models (LLMs) continue to advance, there is increasing interest in their ability to infer human mental states and demonstrate a human-like Theory of Mind (ToM). Most existing ToM evaluations, however, are centered on text-based inputs, while scenarios relying solely on visual information receive far less attention. This leaves a gap, since real-world human-AI interaction typically requires multimodal understanding. In addition, many current methods regard the model as a black box and rarely probe how its internal attention behaves in multiple-choice question answering (QA). The impact of LLM hallucinations on such tasks is also underexplored from an interpretability perspective. To address these issues, we introduce VisionToM, a vision-oriented intervention framework designed to strengthen task-aware reasoning. The core idea is to compute intervention vectors that align visual representations with the correct semantic targets, thereby steering the model's attention through different layers of visual features. This guidance reduces the model's reliance on spurious linguistic priors, leading to more reliable multimodal language model (MLLM) outputs and better QA performance. Experiments on the EgoToM benchmark--an egocentric, real-world video dataset for ToM with three multiple-choice QA settings--demonstrate that our method substantially improves the ToM abilities of MLLMs. Furthermore, results on an additional open-ended generation task show that VisionToM enables MLLMs to produce free-form explanations that more accurately capture agents' mental states, pushing machine-human collaboration toward greater alignment.
Understanding hand-object interaction (HOI) is fundamental to computer vision, robotics, and AR/VR. However, conventional hand videos often lack essential physical information, such as contact forces and motion dynamics, and are prone to frequent occlusions. To address these challenges, we present Glove2Hand, a framework that translates multi-modal sensing glove data in HOI videos into photorealistic bare-hand representations, while faithfully preserving the underlying physical interaction dynamics. We introduce a novel 3D Gaussian hand model that ensures both temporal and multi-view rendering consistency. The rendered hand is seamlessly integrated into the scene using a diffusion-based hand restorer, which effectively handles complex hand-object interactions and non-rigid deformations. Leveraging Glove2Hand, we introduce HandSense, the first multi-modal HOI dataset featuring multi-view bare-hand videos with synchronized tactile and IMU signals. We demonstrate that HandSense significantly enhances downstream bare-hand applications, including video-based contact estimation and hand tracking under severe occlusion.
Multimodal large language models have demonstrated remarkable capabilities in 2D vision, motivating their extension to 3D scene understanding. Recent studies represent 3D scenes as 3D spatial videos composed of image sequences with depth and camera pose information, enabling pre-trained video-language models to perform 3D reasoning tasks. However, the large number of visual tokens in spatial videos remains a major bottleneck for efficient inference. Existing pruning methods overlook the view consistency of spatial videos and the spatial diversity of the remaining tokens, which prevents them from effectively removing inter-frame redundancy and preserving scene completeness. In this paper, we propose Geo3DPruner, a Geometry-Guided 3D Visual Token Pruning framework. Geo3DPruner first models cross-frame relevance through geometry-aware global attention, and then performs a two-stage pruning process. The intra-voxel stage selects representative multi-view features within each voxel, while the inter-voxel stage preserves spatial diversity by selecting a globally distributed subset of voxels. Extensive experiments on multiple 3D scene understanding benchmarks demonstrate that Geo3DPruner retains over 90% of the original performance while pruning 90% of visual tokens, significantly outperforming existing text-guided and vision-guided pruning methods.
We present AToken, the first unified visual tokenizer that achieves both high-fidelity reconstruction and semantic understanding across images, videos, and 3D assets. Unlike existing tokenizers that specialize in either reconstruction or understanding for single modalities, AToken encodes these diverse visual inputs into a shared 4D latent space, unifying both tasks and modalities in a single framework. Specifically, we introduce a pure transformer architecture with 4D rotary position embeddings to process visual inputs of arbitrary resolutions and temporal durations. To ensure stable training, we introduce an adversarial-free training objective that combines perceptual and Gram matrix losses, achieving state-of-the-art reconstruction quality. By employing a progressive training curriculum, AToken gradually expands from single images, videos, and 3D, and supports both continuous and discrete latent tokens. AToken achieves 0.21 rFID with 82.2% ImageNet accuracy for images, 3.01 rFVD with 40.2% MSRVTT retrieval for videos, and 28.28 PSNR with 90.9% classification accuracy for 3D.. In downstream applications, AToken enables both visual generation tasks (e.g., image generation with continuous and discrete tokens, text-to-video generation, image-to-3D synthesis) and understanding tasks (e.g., multimodal LLMs), achieving competitive performance across all benchmarks. These results shed light on the next-generation multimodal AI systems built upon unified visual tokenization.
Recent progress in self- and weakly supervised occupancy estimation has largely relied on 2D projection or rendering-based supervision, which suffers from geometric inconsistencies and severe depth bleeding.We thus introduce ShelfOcc, a vision-only method that overcomes these limitations without relying on LiDAR.ShelfOcc brings supervision into native 3D space by generating metrically consistent semantic voxel labels from video, enabling true 3D supervision without any additional sensors or manual 3D annotations.While recent vision-based 3D geometry foundation models provide a promising source of prior knowledge, they do not work out of the box as a prediction due to sparse or noisy and inconsistent geometry, especially in dynamic driving scenes.Our method introduces a dedicated framework that mitigates these issues by filtering and accumulating static geometry consistently across frames, handling dynamic content and propagating semantic information into a stable voxel representation.This data-centric shift in supervision for weakly/shelf-supervised occupancy estimation allows the use of essentially any SOTA occupancy model architecture without relying on LiDAR data.We argue that such high-quality supervision is essential for robust occupancy learning and constitutes an important complementary avenue to architectural innovation.On the Occ3D-nuScenes benchmark, ShelfOcc substantially outperforms all previous weakly/shelf-supervised methods (up to a 34% relative improvement), establishing a new data-driven direction for LiDAR-free 3D scene understanding.
Recurrent Reasoning with Vision-Language Models for Estimating Long-Horizon Embodied Task Progress
PDF ↗Accurately estimating task progress is critical for embodied agents to plan and execute long-horizon, multi-step tasks. Despite promising advances, existing Vision-Language Models (VLMs) based methods primarily leverage their video understanding capabilities, while neglecting their complex reasoning potential. Furthermore, processing long video trajectories with VLMs is computationally prohibitive for real-world deployment. To address these challenges, we propose the Recurrent Reasoning Vision-Language Model (\text R ^2VLM). Our model features a recurrent reasoning framework that processes local video snippets iteratively, maintaining a global context through an evolving Chain of Thought (CoT). This CoT explicitly records task decomposition, key steps, and their completion status, enabling the model to reason about complex temporal dependencies. This design avoids the high cost of processing long videos while preserving essential reasoning capabilities. We train \text R ^2VLM on large-scale, automatically generated datasets from ALFRED and Ego4D. Extensive experiments on progress estimation and downstream applications, including progress-enhanced policy learning, reward modeling for reinforcement learning, and proactive assistance, demonstrate that \text R ^2VLM achieves strong performance and generalization, achieving a new state-of-the-art in long-horizon task progress estimation. The models and benchmarks are publicly available at https://huggingface.co/collections/zhangyuelin/r2vlm.
Taking advantage of large-scale data and pretrained language models, Video Large Language Models (Video-LLMs) have shown strong capabilities in answering video questions. However, most existing efforts focus on improving performance, with limited attention to understanding their internal mechanisms. This paper aims to bridge this gap through a systematic empirical study. To interpret existing VideoLLMs, we adopt attention knockouts as our primary analytical tool and design three variants: Video Temporal Knockout, Video Spatial Knockout, and Language-to-Video Knockout. Then, we apply these three knockouts on different numbers of layers (window of layers). By carefully controlling the window of layers and types of knockouts, we provide two settings: a global setting and a fine-grained setting. Our study reveals three key findings: (1) Global setting indicates Video information extraction primarily occurs in early layers, forming a clear two-stage process--lower layers focus on perceptual encoding, while higher layers handle abstract reasoning; (2) In the fine-grained setting, certain intermediate layers exert an outsized impact on video question answering, acting as critical outliers, whereas most other layers contribute minimally; (3) In both settings, we observe that spatial-temporal modeling relies more on language-guided retrieval than on intra- and inter-frame self-attention among video tokens, despite the latter's high computational cost. Finally, we demonstrate that these insights can be leveraged to reduce attention computation in Video-LLMs. To our knowledge, this is the first work to systematically uncover how Video-LLMs internally process and understand video content, offering interpretability and efficiency perspectives for future research.
Frame selectoin is crucial due to high frame redundancy and limited context windows when applying Large Vision-Language Models (LVLMs) to long videos. Current methods typically select frames with high relevance to a given query, resulting a disjointed set of frames that disregard the narrative structure of video. In this paper, we introduce Wavelet-based Frame Selection by Detecting Semantic Boundary (WFS-SB), a training-free framework that presents a new perspective: effective video understanding hinges not only on high relevance but, more importantly, on capturing semantic shifts--pivotal moments of narrative change that are essential to comprehending the holistic storyline of video. However, a direct detection of abrupt changes in the query-frame similarity signal is often unreliable due to high-frequency noise arising from model uncertainty and transient visual variations.To address this, we leverage the wavelet transform, which provides an ideal solution through its multi-resolution analysis in both time and frequency domains. By applying this transform, we decompose the noisy signal into multiple scales and extract a clean semantic change signal from the coarsest scale. We identify the local extrema of this signal as semantic boundaries, which segment the video into coherent clips. Building on this, WFS-SB comprises a two-stage strategy: first, adaptively allocating a frame budget to each clip based on a composite importance score; and second, within each clip, employing the Maximal Marginal Relevance approach to select a diverse yet relevant set of frames. Extensive experiments show that WFS-SB significantly boosts LVLM performance, e.g., improving accuracy by 5.5% on VideoMME, 9.5% on MLVU, and 6.2% on LongVideoBench, consistently outperforming state-of-the-art methods. Our code is available at https://github.com/MAC-AutoML/WFS-SB.
Thinking With Videos: Multimodal Tool-Augmented Reinforcement Learning for Long Video Reasoning
PDF ↗The video reasoning ability of multimodal large language models (MLLMs) is crucial for downstream tasks like video question answering and temporal grounding. While recent approaches have explored text-based chain-of-thought (CoT) reasoning for MLLMs, these methods often suffer from limited cross-modal interaction and increased hallucination, especially with longer videos or reasoning chains. To address these challenges, we propose Video Intelligence via Tool-Augmented Learning (VITAL), a novel end-to-end agentic video reasoning framework. With a visual toolbox, the model can densely sample new video frames on demand and generate multimodal CoT for precise long video reasoning. We observe that temporal grounding and question answering are mutually beneficial for video understanding tasks. Therefore, we construct two high-quality multi-task video reasoning datasets MTVR-CoT-72k for supervised fine-tuning and MTVR-RL-110k for reinforcement learning. Moreover, we propose a Difficulty-aware Group Relative Policy Optimization algorithm (DGRPO) to mitigate difficulty imbalance in multi-task reinforcement learning. Extensive experiments on eleven challenging video understanding benchmarks demonstrate the advanced reasoning ability of VITAL, outperforming existing methods in video question answering and temporal grounding tasks, especially in long video scenarios. Code is available at https://github.com/zhang9302002/ThinkingWithVideos.
Streaming Video Large Language Models (VideoLLMs) have demonstrated impressive performance across various video understanding tasks, but they face significant challenges in real-time deployment due to the high computational cost of processing dense visual tokens from continuous video streams. In streaming video scenarios, the primary bottleneck lies in the Vision Transformer (ViT) encoding stage, where redundant processing of temporally similar frames leads to inefficiency. Additionally, inflated token sequences during LLM pre-filling further exacerbate latency and memory overhead. To address these challenges, we propose Streaming Token Compression (STC), a plug-and-play hierarchical framework that seamlessly integrates into existing streaming VideoLLMs, optimizing both ViT encoding and LLM pre-filling stages to accelerate processing. STC introduces two token-level accelerators: STC-Cacher, which reduces ViT encoding overhead by caching and reusing features from temporally similar frames, and STC-Pruner, which compresses the visual token sequence before it enters the LLM, preserving only the most salient tokens based on both spatial and temporal relevance. Extensive experiments on four baseline streaming VideoLLMs across five benchmarks demonstrate that STC outperforms other compression methods. Notably, STC retains up to 99% of accuracy on the ReKV framework while reducing ViT encoding latency and LLM pre-filling latency by 24.5% and 45.3%.
The new era has witnessed a remarkable capability to extend Vision-Language Models (VLMs) for tackling tasks of video understanding. While current VLMs excel at event- or story-level understanding, their ability to capture fine-grained motion details remains limited, primarily due to their focus on high-level static semantic structures and macro-event logic. In contrast, Video Diffusion Models (VDMs) are adept at modeling dynamic motion patterns, benefiting from large-scale video data and the intrinsic requirement of temporal generation. In this paper, we introduce MotionEnhancer, a novel approach that leverages motion priors distilled from a powerful video diffusion model as auxiliary supervision to enhance the motion understanding capability of a VLM via attention alignment. MotionEnhancer comprises two simple parameter-free modules, Motion-sensitive Head Selection (MHS) and Motion-salient Text Token Identification (MTTI), to directly extract and optimize motion-related attentions from the VDM in a computation-only manner. MotionEnhancer provides a scalable solution for motion understanding without additional training parameters, modifications to existing architectures, or tool calling. Extensive experiments demonstrate that MotionEnhancer can achieve consistent improvements over state-of-the-art VLMs on two motion-level video understanding benchmarks, especially on motion-related metrics.
Video Temporal Grounding (VTG) aims to localize the video segment that corresponds to a natural language query, which requires a comprehensive understanding of complex temporal dynamics. Existing Vision-LMMs typically perceive temporal dynamics via positional encoding, text-based timestamps, or visual frame numbering. However, these approaches exhibit notable limitations: assigning each frame a text-based timestamp token introduces additional computational overhead and leads to sparsity in visual attention, positional encoding struggles to capture absolute temporal information, and visual frame numbering often compromises spatial detail. To address these issues, we propose Temporal to Spatial Gridification (T2SGrid), a novel framework that reformulates video temporal understanding as a spatial understanding task. The core idea of T2SGrid is to process video content in clips rather than individual frames. we employ a overlapping sliding windows mechanism to segment the video into temporal clips. Within each window, frames are arranged chronologically in a row-major order into a composite grid image, effectively transforming temporal sequences into structured 2D layouts. The gridification not only encodes temporal information but also enhances local attention within each grid. Furthermore, T2SGrid enables the use of composite text timestamps to establish global temporal awareness. Experiments on standard VTG benchmarks demonstrate that T2SGrid achieves superior performance.
Long video understanding is essential for human-like intelligence, enabling coherent perception and reasoning over extended temporal contexts. While the emerging thinking-with-frames paradigm--which alternates between global temporal reasoning and local frame examination--has advanced the reasoning capabilities of video multi-modal large language models (MLLMs), it suffers from a significant efficiency bottleneck due to the progressively growing and redundant multi-modal context. To address this, we propose SpecTemp, a reinforcement learning-based Speculative Temporal reasoning framework that decouples temporal perception from reasoning via a cooperative dual-model design. In SpecTemp, a lightweight draft MLLM rapidly explores and proposes salient frames from densely sampled temporal regions, while a powerful target MLLM focuses on temporal reasoning and verifies the draft's proposals, iteratively refining its attention until convergence. This design mirrors the collaborative pathways of the human brain, balancing efficiency with accuracy. To support training, we construct the SpecTemp-80K dataset, featuring synchronized dual-level annotations for coarse evidence spans and fine-grained frame-level evidence. Experiments across multiple video understanding benchmarks demonstrate that SpecTemp not only maintains competitive accuracy but also significantly accelerates inference compared with existing thinking-with-frames methods.
Multimodal Summarization (MMS) aims to generate concise textual summaries by understanding and integrating information across videos, transcripts, and images. However, existing approaches still suffer from three main challenges: (1) reliance on domain-specific supervision, (2) implicit fusion with weak cross-modal grounding, and (3) flat temporal modeling without event transitions. To address these issues, we introduce **CoE**, a training-free MMS framework that performs structured reasoning through a **Chain-of-Events** guided by a Hierarchical Event Graph (HEG). The HEG encodes textual semantics into an explicit event hierarchy that scaffolds cross-modal grounding and temporal reasoning. Guided by this structure, **CoE** localizes key visual cues, models event evolution and causal transitions, and refines outputs via lightweight style adaptation for domain alignment. Extensive experiments on eight diverse datasets demonstrate that **CoE** consistently outperforms state-of-the-art video CoT baselines, achieving average gains of **+3.04 ROUGE**, **+9.51 CIDEr**, and **+1.88 BERTScore**, highlighting its robustness, interpretability, and cross-domain generalization. Our code is available at https://github.com/youxiaoxing/CoE.
Video Temporal Grounding (VTG), the task of localizing video segments from text queries, struggles in open-world settings due to limited dataset scale and semantic diversity, causing performance gaps between common and rare concepts. To overcome these limitations, we introduce OmniVTG, a new large-scale dataset for open-world VTG, coupled with a Self-Correction Chain-of-Thought (CoT) training paradigm designed to enhance the grounding capabilities of Multimodal Large Language Models (MLLMs). Our OmniVTG is constructed via a novel Semantic Coverage Iterative Expansion pipeline, which first identifies gaps in the vocabulary of existing datasets and collects videos that are highly likely to contain these target concepts. For high-quality annotation, we leverage the insight that modern MLLMs excel at dense captioning more than direct grounding and design a caption-centric data engine to prompt MLLMs to generate dense, timestamped descriptions. Beyond the dataset, we observe that simple supervised finetuning (SFT) is insufficient, as a performance gap between rare and common concepts still persists. We find that MLLMs' video understanding ability significantly surpasses their direct grounding ability. Based on this, we propose a Self-Correction Chain-of-Thought (CoT) training paradigm. We train the MLLM to first predict, then use its understanding capabilities to reflect on and refine its own predictions. This capability is instilled via a three-stage pipeline of SFT, CoT finetuning, and reinforcement learning. Extensive experiments show our approach not only excels at open-world grounding in our OmniVTG dataset but also achieves state-of-the-art zero-shot performance on four existing VTG benchmarks. Code is available at https://github.com/oceanflowlab/OmniVTG.