We introduce InternVideo2, a new family of video foundation models (ViFM) that achieve the state-of-the-art results in video recognition, video-text tasks, and video-centric dialogue. Our core design is a progressive training approach that unifies the masked video modeling, crossmodal contrastive learning, and next token prediction, scaling up the video encoder size to 6B parameters. At the data level, we prioritize spatiotemporal consistency by semantically segmenting videos and generating video-audio-speech captions. This improves the alignment between video and text. Through extensive experiments, we validate our designs and demonstrate superior performance on over 60 video and audio tasks. Notably, our model outperforms others on various video-related dialogue and long video understanding benchmarks, highlighting its ability to reason and comprehend longer contexts. Code and models are available at https://github.com/OpenGVLab/InternVideo/tree/main/InternVideo2/.
Recent studies have demonstrated the effectiveness of Large Language Models (LLMs) as reasoning modules that can deconstruct complex tasks into more manageable sub-tasks, particularly when applied to visual reasoning tasks for images. In contrast, this paper introduces a Video Understanding and Reasoning Framework (VURF) based on the reasoning power of LLMs. Ours is a novel approach to extend the utility of LLMs in the context of video tasks, leveraging their capacity to generalize from minimal input and output demonstrations within a contextual framework. By presenting LLMs with pairs of instructions and their corresponding high-level programs, we harness their contextual learning capabilities to generate executable visual programs for video understanding. To enhance program's accuracy and robustness, we implement two important strategies. Firstly, we employ a feedback-generation approach, powered by GPT-3.5, to rectify errors in programs utilizing unsupported functions. Secondly, taking motivation from recent works on self refinement of LLM outputs, we introduce an iterative procedure for improving the quality of the in-context examples by aligning the initial outputs to the outputs that would have been generated had the LLM not been bound by the structure of the in-context examples. Our results on several video-specific tasks, including visual QA, video anticipation, pose estimation and multi-video QA illustrate the efficacy of these enhancements in improving the performance of visual programming approaches for video tasks.
Language has become a prominent modality in computer vision with the rise of LLMs. Despite supporting long context-lengths, their effectiveness in handling long-term information gradually declines with input length. This becomes critical, especially in applications such as long-form video understanding. In this paper, we introduce a Language Repository (LangRepo) for LLMs, that maintains concise and structured information as an interpretable (i.e., all-textual) representation. Our repository is updated iteratively based on multi-scale video chunks. We introduce write and read operations that focus on pruning redundancies in text, and extracting information at various temporal scales. The proposed framework is evaluated on zero-shot visual question-answering benchmarks including EgoSchema, NExT-QA, IntentQA and NExT-GQA, showing state-of-the-art performance at its scale. Our code is available at https://github.com/kkahatapitiya/LangRepo.
In this paper, we explore the visual representations produced from a pre-trained text-to-video (T2V) diffusion model for video understanding tasks. We hypothesize that the latent representation learned from a pretrained generative T2V model encapsulates rich semantics and coherent temporal correspondences, thereby naturally facilitating video understanding. Our hypothesis is validated through the classic referring video object segmentation (R-VOS) task. We introduce a novel framework, termed "VD-IT", tailored with dedicatedly designed components built upon a fixed pretrained T2V model. Specifically, VD-IT uses textual information as a conditional input, ensuring semantic consistency across time for precise temporal instance matching. It further incorporates image tokens as supplementary textual inputs, enriching the feature set to generate detailed and nuanced masks. Besides, instead of using the standard Gaussian noise, we propose to predict the video-specific noise with an extra noise prediction module, which can help preserve the feature fidelity and elevates segmentation quality. Through extensive experiments, we surprisingly observe that fixed generative T2V diffusion models, unlike commonly used video backbones (e.g., Video Swin Transformer) pretrained with discriminative image/video pre-tasks, exhibit better potential to maintain semantic alignment and temporal consistency. On existing standard benchmarks, our VD-IT achieves highly competitive results, surpassing many existing state-of-the-art methods. The code is available at https://github.com/buxiangzhiren/VD-IT.
We explore how reconciling several foundation models (large language models and vision-language models) with a novel unified memory mechanism could tackle the challenging video understanding problem, especially capturing the long-term temporal relations in lengthy videos. In particular, the proposed multimodal agent VideoAgent: 1) constructs a structured memory to store both the generic temporal event descriptions and object-centric tracking states of the video; 2) given an input task query, it employs tools including video segment localization and object memory querying along with other visual foundation models to interactively solve the task, utilizing the zero-shot tool-use ability of LLMs. VideoAgent demonstrates impressive performances on several long-horizon video understanding benchmarks, an average increase of 6.6% on NExT-QA and 26.0% on EgoSchema over baselines, closing the gap between open-sourced models and private counterparts including Gemini 1.5 Pro.
Long-form video understanding represents a significant challenge within computer vision, demanding a model capable of reasoning over long multi-modal sequences. Motivated by the human cognitive process for long-form video understanding, we emphasize interactive reasoning and planning over the ability to process lengthy visual inputs. We introduce a novel agent-based system, VideoAgent, that employs a large language model as a central agent to iteratively identify and compile crucial information to answer a question, with vision-language foundation models serving as tools to translate and retrieve visual information. Evaluated on the challenging EgoSchema and NExT-QA benchmarks, VideoAgent achieves 54.1% and 71.3% zero-shot accuracy with only 8.4 and 8.2 frames used on average. These results demonstrate superior effectiveness and efficiency of our method over the current state-of-the-art methods, highlighting the potential of agent-based approaches in advancing long-form video understanding.
Understanding videos is one of the fundamental directions in computer vision research, with extensive efforts dedicated to exploring various architectures such as RNN, 3D CNN, and Transformers. The newly proposed architecture of state space model, e.g., Mamba, shows promising traits to extend its success in long sequence modeling to video modeling. To assess whether Mamba can be a viable alternative to Transformers in the video understanding domain, in this work, we conduct a comprehensive set of studies, probing different roles Mamba can play in modeling videos, while investigating diverse tasks where Mamba could exhibit superiority. We categorize Mamba into four roles for modeling videos, deriving a Video Mamba Suite composed of 14 models/modules, and evaluating them on 12 video understanding tasks. Our extensive experiments reveal the strong potential of Mamba on both video-only and video-language tasks while showing promising efficiency-performance trade-offs. We hope this work could provide valuable data points and insights for future research on video understanding. Code is public: https://github.com/OpenGVLab/video-mamba-suite.
Current training pipelines in object recognition neglect Hue Jittering when doing data augmentation as it not only brings appearance changes that are detrimental to classification, but also the implementation is inefficient in practice. In this study, we investigate the effect of hue variance in the context of video understanding and find this variance to be beneficial since static appearances are less important in videos that contain motion information. Based on this observation, we propose a data augmentation method for video understanding, named Motion Coherent Augmentation (MCA), that introduces appearance variation in videos and implicitly encourages the model to prioritize motion patterns, rather than static appearances. Concretely, we propose an operation SwapMix to efficiently modify the appearance of video samples, and introduce Variation Alignment (VA) to resolve the distribution shift caused by SwapMix, enforcing the model to learn appearance invariant representations. Comprehensive empirical evaluation across various architectures and different datasets solidly validates the effectiveness and generalization ability of MCA, and the application of VA in other augmentation methods. Code is available at https://github.com/BeSpontaneous/MCA-pytorch.
We introduce a novel text-to-pose video editing method, ReimaginedAct. While existing video editing tasks are limited to changes in attributes, backgrounds, and styles, our method aims to predict open-ended human action changes in video. Moreover, our method can accept not only direct instructional text prompts but also `what if' questions to predict possible action changes. ReimaginedAct comprises video understanding, reasoning, and editing modules. First, an LLM is utilized initially to obtain a plausible answer for the instruction or question, which is then used for (1) prompting Grounded-SAM to produce bounding boxes of relevant individuals and (2) retrieving a set of pose videos that we have collected for editing human actions. The retrieved pose videos and the detected individuals are then utilized to alter the poses extracted from the original video. We also employ a timestep blending module to ensure the edited video retains its original content except where necessary modifications are needed. To facilitate research in text-to-pose video editing, we introduce a new evaluation dataset, WhatifVideo-1.0. This dataset includes videos of different scenarios spanning a range of difficulty levels, along with questions and text prompts. Experimental results demonstrate that existing video editing methods struggle with human action editing, while our approach can achieve effective action editing and even imaginary editing from counterfactual questions.
Addressing the dual challenges of local redundancy and global dependencies in video understanding, this work innovatively adapts the Mamba to the video domain. The proposed VideoMamba overcomes the limitations of existing 3D convolution neural networks and video transformers. Its linear-complexity operator enables efficient long-term modeling, which is crucial for high-resolution long video understanding. Extensive evaluations reveal VideoMamba's four core abilities: (1) Scalability in the visual domain without extensive dataset pretraining, thanks to a novel self-distillation technique; (2) Sensitivity for recognizing short-term actions even with fine-grained motion differences; (3) Superiority in long-term video understanding, showcasing significant advancements over traditional feature-based models; and (4) Compatibility with other modalities, demonstrating robustness in multi-modal contexts. Through these distinct advantages, VideoMamba sets a new benchmark for video understanding, offering a scalable and efficient solution for comprehensive video understanding. All the code and models are available at https://github.com/OpenGVLab/VideoMamba.
In this study, we identify the inefficient attention phenomena in Large Vision-Language Models (LVLMs), notably within prominent models like LLaVA-1.5, QwenVL-Chat and Video-LLaVA. We find out that the attention computation over visual tokens is of extreme inefficiency in the deep layers of popular LVLMs, suggesting a need for a sparser approach compared to textual data handling. To this end, we introduce FastV, a versatile plug-and-play method designed to optimize computational efficiency by learning adaptive attention patterns in early layers and pruning visual tokens in subsequent ones. Our evaluations demonstrate FastV's ability to dramatically reduce computational costs (e.g., a 45 reduction in FLOPs for LLaVA-1.5-13B) without sacrificing performance in a wide range of image and video understanding tasks. The computational efficiency and performance trade-off of FastV are highly customizable and pareto-efficient. It can compress the FLOPs of a 13B-parameter model to achieve a lower budget than that of a 7B-parameter model, while still maintaining superior performance. We believe FastV has practical values for deployment of LVLMs in edge devices and commercial models. Code is released at https://github.com/pkunlp-icler/FastV.
Current multi-object tracking (MOT) aims to predict trajectories of targets (i.e., ''where'') in videos. Yet, knowing merely ''where'' is insufficient in many crucial applications. In comparison, semantic understanding such as fine-grained behaviors, interactions, and overall summarized captions (i.e., ''what'') from videos, associated with ''where'', is highly-desired for comprehensive video analysis. Thus motivated, we introduce Semantic Multi-Object Tracking (SMOT), that aims to estimate object trajectories and meanwhile understand semantic details of associated trajectories including instance captions, instance interactions, and overall video captions, integrating ''where'' and ''what'' for tracking. In order to foster the exploration of SMOT, we propose BenSMOT, a large-scale Benchmark for Semantic MOT. Specifically, BenSMOT comprises 3,292 videos with 151K frames, covering various scenarios for semantic tracking of humans. BenSMOT provides annotations for the trajectories of targets, along with associated instance captions in natural language, instance interactions, and overall caption for each video sequence. To our best knowledge, BenSMOT is the first publicly available benchmark for SMOT. Besides, to encourage future research, we present a novel tracker named SMOTer, which is specially designed and end-to-end trained for SMOT, showing promising performance. By releasing BenSMOT, we expect to go beyond conventional MOT by predicting ''where'' and ''what'' for SMOT, opening up a new direction in tracking for video understanding. We will release BenSMOT and SMOTer at https://github.com/Nathan-Li123/SMOTer.
Development of multimodal models has marked a significant step forward in how machines understand videos. These models have shown promise in analyzing short video clips. However, when it comes to longer formats like movies, they often fall short. The main hurdles are the lack of high-quality, diverse video data and the intensive work required to collect or annotate such data. In face of these challenges, we propose MovieLLM, a novel framework designed to synthesize consistent and high-quality video data for instruction tuning. The pipeline is carefully designed to control the style of videos by improving textual inversion technique with powerful text generation capability of GPT-4. As the first framework to do such thing, our approach stands out for its flexibility and scalability, empowering users to create customized movies with only one description. This makes it a superior alternative to traditional data collection methods. Our extensive experiments validate that the data produced by MovieLLM significantly improves the performance of multimodal models in understanding complex video narratives, overcoming the limitations of existing datasets regarding scarcity and bias.
Human comprehension of a video stream is naturally broad: in a few instants, we are able to understand what is happening, the relevance and relationship of objects, and forecast what will follow in the near future, everything all at once. We believe that - to effectively transfer such an holistic perception to intelligent machines - an important role is played by learning to correlate concepts and to abstract knowledge coming from different tasks, to synergistically exploit them when learning novel skills. To accomplish this, we seek for a unified approach to video understanding which combines shared temporal modelling of human actions with minimal overhead, to support multiple downstream tasks and enable cooperation when learning novel skills. We then propose EgoPack, a solution that creates a collection of task perspectives that can be carried across downstream tasks and used as a potential source of additional insights, as a backpack of skills that a robot can carry around and use when needed. We demonstrate the effectiveness and efficiency of our approach on four Ego4D benchmarks, outperforming current state-of-the-art methods.
We present MM-AU, a novel dataset for Multi-Modal Accident video Understanding. MM-AU contains 11,727 in-the-wild ego-view accident videos, each with temporally aligned text descriptions. We annotate over 2.23 million object boxes and 58,650 pairs of video-based accident reasons, covering 58 accident categories. MM-AU supports various accident understanding tasks, particularly multimodal video diffusion to understand accident cause-effect chains for safe driving. With MM-AU, we present an Abductive accident Video understanding framework for Safe Driving perception (AdVersa-SD). AdVersa-SD performs video diffusion via an Object-Centric Video Diffusion (OAVD) method which is driven by an abductive CLIP model. This model involves a contrastive interaction loss to learn the pair co-occurrence of normal, near-accident, accident frames with the corresponding text descriptions, such as accident reasons, prevention advice, and accident categories. OAVD enforces the causal region learning while fixing the content of the original frame background in video generation, to find the dominant cause-effect chain for certain accidents. Extensive experiments verify the abductive ability of AdVersa-SD and the superiority of OAVD against the state-of-the-art diffusion models. Additionally, we provide careful benchmark evaluations for object detection and accident reason answering since AdVersa-SD relies on precise object and accident reason information.
It is challenging for models to understand complex, multimodal content such as television clips, and this is in part because video-language models often rely on single-modality reasoning and lack interpretability. To combat these issues we propose TV-TREES, the first multimodal entailment tree generator. TV-TREES serves as an approach to video understanding that promotes interpretable joint-modality reasoning by searching for trees of entailment relationships between simple text-video evidence and higher-level conclusions that prove question-answer pairs. We also introduce the task of multimodal entailment tree generation to evaluate reasoning quality. Our method's performance on the challenging TVQA benchmark demonstrates interpretable, state-of-the-art zero-shot performance on full clips, illustrating that multimodal entailment tree generation can be a best-of-both-worlds alternative to black-box systems.
Long video understanding is a significant and ongoing challenge in the intersection of multimedia and artificial intelligence. Employing large language models (LLMs) for comprehending video becomes an emerging and promising method. However, this approach incurs high computational costs due to the extensive array of video tokens, experiences reduced visual clarity as a consequence of token aggregation, and confronts challenges arising from irrelevant visual tokens while answering video-related questions. To alleviate these issues, we present an Interactive Visual Adapter (IVA) within LLMs, designed to enhance interaction with fine-grained visual elements. Specifically, we first transform long videos into temporal video tokens via leveraging a visual encoder alongside a pretrained causal transformer, then feed them into LLMs with the video instructions. Subsequently, we integrated IVA, which contains a lightweight temporal frame selector and a spatial feature interactor, within the internal blocks of LLMs to capture instruction-aware and fine-grained visual signals. Consequently, the proposed video-LLM facilitates a comprehensive understanding of long video content through appropriate long video modeling and precise visual interactions. We conducted extensive experiments on nine video understanding benchmarks and experimental results show that our interactive visual adapter significantly improves the performance of video LLMs on long video QA tasks. Ablation studies further verify the effectiveness of IVA in understanding long and short video.
Extensive literature has drawn comparisons between recordings of biological neurons in the brain and deep neural networks. This comparative analysis aims to advance and interpret deep neural networks and enhance our understanding of biological neural systems. However, previous works did not consider the time aspect and how the encoding of video and dynamics in deep networks relate to the biological neural systems within a large-scale comparison. Towards this end, we propose the first large-scale study focused on comparing video understanding models with respect to the visual cortex recordings using video stimuli. The study encompasses more than two million regression fits, examining image vs. video understanding, convolutional vs. transformer-based and fully vs. self-supervised models. Additionally, we propose a novel neural encoding scheme to better encode biological neural systems. We provide key insights on how video understanding models predict visual cortex responses; showing video understanding better than image understanding models, convolutional models are better in the early-mid visual cortical regions than transformer based ones except for multiscale transformers, and that two-stream models are better than single stream. Furthermore, we propose a novel neural encoding scheme that is built on top of the best performing video understanding models, while incorporating inter-intra region connectivity across the visual cortex. Our neural encoding leverages the encoded dynamics from video stimuli, through utilizing two-stream networks and multiscale transformers, while taking connectivity priors into consideration. Our results show that merging both intra and inter-region connectivity priors increases the encoding performance over each one of them standalone or no connectivity priors. It also shows the necessity for encoding dynamics to fully benefit from such connectivity priors.
Most video captioning models are designed to process short video clips of few seconds and output text describing low-level visual concepts (e.g., objects, scenes, atomic actions). However, most real-world videos last for minutes or hours and have a complex hierarchical structure spanning different temporal granularities. We propose Video ReCap, a recursive video captioning model that can process video inputs of dramatically different lengths (from 1 second to 2 hours) and output video captions at multiple hierarchy levels. The recursive video-language architecture exploits the synergy between different video hierarchies and can process hour-long videos efficiently. We utilize a curriculum learning training scheme to learn the hierarchical structure of videos, starting from clip-level captions describing atomic actions, then focusing on segment-level descriptions, and concluding with generating summaries for hour-long videos. Furthermore, we introduce Ego4D-HCap dataset by augmenting Ego4D with 8,267 manually collected long-range video summaries. Our recursive model can flexibly generate captions at different hierarchy levels while also being useful for other complex video understanding tasks, such as VideoQA on EgoSchema. Data, code, and models are available at: https://sites.google.com/view/vidrecap
Video-Language Models (VLMs), powered by the advancements in Large Language Models (LLMs), are charting new frontiers in video understanding. A pivotal challenge is the development of an efficient method to encapsulate video content into a set of representative tokens to align with LLMs. In this work, we introduce Slot-VLM, a novel framework designed to generate semantically decomposed video tokens, in terms of object-wise and event-wise visual representations, to facilitate LLM inference. Particularly, we design a SlowFast Slots module, i.e., SF-Slots, that adaptively aggregates the dense video tokens from the CLIP vision encoder to a set of representative slots. In order to take into account both the spatial object details and the varied temporal dynamics, SF-Slots is built with a dual-branch structure. The Slow-Slots branch focuses on extracting object-centric slots from features at high spatial resolution but low (slow) frame sample rate, emphasizing detailed object information. Conversely, Fast-Slots branch is engineered to learn event-centric slots from high temporal sample rate but low spatial resolution features. These complementary slots are combined to form the vision context, serving as the input to the LLM for efficient question answering. Our experimental results demonstrate the effectiveness of our Slot-VLM, which achieves the state-of-the-art performance on video question-answering.