Current video-language models struggle with long-video understanding due to limited context lengths and reliance on sparse frame subsampling, often leading to information loss. This paper introduces $\infty$-Video, which can process arbitrarily long videos through a continuous-time long-term memory (LTM) consolidation mechanism. Our framework augments video Q-formers by allowing them to process unbounded video contexts efficiently and without requiring additional training. Through continuous attention, our approach dynamically allocates higher granularity to the most relevant video segments, forming "sticky" memories that evolve over time. Experiments with Video-LLaMA and VideoChat2 demonstrate improved performance in video question-answering tasks, showcasing the potential of continuous-time LTM mechanisms to enable scalable and training-free comprehension of long videos.
Applying Multimodal Large Language Models (MLLMs) to video understanding presents significant challenges due to the need to model temporal relations across frames. Existing approaches adopt either implicit temporal modeling, relying solely on the LLM decoder, or explicit temporal modeling, employing auxiliary temporal encoders. To investigate this debate between the two paradigms, we propose the Stackable Temporal Encoder (STE). STE enables flexible explicit temporal modeling with adjustable temporal receptive fields and token compression ratios. Using STE, we systematically compare implicit and explicit temporal modeling across dimensions such as overall performance, token compression effectiveness, and temporal-specific understanding. We also explore STE's design considerations and broader impacts as a plug-in module and in image modalities. Our findings emphasize the critical role of explicit temporal modeling, providing actionable insights to advance video MLLMs.
In human-centric scenes, the ability to simultaneously understand visual and auditory information is crucial. While recent omni models can process multiple modalities, they generally lack effectiveness in human-centric scenes due to the absence of large-scale, specialized datasets and non-targeted architectures. In this work, we developed HumanOmni, the industry's first human-centric Omni-multimodal large language model. We constructed a dataset containing over 2.4 million human-centric video clips with detailed captions and more than 14 million instructions, facilitating the understanding of diverse human-centric scenes. HumanOmni includes three specialized branches for understanding different types of scenes. It adaptively fuses features from these branches based on user instructions, significantly enhancing visual understanding in scenes centered around individuals. Moreover, HumanOmni integrates audio features to ensure a comprehensive understanding of environments and individuals. Our experiments validate HumanOmni's advanced capabilities in handling human-centric scenes across a variety of tasks, including emotion recognition, facial expression description, and action understanding. Our model will be open-sourced to facilitate further development and collaboration within both academia and industry.
The analysis of extended video content poses unique challenges in artificial intelligence, particularly when dealing with the complexity of tracking and understanding visual elements across time. Current methodologies that process video frames sequentially struggle to maintain coherent tracking of objects, especially when these objects temporarily vanish and later reappear in the footage. A critical limitation of these approaches is their inability to effectively identify crucial moments in the video, largely due to their limited grasp of temporal relationships. To overcome these obstacles, we present GraphVideoAgent, a cutting-edge system that leverages the power of graph-based object tracking in conjunction with large language model capabilities. At its core, our framework employs a dynamic graph structure that maps and monitors the evolving relationships between visual entities throughout the video sequence. This innovative approach enables more nuanced understanding of how objects interact and transform over time, facilitating improved frame selection through comprehensive contextual awareness. Our approach demonstrates remarkable effectiveness when tested against industry benchmarks. In evaluations on the EgoSchema dataset, GraphVideoAgent achieved a 2.2 improvement over existing methods while requiring analysis of only 8.2 frames on average. Similarly, testing on the NExT-QA benchmark yielded a 2.0 performance increase with an average frame requirement of 8.1. These results underscore the efficiency of our graph-guided methodology in enhancing both accuracy and computational performance in long-form video understanding tasks.
We present the TinyLLaVA-Video, a video understanding model with parameters not exceeding 4B that processes video sequences in a simple manner, without the need for complex architectures, supporting both fps sampling and uniform frame sampling. Our model is characterized by modularity and scalability, allowing training and inference with limited computational resources and enabling users to replace components based on their needs. We validate the effectiveness of this framework through experiments, the best model achieving performance comparable to certain existing 7B models on multiple video understanding benchmarks. The code and training recipes are fully open source, with all components and training data publicly available. We hope this work can serve as a baseline for practitioners exploring small-scale multimodal models for video understanding. It is available at \url{https://github.com/ZhangXJ199/TinyLLaVA-Video}.
Despite significant advancements in video large multimodal models (video-LMMs), achieving effective temporal grounding in long-form videos remains a challenge for existing models. To address this limitation, we propose Temporal Preference Optimization (TPO), a novel post-training framework designed to enhance the temporal grounding capabilities of video-LMMs through preference learning. TPO adopts a self-training approach that enables models to differentiate between well-grounded and less accurate temporal responses by leveraging curated preference datasets at two granularities: localized temporal grounding, which focuses on specific video segments, and comprehensive temporal grounding, which captures extended temporal dependencies across entire video sequences. By optimizing on these preference datasets, TPO significantly enhances temporal understanding while reducing reliance on manually annotated data. Extensive experiments on three long-form video understanding benchmarks--LongVideoBench, MLVU, and Video-MME--demonstrate the effectiveness of TPO across two state-of-the-art video-LMMs. Notably, LLaVA-Video-TPO establishes itself as the leading 7B model on the Video-MME benchmark, underscoring the potential of TPO as a scalable and efficient solution for advancing temporal reasoning in long-form video understanding. Project page: https://ruili33.github.io/tpo_website.
Recent advances in Large Language Models (LLMs) have enabled the development of Video-LLMs, advancing multimodal learning by bridging video data with language tasks. However, current video understanding models struggle with processing long video sequences, supporting multi-turn dialogues, and adapting to real-world dynamic scenarios. To address these issues, we propose StreamChat, a training-free framework for streaming video reasoning and conversational interaction. $\StreamChat$ leverages a novel hierarchical memory system to efficiently process and compress video features over extended sequences, enabling real-time, multi-turn dialogue. Our framework incorporates a parallel system scheduling strategy that enhances processing speed and reduces latency, ensuring robust performance in real-world applications. Furthermore, we introduce StreamBench, a versatile benchmark that evaluates streaming video understanding across diverse media types and interactive scenarios, including multi-turn interactions and complex reasoning tasks. Extensive evaluations on StreamBench and other public benchmarks demonstrate that StreamChat significantly outperforms existing state-of-the-art models in terms of accuracy and response times, confirming its effectiveness for streaming video understanding. Code is available at StreamChat: https://github.com/hmxiong/StreamChat.
In this paper, we propose VideoLLaMA3, a more advanced multimodal foundation model for image and video understanding. The core design philosophy of VideoLLaMA3 is vision-centric. The meaning of "vision-centric" is two-fold: the vision-centric training paradigm and vision-centric framework design. The key insight of our vision-centric training paradigm is that high-quality image-text data is crucial for both image and video understanding. Instead of preparing massive video-text datasets, we focus on constructing large-scale and high-quality image-text datasets. VideoLLaMA3 has four training stages: 1) Vision Encoder Adaptation, which enables vision encoder to accept images of variable resolutions as input; 2) Vision-Language Alignment, which jointly tunes the vision encoder, projector, and LLM with large-scale image-text data covering multiple types (including scene images, documents, charts) as well as text-only data. 3) Multi-task Fine-tuning, which incorporates image-text SFT data for downstream tasks and video-text data to establish a foundation for video understanding. 4) Video-centric Fine-tuning, which further improves the model's capability in video understanding. As for the framework design, to better capture fine-grained details in images, the pretrained vision encoder is adapted to encode images of varying sizes into vision tokens with corresponding numbers, rather than a fixed number of tokens. For video inputs, we reduce the number of vision tokens according to their similarity so that the representation of videos will be more precise and compact. Benefit from vision-centric designs, VideoLLaMA3 achieves compelling performances in both image and video understanding benchmarks.
In recent years, action recognition has received much attention and wide application due to its important role in video understanding. Most of the researches on action recognition methods focused on improving the performance via various deep learning methods rather than the classification of skeleton points. The topological modeling between skeleton points and body parts was seldom considered. Although some studies have used a data-driven approach to classify the topology of the skeleton point, the nature of the skeleton point in terms of kinematics has not been taken into consideration. Therefore, in this paper, we draw on the theory of kinematics to adapt the topological relations of the skeleton point and propose a topological relation classification based on body parts and distance from core of body. To synthesize these topological relations for action recognition, we propose a novel Hypergraph Fusion Graph Convolutional Network (HFGCN). In particular, the proposed model is able to focus on the human skeleton points and the different body parts simultaneously, and thus construct the topology, which improves the recognition accuracy obviously. We use a hypergraph to represent the categorical relationships of these skeleton points and incorporate the hypergraph into a graph convolution network to model the higher-order relationships among the skeleton points and enhance the feature representation of the network. In addition, our proposed hypergraph attention module and hypergraph graph convolution module optimize topology modeling in temporal and channel dimensions, respectively, to further enhance the feature representation of the network. We conducted extensive experiments on three widely used datasets.The results validate that our proposed method can achieve the best performance when compared with the state-of-the-art skeleton-based methods.
This paper aims to improve the performance of video multimodal large language models (MLLM) via long and rich context (LRC) modeling. As a result, we develop a new version of InternVideo2.5 with a focus on enhancing the original MLLMs' ability to perceive fine-grained details and capture long-form temporal structure in videos. Specifically, our approach incorporates dense vision task annotations into MLLMs using direct preference optimization and develops compact spatiotemporal representations through adaptive hierarchical token compression. Experimental results demonstrate this unique design of LRC greatly improves the results of video MLLM in mainstream video understanding benchmarks (short & long), enabling the MLLM to memorize significantly longer video inputs (at least 6x longer than the original), and master specialized vision capabilities like object tracking and segmentation. Our work highlights the importance of multimodal context richness (length and fineness) in empowering MLLM's innate abilites (focus and memory), providing new insights for future research on video MLLM. Code and models are available at https://github.com/OpenGVLab/InternVideo/tree/main/InternVideo2.5
Despite the promising performance of Large Vision Language Models (LVLMs) in visual understanding, they occasionally generate incorrect outputs. While reward models (RMs) with reinforcement learning or test-time scaling offer the potential for improving generation quality, a critical gap remains: publicly available multi-modal RMs for LVLMs are scarce, and the implementation details of proprietary models are often unclear. We bridge this gap with InternLM-XComposer2.5-Reward (IXC-2.5-Reward), a simple yet effective multi-modal reward model that aligns LVLMs with human preferences. To ensure the robustness and versatility of IXC-2.5-Reward, we set up a high-quality multi-modal preference corpus spanning text, image, and video inputs across diverse domains, such as instruction following, general understanding, text-rich documents, mathematical reasoning, and video understanding. IXC-2.5-Reward achieves excellent results on the latest multi-modal reward model benchmark and shows competitive performance on text-only reward model benchmarks. We further demonstrate three key applications of IXC-2.5-Reward: (1) Providing a supervisory signal for RL training. We integrate IXC-2.5-Reward with Proximal Policy Optimization (PPO) yields IXC-2.5-Chat, which shows consistent improvements in instruction following and multi-modal open-ended dialogue; (2) Selecting the best response from candidate responses for test-time scaling; and (3) Filtering outlier or noisy samples from existing image and video instruction tuning training data. To ensure reproducibility and facilitate further research, we have open-sourced all model weights and training recipes at https://github.com/InternLM/InternLM-XComposer/tree/main/InternLM-XComposer-2.5-Reward
We introduce MMVU, a comprehensive expert-level, multi-discipline benchmark for evaluating foundation models in video understanding. MMVU includes 3,000 expert-annotated questions spanning 27 subjects across four core disciplines: Science, Healthcare, Humanities & Social Sciences, and Engineering. Compared to prior benchmarks, MMVU features three key advancements. First, it challenges models to apply domain-specific knowledge and perform expert-level reasoning to analyze specialized-domain videos, moving beyond the basic visual perception typically assessed in current video benchmarks. Second, each example is annotated by human experts from scratch. We implement strict data quality controls to ensure the high quality of the dataset. Finally, each example is enriched with expert-annotated reasoning rationals and relevant domain knowledge, facilitating in-depth analysis. We conduct an extensive evaluation of 32 frontier multimodal foundation models on MMVU. The latest System-2-capable models, o1 and Gemini 2.0 Flash Thinking, achieve the highest performance among the tested models. However, they still fall short of matching human expertise. Through in-depth error analyses and case studies, we offer actionable insights for future advancements in expert-level, knowledge-intensive video understanding for specialized domains.
We present Omni-RGPT, a multimodal large language model designed to facilitate region-level comprehension for both images and videos. To achieve consistent region representation across spatio-temporal dimensions, we introduce Token Mark, a set of tokens highlighting the target regions within the visual feature space. These tokens are directly embedded into spatial regions using region prompts (e.g., boxes or masks) and simultaneously incorporated into the text prompt to specify the target, establishing a direct connection between visual and text tokens. To further support robust video understanding without requiring tracklets, we introduce an auxiliary task that guides Token Mark by leveraging the consistency of the tokens, enabling stable region interpretation across the video. Additionally, we introduce a large-scale region-level video instruction dataset (RegVID-300k). Omni-RGPT achieves state-of-the-art results on image and video-based commonsense reasoning benchmarks while showing strong performance in captioning and referring expression comprehension tasks.
The essence of audio-visual segmentation (AVS) lies in locating and delineating sound-emitting objects within a video stream. While Transformer-based methods have shown promise, their handling of long-range dependencies struggles due to quadratic computational costs, presenting a bottleneck in complex scenarios. To overcome this limitation and facilitate complex multi-modal comprehension with linear complexity, we introduce AVS-Mamba, a selective state space model to address the AVS task. Our framework incorporates two key components for video understanding and cross-modal learning: Temporal Mamba Block for sequential video processing and Vision-to-Audio Fusion Block for advanced audio-vision integration. Building on this, we develop the Multi-scale Temporal Encoder, aimed at enhancing the learning of visual features across scales, facilitating the perception of intra- and inter-frame information. To perform multi-modal fusion, we propose the Modality Aggregation Decoder, leveraging the Vision-to-Audio Fusion Block to integrate visual features into audio features across both frame and temporal levels. Further, we adopt the Contextual Integration Pyramid to perform audio-to-vision spatial-temporal context collaboration. Through these innovative contributions, our approach achieves new state-of-the-art results on the AVSBench-object and AVSBench-semantic datasets. Our source code and model weights are available at AVS-Mamba.
Facial expression captioning has found widespread application across various domains. Recently, the emergence of video Multimodal Large Language Models (MLLMs) has shown promise in general video understanding tasks. However, describing facial expressions within videos poses two major challenges for these models: (1) the lack of adequate datasets and benchmarks, and (2) the limited visual token capacity of video MLLMs. To address these issues, this paper introduces a new instruction-following dataset tailored for dynamic facial expression caption. The dataset comprises 5,033 high-quality video clips annotated manually, containing over 700,000 tokens. Its purpose is to improve the capability of video MLLMs to discern subtle facial nuances. Furthermore, we propose FaceTrack-MM, which leverages a limited number of tokens to encode the main character's face. This model demonstrates superior performance in tracking faces and focusing on the facial expressions of the main characters, even in intricate multi-person scenarios. Additionally, we introduce a novel evaluation metric combining event extraction, relation classification, and the longest common subsequence (LCS) algorithm to assess the content consistency and temporal sequence consistency of generated text. Moreover, we present FEC-Bench, a benchmark designed to assess the performance of existing video MLLMs in this specific task. All data and source code will be made publicly available.
We introduce Tarsier2, a state-of-the-art large vision-language model (LVLM) designed for generating detailed and accurate video descriptions, while also exhibiting superior general video understanding capabilities. Tarsier2 achieves significant advancements through three key upgrades: (1) Scaling pre-training data from 11M to 40M video-text pairs, enriching both volume and diversity; (2) Performing fine-grained temporal alignment during supervised fine-tuning; (3) Using model-based sampling to automatically construct preference data and applying DPO training for optimization. Extensive experiments show that Tarsier2-7B consistently outperforms leading proprietary models, including GPT-4o and Gemini 1.5 Pro, in detailed video description tasks. On the DREAM-1K benchmark, Tarsier2-7B improves F1 by 2.8% over GPT-4o and 5.8% over Gemini-1.5-Pro. In human side-by-side evaluations, Tarsier2-7B shows a +8.6% performance advantage over GPT-4o and +24.9% over Gemini-1.5-Pro. Tarsier2-7B also sets new state-of-the-art results across 15 public benchmarks, spanning tasks such as video question-answering, video grounding, hallucination test, and embodied question-answering, demonstrating its versatility as a robust generalist vision-language model.
Despite the advancements of Video Large Language Models (VideoLLMs) in various tasks, they struggle with fine-grained temporal understanding, such as Dense Video Captioning (DVC). DVC is a complicated task of describing all events within a video while also temporally localizing them, which integrates multiple fine-grained tasks, including video segmentation, video captioning, and temporal video grounding. Previous VideoLLMs attempt to solve DVC in a single step, failing to utilize their reasoning capability. Moreover, previous training objectives for VideoLLMs do not fully reflect the evaluation metrics, therefore not providing supervision directly aligned to target tasks. To address such a problem, we propose a novel framework named VidChain comprised of Chain-of-Tasks (CoTasks) and Metric-based Direct Preference Optimization (M-DPO). CoTasks decompose a complex task into a sequence of sub-tasks, allowing VideoLLMs to leverage their reasoning capabilities more effectively. M-DPO aligns a VideoLLM with evaluation metrics, providing fine-grained supervision to each task that is well-aligned with metrics. Applied to two different VideoLLMs, VidChain consistently improves their fine-grained video understanding, thereby outperforming previous VideoLLMs on two different DVC benchmarks and also on the temporal video grounding task. Code is available at \url{https://github.com/mlvlab/VidChain}.
Video causal reasoning aims to achieve a high-level understanding of videos from a causal perspective. However, it exhibits limitations in its scope, primarily executed in a question-answering paradigm and focusing on brief video segments containing isolated events and basic causal relations, lacking comprehensive and structured causality analysis for videos with multiple interconnected events. To fill this gap, we introduce a new task and dataset, Multi-Event Causal Discovery (MECD). It aims to uncover the causal relations between events distributed chronologically across long videos. Given visual segments and textual descriptions of events, MECD identifies the causal associations between these events to derive a comprehensive and structured event-level video causal graph explaining why and how the result event occurred. To address the challenges of MECD, we devise a novel framework inspired by the Granger Causality method, incorporating an efficient mask-based event prediction model to perform an Event Granger Test. It estimates causality by comparing the predicted result event when premise events are masked versus unmasked. Furthermore, we integrate causal inference techniques such as front-door adjustment and counterfactual inference to mitigate challenges in MECD like causality confounding and illusory causality. Additionally, context chain reasoning is introduced to conduct more robust and generalized reasoning. Experiments validate the effectiveness of our framework in reasoning complete causal relations, outperforming GPT-4o and VideoChat2 by 5.77% and 2.70%, respectively. Further experiments demonstrate that causal relation graphs can also contribute to downstream video understanding tasks such as video question answering and video event prediction.
Long-form egocentric video understanding provides rich contextual information and unique insights into long-term human behaviors, holding significant potential for applications in embodied intelligence, long-term activity analysis, and personalized assistive technologies. However, existing benchmark datasets primarily focus on single, short-duration videos or moderately long videos up to dozens of minutes, leaving a substantial gap in evaluating extensive, ultra-long egocentric video recordings. To address this, we introduce X-LeBench, a novel benchmark dataset specifically crafted for evaluating tasks on extremely long egocentric video recordings. Leveraging the advanced text processing capabilities of large language models (LLMs), X-LeBench develops a life-logging simulation pipeline that produces realistic, coherent daily plans aligned with real-world video data. This approach enables the flexible integration of synthetic daily plans with real-world footage from Ego4D-a massive-scale egocentric video dataset covers a wide range of daily life scenarios-resulting in 432 simulated video life logs that mirror realistic daily activities in contextually rich scenarios. The video life-log durations span from 23 minutes to 16.4 hours. The evaluation of several baseline systems and multimodal large language models (MLLMs) reveals their poor performance across the board, highlighting the inherent challenges of long-form egocentric video understanding and underscoring the need for more advanced models.
With the rapid development of multimedia processing and deep learning technologies, especially in the field of video understanding, video quality assessment (VQA) has achieved significant progress. Although researchers have moved from designing efficient video quality mapping models to various research directions, in-depth exploration of the effectiveness-efficiency trade-offs of spatio-temporal modeling in VQA models is still less sufficient. Considering the fact that videos have highly redundant information, this paper investigates this problem from the perspective of joint spatial and temporal sampling, aiming to seek the answer to how little information we should keep at least when feeding videos into the VQA models while with acceptable performance sacrifice. To this end, we drastically sample the video's information from both spatial and temporal dimensions, and the heavily squeezed video is then fed into a stable VQA model. Comprehensive experiments regarding joint spatial and temporal sampling are conducted on six public video quality databases, and the results demonstrate the acceptable performance of the VQA model when throwing away most of the video information. Furthermore, with the proposed joint spatial and temporal sampling strategy, we make an initial attempt to design an online VQA model, which is instantiated by as simple as possible a spatial feature extractor, a temporal feature fusion module, and a global quality regression module. Through quantitative and qualitative experiments, we verify the feasibility of online VQA model by simplifying itself and reducing input.