In recent years, there has been a significant surge of interest in unifying image comprehension and generation within Large Language Models (LLMs). This growing interest has prompted us to explore extending this unification to videos. The core challenge lies in developing a versatile video tokenizer that captures both the spatial characteristics and temporal dynamics of videos to obtain representations for LLMs, and the representations can be further decoded into realistic video clips to enable video generation. In this work, we introduce Divot, a Diffusion-Powered Video Tokenizer, which leverages the diffusion process for self-supervised video representation learning. We posit that if a video diffusion model can effectively de-noise video clips by taking the features of a video tokenizer as the condition, then the tokenizer has successfully captured robust spatial and temporal information. Additionally, the video diffusion model inherently functions as a de-tokenizer, decoding videos from their representations. Building upon the Divot tokenizer, we present Divot-LLM through video-to-text auto-regression and text-to-video generation by modeling the distributions of continuous-valued Divot features with a Gaussian Mixture Model. Experimental results demonstrate that our diffusion-based video tokenizer, when integrated with a pre-trained LLM, achieves competitive performance across various video comprehension and generation benchmarks. The instruction tuned Divot-LLM also excels in video storytelling, generating interleaved narratives and corresponding videos. Models and codes are available at https://github.com/TencentARC/Divot.

With the rapid development of Multi-modal Large language Models (MLLMs), a number of diagnostic bench-marks have recently emerged to evaluate the comprehension capabilities of these models. However, most bench-marks predominantly assess spatial understanding in the static image tasks, while overlooking temporal understanding in the dynamic video tasks. To alleviate this issue, we introduce a comprehensive Multi-modal Video understanding Benchmark, namely MVBench, which covers 20 chal-lenging video tasks that cannot be effectively solved with a single frame. Specifically, we first introduce a novel static-to-dynamic method to define these temporal-related tasks. By transforming various static tasks into dynamic ones, we enable the systematic generation of video tasks that require a broad spectrum of temporal skills, ranging from perception to cognition. Then, guided by the task definition, we au-tomatically convert public video annotations into multiple-choice QA to evaluate each task. On one hand, such a distinct paradigm allows us to build MVBench efficiently, without much manual intervention. On the other hand, it guarantees evaluation fairness with ground-truth video an-notations, avoiding the biased scoring of LLMs. More-over, we further develop a robust video MLLM baseline, i.e., VideoChat2, by progressive multi-modal training with di-verse instruction-tuning data. The extensive results on our MVBench reveal that, the existing MLLMs are far from sat-isfactory in temporal understanding, while our VideoChat2 largely surpasses these leading models by over 15% on MVBench. All models and data are available at https://github.com/OpenGVLab/Ask-Anything.

This paper introduces InternVid, a large-scale video-centric multimodal dataset that enables learning powerful and transferable video-text representations for multimodal understanding and generation. InternVid contains over 7 million videos lasting nearly 760K hours, yielding 234M video clips accompanied by detailed descriptions of total 4.1B words. Our core contribution is to develop a scalable approach to autonomously build a high-quality video-text dataset with large language models (LLM), thereby showcasing its efficacy in learning video-language representation at scale. Specifically, we utilize a multi-scale approach to generate video-related descriptions. Furthermore, we introduce ViCLIP, a video-text representation learning model based on ViT-L. Learned on InternVid via contrastive learning, this model demonstrates leading zero-shot action recognition and competitive video retrieval performance. Beyond basic video understanding tasks like recognition and retrieval, our dataset and model have broad applications. They are particularly beneficial for generating interleaved video-text data for learning a video-centric dialogue system, advancing video-to-text and text-to-video generation research. These proposed resources provide a tool for researchers and practitioners interested in multimodal video understanding and generation.

The computational complexity of video models increases linearly with the square number of frames. Thus, constrained bycomputational resources, training video models to learn long-term temporal semantics end-to-end is quite a challenge. Currently, the main-stream method is to split a raw video into clips, leading to incomplete fragmentary temporal information flow and failure of modeling long-term semantics. In this paper, we design the Markov Progressive framework (MaPro), a theoretical framework consisting of the progressive modeling method and a paradigm model tailored for it. Thecore idea of MaPro is to find a paradigm model consisting of proposed Markov operators which can be trained in multiple sequential steps and ensure that the multi-step progressive modeling is equivalent to the conventional end-to-endmodeling. By training the paradigm model under the progressive method, we are able to model long videos end-to-endwith limited resources and ensure the effective transmission of long-term temporal information. We provide implementations of this theoretical system on the mainstream CNN- and Transformer-based models, where they are modified to conform to the Markov paradigm. As a general and robust training method, we experimentally demonstrate that it yields significant performance improvements on different backbones and datasets. As an illustrative example, the proposed method improves the SlowOnly network by 4.1 mAP on Charades and 2.5 top-1 accuracy on Kinetics. And for TimeSformer, MaPro improves its performance on Kinetics by 2.0 top-1 accuracy. Importantly, all these improvements areachieved with a little parameter and computation overhead.

Video Foundation Models (VFMs) have received limited exploration due to high computational costs and data scarcity. Previous VFMs rely on Image Foundation Models (IFMs), which face challenges in transferring to the video domain. Although VideoMAE has trained a robust ViT from limited data, its low-level reconstruction poses convergence difficulties and conflicts with high-level cross-modal alignment. This paper proposes a training-efficient method for temporal-sensitive VFMs that integrates the benefits of existing methods. To increase data efficiency, we mask out most of the low-semantics video tokens, but selectively align the unmasked tokens with IFM, which serves as the UnMasked Teacher (UMT). By providing semantic guidance, our method enables faster convergence and multi-modal friendliness. With a progressive pre-training framework, our model can handle various tasks including scene-related, temporal-related, and complex video-language understanding. Using only public sources for pre-training in 6 days on 32 A100 GPUs, our scratch-built ViT-L/16 achieves state-of-the-art performances on various video tasks.

The prolific performances of Vision Transformers (ViTs) in image tasks have prompted research into adapting the image ViTs for video tasks. However, the substantial gap between image and video impedes the spatiotemporal learning of these image-pretrained models. Though video-specialized models like UniFormer can transfer to the video domain more seamlessly, their unique architectures require prolonged image pretraining, limiting the scalability. Given the emergence of powerful open-source image ViTs, we propose unlocking their potential for video understanding with efficient UniFormer designs. We call the resulting model UniFormerV2, since it inherits the concise style of the Uni-Former block, while redesigning local and global relation aggregators that seamlessly integrate advantages from both ViTs and UniFormer. Our UniFormerV2 achieves state-of-the-art performances on 8 popular video benchmarks, including scene-related Kinetics-400/600/700, heterogeneous Moments in Time, temporal-related Something-Something V1/V2, and untrimmed ActivityNet and HACS. It is note-worthy that to the best of our knowledge, UniFormerV2 is the first to elicit 90% top-1 accuracy on Kinetics-400.

Human activity understanding is of widespread interest in artificial intelligence and spans diverse applications like health care and behavior analysis. Although there have been advances with deep learning, it remains challenging. The object recognition-like solutions usually try to map pixels to semantics directly, but activity patterns are much different from object patterns, thus hindering another success. In this article, we propose a novel paradigm to reformulate this task in two-stage: first mapping pixels to an intermediate space spanned by atomic activity primitives, then programming detected primitives with interpretable logic rules to infer semantics. To afford a representative primitive space, we build a knowledge base including 26+ M primitive labels and logic rules from human priors or automatic discovering. Our framework, Human Activity Knowledge Engine (HAKE), exhibits superior generalization ability and performance upon canonical methods on challenging benchmarks. Code and data are available at http://hake-mvig.cn/.

We study the unsupervised representation learning for the semantic segmentation task. Different from previous works that aim at providing unsupervised pre-trained backbones for segmentation models which need further supervised fine-tune, here, we focus on providing representation that is only trained by unsupervised methods. This means models need to directly generate pixel-level, linearly separable semantic results. We first explore and present two factors that have significant effects on segmentation under the contrastive learning framework: 1) the difficulty and diversity of the positive contrastive pairs, 2) the balance of global and local features. With the intention of optimizing these factors, we propose the cycle-attention contrastive learning (CACL). CACL makes use of semantic continuity of video frames, adopting unsupervised cycle-consistent attention mechanism to implicitly conduct contrastive learning with difficult, global-local-balanced positive pixel pairs. Compared with baseline model MoCo-v2 and other unsupervised methods, CACL demonstrates consistently superior performance on PASCAL VOC (+4.5 mIoU) and Cityscapes (+4.5 mIoU) datasets.

When tracking objects, humans rely on a memory mechanism, memorize the track of an object then look for it in the current scene. In this paper, we propose Memory-based Multi-object Tracking with Transformers (MMTT) to mimic human behavior in multi-object tracking. Unlike Re-ID-based methods, MMTT solves multi-object tracking in an explicit way, with a Track Encoder to extract track memory, a Detection Encoder to extract detection interactions, and a Memory Decoder to simulate the "look" process. The design of MMTT has the ability to model both spatial and temporal in-formation of a single track. We evaluate on commonly used MOT datasets and the experimental results demonstrate its superior effectiveness. We hope this paper can provide a novel direction for the MOT task. The code and models will be made publicly available upon acceptance.

Human attention mechanisms often work in a top-down manner, yet it is not well explored in vision research. Here, we propose the Top-Down Attention Framework (TDAF) to capture top-down attentions, which can be easily adopted in most existing models. The designed Recursive Dual-Directional Nested Structure in it forms two sets of orthogonal paths, recursive and structural ones, where bottom-up spatial features and top-down attention features are extracted respectively. Such spatial and attention features are nested deeply, therefore, the proposed framework works in a mixed top-down and bottom-up manner. Empirical evidence shows that our TDAF can capture effective stratified attention information and boost performance. ResNet with TDAF achieves 2.0% improvements on ImageNet. For object detection, the performance is improved by 2.7% AP over FCOS. For pose estimation, TDAF improves the baseline by 1.6%. And for action recognition, the 3D-ResNet adopting TDAF achieves improvements of 1.7% accuracy.

Convolutional video models have an order of magnitude larger computational complexity than their counter-part image-level models. Constrained by computational resources, there is no model or training method that can train long video sequences end-to-end. Currently, the main-stream method is to split a raw video into clips, leading to incomplete fragmentary temporal information flow. Inspired by natural language processing techniques dealing with long sentences, we propose to treat videos as serial fragments satisfying Markov property, and train it as a whole by progressively propagating information through the temporal dimension in multiple steps. This progressive training (PGT) method is able to train long videos end-to-end with limited resources and ensures the effective transmission of information. As a general and robust training method, we empirically demonstrate that it yields significant performance improvements on different models and datasets. As an illustrative example, the proposed method improves SlowOnly network by 3.7 mAP on Charades and 1.9 top-1 accuracy on Kinetics with negligible parameter and computation overhead. Code is available at: https://github.com/BoPang1996/PGT.

We propose to personalize a 2D human pose estimator given a set of test images of a person without using any manual annotations. While there is a significant advancement in human pose estimation, it is still very challenging for a model to generalize to different unknown environments and unseen persons. Instead of using a fixed model for every test case, we adapt our pose estimator during test time to exploit person-specific information. We first train our model on diverse data with both a supervised and a self-supervised pose estimation objectives jointly. We use a Transformer model to build a transformation between the self-supervised keypoints and the supervised keypoints. During test time, we personalize and adapt our model by fine-tuning with the self-supervised objective. The pose is then improved by transforming the updated self-supervised keypoints. We experiment with multiple datasets and show significant improvements on pose estimations with our self-supervised personalization. Project page with code is available at https://liyz15.github.io/TTP/.

Multi-object tracking is a fundamental vision problem that has been studied for a long time. As deep learning brings excellent performances to object detection algorithms, Tracking by Detection (TBD) has become the mainstream tracking framework. Despite the success of TBD, this two-step method is too complicated to train in an end-to-end manner and induces many challenges as well, such as insufficient exploration of video spatial-temporal information, vulnerability when facing object occlusion, and excessive reliance on detection results. To address these challenges, we propose a concise end-to-end model TubeTK which only needs one step training by introducing the "bounding-tube" to indicate temporal-spatial locations of objects in a short video clip. TubeTK provides a novel direction of multi-object tracking, and we demonstrate its potential to solve the above challenges without bells and whistles. We analyze the performance of TubeTK on several MOT benchmarks and provide empirical evidence to show that TubeTK has the ability to overcome occlusions to some extent without any ancillary technologies like Re-ID. Compared with other methods that adopt private detection results, our one-stage end-to-end model achieves state-of-the-art performances even if it adopts no ready-made detection results. We hope that the proposed TubeTK model can serve as a simple but strong alternative for video-based MOT task. The code and model will be publicly available accompanying this paper.

Human-Object Interaction (HOI) consists of human, object and implicit interaction/verb. Different from previous methods that directly map pixels to HOI semantics, we propose a novel perspective for HOI learning in an analytical manner. In analogy to Harmonic Analysis, whose goal is to study how to represent the signals with the superposition of basic waves, we propose the HOI Analysis. We argue that coherent HOI can be decomposed into isolated human and object. Meanwhile, isolated human and object can also be integrated into coherent HOI again. Moreover, transformations between human-object pairs with the same HOI can also be easier approached with integration and decomposition. As a result, the implicit verb will be represented in the transformation function space. In light of this, we propose an Integration-Decomposition Network (IDN) to implement the above transformations and achieve state-of-the-art performance on widely-used HOI detection benchmarks. Code is available at https://github.com/DirtyHarryLYL/HAKE-Action-Torch/tree/IDN-(Integrating-Decomposing-Network).