Large language model (LLM) agents have emerged as a promising solution for enhancing recommendation systems via user simulation. However, existing studies predominantly resort to prompt-based simulation using frozen LLMs, which frequently results in suboptimal item modeling and user preference learning, thereby ultimately constraining recommendation performance. To address these challenges, we introduce VRAgent-R1, a novel agent-based paradigm that incorporates human-like intelligence in user simulation. Specifically, VRAgent-R1 comprises two distinct agents: the Item Perception (IP) Agent and the User Simulation (US) Agent, designed for interactive user-item modeling. Firstly, the IP Agent emulates human-like progressive thinking based on MLLMs, effectively capturing hidden recommendation semantics in videos. With a more comprehensive multimodal content understanding provided by the IP Agent, the video recommendation system is equipped to provide higher-quality candidate items. Subsequently, the US Agent refines the recommended video sets based on in-depth chain-of-thought (CoT) reasoning and achieves better alignment with real user preferences through reinforcement learning. Experimental results on a large-scale video recommendation benchmark MicroLens-100k have demonstrated the effectiveness of our proposed VRAgent-R1 method, e.g., the IP Agent achieves a 6.0% improvement in NDCG@10, while the US Agent shows approximately 45.0% higher accuracy in user decision simulation compared to state-of-the-art baselines.

Existing video recommendation systems, relying mainly on ID-based embedding mapping and collaborative filtering, often fail to capture in-depth video content semantics. Moreover, most struggle to address biased user behaviors (e.g., accidental clicks, fast skips), leading to inaccurate interest modeling and frequent negative feedback in top recommendations with unclear causes. To tackle this issue, we collect real-world user video-watching sequences, annotate the reasons for users' dislikes, and construct a benchmark dataset for personalized explanations. We then introduce the Agentic Explainable Negative Feedback (ENF) framework, which integrates three core components: (1) the Profile Agent, extracting behavioral cues from users' historical data to derive psychological and personality profiles; (2) the Video Agent, performing comprehensive multimodal video analysis; and (3) the Reason Agent, synthesizing information from the other two agents to predict user engagement and generate explanations. Additionally, we propose the S-GRPO algorithm, enabling the model to progressively address complex tasks during reinforcement fine-tuning. Experimental results on the collected dataset show that our method significantly outperforms state-of-the-art baselines in negative feedback prediction and reason explanation. Notably, it achieves an 8.6% improvement over GPT-4o in reason classification. Deployment on the business platform further validates its benefits: increasing average user watch time by 6.2%, reducing the fast-skip rate by 9.4% , and significantly enhancing user satisfaction.

User feedback is critical for refining recommendation systems, yet explicit feedback (e.g., likes or dislikes) remains scarce in practice. As a more feasible alternative, inferring user preferences from massive implicit feedback has shown great potential (e.g., a user quickly skipping a recommended video usually indicates disinterest). Unfortunately, implicit feedback is often noisy: a user might skip a video due to accidental clicks or other reasons, rather than disliking it. Such noise can easily misjudge user interests, thereby undermining recommendation performance. To address this issue, we propose a novel Group-aware User Behavior Simulation (G-UBS) paradigm, which leverages contextual guidance from relevant user groups, enabling robust and in-depth interpretation of implicit feedback for individual users. Specifically, G-UBS operates via two key agents. First, the User Group Manager (UGM) effectively clusters users to generate group profiles utilizing a ``summarize-cluster-reflect" workflow based on LLMs. Second, the User Feedback Modeler (UFM) employs an innovative group-aware reinforcement learning approach, where each user is guided by the associated group profiles during the reinforcement learning process, allowing UFM to robustly and deeply examine the reasons behind implicit feedback. To assess our G-UBS paradigm, we have constructed a Video Recommendation benchmark with Implicit Feedback (IF-VR). To the best of our knowledge, this is the first multi-modal benchmark for implicit feedback evaluation in video recommendation, encompassing 15k users, 25k videos, and 933k interaction records with implicit feedback. Extensive experiments on IF-VR demonstrate that G-UBS significantly outperforms mainstream LLMs and MLLMs, with a 4.0% higher proportion of videos achieving a play rate > 30% and 14.9% higher reasoning accuracy on IF-VR.

Recent advances in video understanding have been driven by MLLMs. But these MLLMs are good at analyzing short videos, while suffering from difficulties in understanding videos with a longer context. To address this difficulty, several agent paradigms have recently been proposed, using MLLMs as agents for retrieving extra contextual knowledge in a long video. However, most existing agents ignore the key fact that a long video is composed with multiple shots, i.e., to answer the user question from a long video, it is critical to deeply understand its relevant shots like human. Without such insight, these agents often mistakenly find redundant even noisy temporal context, restricting their capacity for long video understanding. To fill this gap, we propose VideoChat-A1, a novel long video agent paradigm. Different from the previous works, our VideoChat-A1 can deeply think with long videos, via a distinct chain-of-shot reasoning paradigm. More specifically, it can progressively select the relevant shots of user question, and look into these shots in a coarse-to-fine partition. By multi-modal reasoning along the shot chain, VideoChat-A1 can effectively mimic step-by-step human thinking process, allowing the interactive discovery of preferable temporal context for thoughtful understanding in long videos. Extensive experiments show that, VideoChat-A1 achieves the state-of-the-art performance on the mainstream long video QA benchmarks, e.g., it achieves 77.0 on VideoMME(w/ subs) and 70.1 on EgoSchema, outperforming its strong baselines (e.g., InternVL2.5-8B and InternVideo2.5-8B), by up to 10.1% and 6.2%. Compared to leading closed-source GPT-4o and Gemini 1.5 Pro, VideoChat-A1 offers competitive accuracy, but only with 7% input frames and 12% inference time on average.

Video temporal grounding (VTG) aims to locate precise segments in videos based on language queries, which is a fundamental challenge in video understanding. While recent Multimodal Large Language Models (MLLMs) have shown promise in tackling VTG through reinforcement learning (RL), they overlook the challenges arising from both the quality and difficulty of training samples. (1) Partially annotated samples. Many manually annotated samples contain relevant segments beyond the annotated interval, introducing ambiguous supervision. (2) Hard-to-ground samples. Samples with poor zero-shot performance produce consistently low and indistinguishable rewards during RL training, exhibiting no clear preference among multiple outputs and thus hindering learning efficiency. To address these challenges, we propose VideoTG-R1, a novel curriculum RL framework with reflected boundary annotations, enabling data-efficient training. Specifically, we propose a Boundary Reflection Agent that utilizes MLLMs to predict query-relevant timestamps outside the annotated intervals, allowing us to identify and filter out partially annotated samples, thereby reducing ambiguity. Furthermore, we introduce a Difficulty Estimation Agent to assess the training difficulty of each sample and design a curriculum RL strategy that dynamically masks the videos of hard-to-ground samples according to the training steps, easing the training difficulty and providing clearer preference. Experiments on the VTG and grounded VideoQA tasks demonstrate the effectiveness of our method. Remarkably, with only 10% of the training samples and 21% of the computational budget, VideoTG-R1 outperforms full-data counterparts under both group relative policy optimization (GRPO) and supervised fine-tuning (SFT). The code is available at https://github.com/ldong1111/VideoTG-R1.

With the prevalence of Multimodal Large Language Models(MLLMs), autonomous driving has encountered new opportunities and challenges. In particular, multi-modal video understanding is critical to interactively analyze what will happen in the procedure of autonomous driving. However, videos in such a dynamical scene that often contains complex spatial-temporal movements, which restricts the generalization capacity of the existing MLLMs in this field. To bridge the gap, we propose a novel Hierarchical Mamba Adaptation (H-MBA) framework to fit the complicated motion changes in autonomous driving videos. Specifically, our H-MBA consists of two distinct modules, including Context Mamba (C-Mamba) and Query Mamba (Q-Mamba). First, C-Mamba contains various types of structure state space models, which can effectively capture multi-granularity video context for different temporal resolution. Second, Q-Mamba flexibly transforms the current frame as the learnable query, and attentively select multi-granularity video context into query. Consequently, it can adaptively integrate all the video contexts of multi-scale temporal resolutions to enhance video understanding. Via a plug-and-play paradigm in MLLMs, our H-MBA shows the remarkable performance on multi-modal video tasks in autonomous driving, e.g., for risk object detection, it outperforms the previous SOTA method with 5.5% mIoU improvement.

Despite recent advancements in text-to-image generation, most existing methods struggle to create images with multiple objects and complex spatial relationships in the 3D world. To tackle this limitation, we introduce a generic AI system, namely MUSES, for 3D-controllable image generation from user queries. Specifically, our MUSES develops a progressive workflow with three key components, including (1) Layout Manager for 2D-to-3D layout lifting, (2) Model Engineer for 3D object acquisition and calibration, (3) Image Artist for 3D-to-2D image rendering. By mimicking the collaboration of human professionals, this multi-modal agent pipeline facilitates the effective and automatic creation of images with 3D-controllable objects, through an explainable integration of top-down planning and bottom-up generation. Additionally, existing benchmarks lack detailed descriptions of complex 3D spatial relationships of multiple objects. To fill this gap, we further construct a new benchmark of T2I-3DisBench (3D image scene), which describes diverse 3D image scenes with 50 detailed prompts. Extensive experiments show the state-of-the-art performance of MUSES on both T2I-CompBench and T2I-3DisBench, outperforming recent strong competitors such as DALL-E 3 and Stable Diffusion 3. These results demonstrate a significant step forward for MUSES in bridging natural language, 2D image generation, and 3D world.

Existing multimodal generative models fall short as qualified design copilots, as they often struggle to generate imaginative outputs once instructions are less detailed or lack the ability to maintain consistency with the provided references. In this work, we introduce WeGen, a model that unifies multimodal generation and understanding, and promotes their interplay in iterative generation. It can generate diverse results with high creativity for less detailed instructions. And it can progressively refine prior generation results or integrating specific contents from references following the instructions in its chat with users. During this process, it is capable of preserving consistency in the parts that the user is already satisfied with. To this end, we curate a large-scale dataset, extracted from Internet videos, containing rich object dynamics and auto-labeled dynamics descriptions by advanced foundation models to date. These two information are interleaved into a single sequence to enable WeGen to learn consistency-aware generation where the specified dynamics are generated while the consistency of unspecified content is preserved aligned with instructions. Besides, we introduce a prompt self-rewriting mechanism to enhance generation diversity. Extensive experiments demonstrate the effectiveness of unifying multimodal understanding and generation in WeGen and show it achieves state-of-the-art performance across various visual generation benchmarks. These also demonstrate the potential of WeGen as a user-friendly design copilot as desired. The code and models will be available at https://github.com/hzphzp/WeGen.

Current multimodal large language models (MLLMs) struggle with fine-grained or precise understanding of visuals although they give comprehensive perception and reasoning in a spectrum of vision applications. Recent studies either develop tool-using or unify specific visual tasks into the autoregressive framework, often at the expense of overall multimodal performance. To address this issue and enhance MLLMs with visual tasks in a scalable fashion, we propose Task Preference Optimization (TPO), a novel method that utilizes differentiable task preferences derived from typical fine-grained visual tasks. TPO introduces learnable task tokens that establish connections between multiple task-specific heads and the MLLM. By leveraging rich visual labels during training, TPO significantly enhances the MLLM’s multimodal capabilities and task-specific performance. Through multi-task co-training within TPO, we observe synergistic benefits that elevate individual task performance beyond what is achievable through single-task training methodologies. Our instantiation of this approach with VideoChat and LLaVA demonstrates an overall 14.6% improvement in multimodal performance compared to baseline models. Additionally, MLLM-TPO demonstrates robust zero-shot capabilities across various tasks, performing comparably to state-of-the-art supervised models.

Despite the recent advancement in video stylization, most existing methods struggle to render any video with complex transitions, based on an open style description of user query. To fill this gap, we introduce a generic multi-agent system for video stylization, V-Stylist, by a novel collaboration and reflection paradigm of multi-modal large language models. Specifically, our V-Stylist is a systematical workflow with three key roles: (1) Video Parser decomposes the input video into a number of shots and generates their text prompts of key shot content. Via a concise video-to-shot prompting paradigm, it allows our V-Stylist to effectively handle videos with complex transitions. (2) Style Parser identifies the style in the user query and progressively search the matched style model from a style tree. Via a robust tree-of-thought searching paradigm, it allows our V-Stylist to precisely specify vague style preference in the open user query. (3) Style Artist leverages the matched model to render all the video shots into the required style. Via a novel multi-round self-reflection paradigm, it allows our V-Stylist to adaptively adjust detail control, according to the style requirement. With such a distinct design of mimicking human professionals, our V-Stylist achieves a major breakthrough over the primary challenges for effective and automatic video stylization. Moreover, we further construct a new benchmark Text-driven Video Stylization Benchmark (TVSBench), which fills the gap to assess various stylization of complex videos on open user queries. Extensive experiments show that, V-Stylist achieves the state-of-the-art, e.g.,V-Stylist surpasses FRESCO and ControlVideo by 6.05% and 4.51% respectively in overall average metrics, marking a significant advance in video stylization.

Existing MLLMs encounter significant challenges in modeling the temporal context within long videos. Currently, mainstream Agent-based methods use external tools to assist a single MLLM in answering long video questions. Despite such tool-based support, a solitary MLLM still offers only a partial understanding of long videos, resulting in limited performance. In order to better address long video tasks, we introduce LVAgent, the first framework enabling multi-round dynamic collaboration of MLLM agents in long video understanding. Our method consists of four key steps: 1) Selection: We pre-select appropriate agents from the model library to form optimal agent teams based on different tasks. 2) Perception: We design an effective retrieval scheme for long videos to improve the coverage of critical temporal segments while maintaining computational efficiency. 3) Action: Agents answer long video questions and exchange reasons. 4) Reflection: We evaluate each agent's performance in each round of discussion and optimize the agent team for dynamic collaboration. The agents iteratively refine their answers by multi-round dynamical collaboration of MLLM agents. LVAgent is the first agent system method that outperforms all closed-source models (like GPT-4o) and open-source models (like InternVL-2.5 and Qwen2-VL) in the long video understanding tasks. Our LVAgent achieves an accuracy of 80\% on four mainstream long video understanding tasks. Notably, LVAgent improves accuracy by 13.3\% on LongVideoBench. Code is available at https://github.com/64327069/LVAgent.

We present VRBench, the first long narrative video benchmark crafted for evaluating large models' multi-step reasoning capabilities, addressing limitations in existing evaluations that overlook temporal reasoning and procedural validity. It comprises 960 long videos (with an average duration of 1.6 hours), along with 8,243 human-labeled multi-step question-answering pairs and 25,106 reasoning steps with timestamps. These videos are curated via a multi-stage filtering process including expert inter-rater reviewing to prioritize plot coherence. We develop a human-AI collaborative framework that generates coherent reasoning chains, each requiring multiple temporally grounded steps, spanning seven types (e.g., event attribution, implicit inference). VRBench designs a multi-phase evaluation pipeline that assesses models at both the outcome and process levels. Apart from the MCQs for the final results, we propose a progress-level LLM-guided scoring metric to evaluate the quality of the reasoning chain from multiple dimensions comprehensively. Through extensive evaluations of 12 LLMs and 19 VLMs on VRBench, we undertake a thorough analysis and provide valuable insights that advance the field of multi-step reasoning.

The existing video understanding benchmarks for multimodal large language models (MLLMs) mainly focus on short videos. The few benchmarks for long video understanding often rely on multiple-choice questions (MCQs). Due to the limitations of MCQ evaluations and the advanced reasoning abilities of MLLMs, models can often answer correctly by combining short video insights with elimination, without truly understanding the content. To bridge this gap, we introduce CG-Bench, a benchmark for clue-grounded question answering in long videos. CG-Bench emphasizes the model's ability to retrieve relevant clues, enhancing evaluation credibility. It includes 1,219 manually curated videos organized into 14 primary, 171 secondary, and 638 tertiary categories, making it the largest benchmark for long video analysis. The dataset features 12,129 QA pairs in three question types: perception, reasoning, and hallucination. To address the limitations of MCQ-based evaluation, we develop two novel clue-based methods: clue-grounded white box and black box evaluations, assessing whether models generate answers based on accurate video understanding. We evaluated multiple closed-source and open-source MLLMs on CG-Bench. The results show that current models struggle significantly with long videos compared to short ones, and there is a notable gap between open-source and commercial models. We hope CG-Bench will drive the development of more reliable and capable MLLMs for long video comprehension.

In egocentric video understanding, the motion of hands and objects as well as their interactions play a significant role by nature. However, existing egocentric video representation learning methods mainly focus on aligning video representation with high-level narrations, overlooking the intricate dynamics between hands and objects. In this work, we aim to integrate the modeling of fine-grained hand-object dynamics into the video representation learning process. Since no suitable data is available, we introduce HOD, a novel pipeline employing a hand-object detector and a large language model to generate high-quality narrations with detailed descriptions of hand-object dynamics. To learn these fine-grained dynamics, we propose EgoVideo, a model with a new lightweight motion adapter to capture fine-grained hand-object motion information. Through our co-training strategy, EgoVideo effectively and efficiently leverages the fine-grained hand-object dynamics in the HOD data. Extensive experiments demonstrate that our method achieves state-of-the-art performance across multiple egocentric downstream tasks, including improvements of 6.3% in EK-100 multi-instance retrieval, 5.7% in EK-100 classification, and 16.3% in EGTEA classification in zero-shot settings. Furthermore, our model exhibits robust generalization capabilities in hand-object interaction and robot manipulation tasks.

Creating high-quality data for training robust language-instructed agents is a long-lasting challenge in embodied AI. In this paper, we introduce a Self-Refining Data Flywheel (SRDF) that generates high-quality and large-scale navigational instruction-trajectory pairs by iteratively refining the data pool through the collaboration between two models, the instruction generator and the navigator, without any human-in-the-loop annotation. Specifically, SRDF starts with using a base generator to create an initial data pool for training a base navigator, followed by applying the trained navigator to filter the data pool. This leads to higher-fidelity data to train a better generator, which can, in turn, produce higher-quality data for training the next-round navigator. Such a flywheel establishes a data self-refining process, yielding a continuously improved and highly effective dataset for large-scale language-guided navigation learning. Our experiments demonstrate that after several flywheel rounds, the navigator elevates the performance boundary from 70\% to 78\% SPL on the classic R2R test set, surpassing human performance (76\%) for the first time. Meanwhile, this process results in a superior generator, evidenced by a SPICE increase from 23.5 to 26.2, better than all previous VLN instruction generation methods. Finally, we demonstrate the scalability of our method through increasing environment and instruction diversity, and the generalization ability of our pre-trained navigator across various downstream navigation tasks, surpassing state-of-the-art methods by a large margin in all cases.

Multimodal Large Language Models (MLLMs) have demonstrated impressive performance in short video understanding. However, understanding long-form videos still remains challenging for MLLMs. This paper proposes TimeSuite, a collection of new designs to adapt the existing short-form video MLLMs for long video understanding, including a simple yet efficient framework to process long video sequence, a high-quality video dataset for grounded tuning of MLLMs, and a carefully-designed instruction tuning task to explicitly incorporate the grounding supervision in the traditional QA format. Specifically, based on VideoChat, we propose our long-video MLLM, coined as VideoChat-T, by implementing a token shuffling to compress long video tokens and introducing Temporal Adaptive Position Encoding (TAPE) to enhance the temporal awareness of visual representation. Meanwhile, we introduce the TimePro, a comprehensive grounding-centric instruction tuning dataset composed of 9 tasks and 349k high-quality grounded annotations. Notably, we design a new instruction tuning task type, called Temporal Grounded Caption, to peform detailed video descriptions with the corresponding time stamps prediction. This explicit temporal location prediction will guide MLLM to correctly attend on the visual content when generating description, and thus reduce the hallucination risk caused by the LLMs. Experimental results demonstrate that our TimeSuite provides a successful solution to enhance the long video understanding capability of short-form MLLM, achieving improvement of 5.6% and 6.8% on the benchmarks of Egoschema and VideoMME, respectively. In addition, VideoChat-T exhibits robust zero-shot temporal grounding capabilities, significantly outperforming the existing state-of-the-art MLLMs. After fine-tuning, it performs on par with the traditional supervised expert models.

Diffusion models have driven the advancement of vision generation over the past years. However, it is often difficult to apply these large models in downstream tasks, due to massive fine-tuning cost. Recently, Low-Rank Adaptation (LoRA) has been applied for efficient tuning of diffusion models. Unfortunately, the capabilities of LoRA-tuned diffusion models are limited, since the same LoRA is used for different timesteps of the diffusion process. To tackle this problem, we introduce a general and concise TimeStep Master (TSM) paradigm with two key fine-tuning stages. In the fostering stage (1-stage), we apply different LoRAs to fine-tune the diffusion model at different timestep intervals. This results in different TimeStep LoRA experts that can effectively capture different noise levels. In the assembling stage (2-stage), we design a novel asymmetrical mixture of TimeStep LoRA experts, via core-context collaboration of experts at multi-scale intervals. For each timestep, we leverage TimeStep LoRA expert within the smallest interval as the core expert without gating, and use experts within the bigger intervals as the context experts with time-dependent gating. Consequently, our TSM can effectively model the noise level via the expert in the finest interval, and adaptively integrate contexts from the experts of other scales, boosting the versatility of diffusion models. To show the effectiveness of our TSM paradigm, we conduct extensive experiments on three typical and popular LoRA-related tasks of diffusion models, including domain adaptation, post-pretraining, and model distillation. Our TSM achieves the state-of-the-art results on all these tasks, throughout various model structures (UNet, DiT and MM-DiT) and visual data modalities (Image, Video), showing its remarkable generalization capacity.

Inducing reasoning in multimodal large language models (MLLMs) is critical for achieving human-level perception and understanding. Existing methods mainly leverage LLM reasoning to analyze parsed visuals, often limited by static perception stages. This paper introduces Visual Test-Time Scaling (VTTS), a novel approach to enhance MLLMs' reasoning via iterative perception during inference. VTTS mimics humans' hierarchical attention by progressively refining focus on high-confidence spatio-temporal regions, guided by updated textual predictions. Specifically, VTTS employs an Iterative Perception (ITP) mechanism, incorporating reinforcement learning with spatio-temporal supervision to optimize reasoning. To support this paradigm, we also present VTTS-80K, a dataset tailored for iterative perception. These designs allows a MLLM to enhance its performance by increasing its perceptual compute. Extensive experiments validate VTTS's effectiveness and generalization across diverse tasks and benchmarks. Our newly introduced Videochat-R1.5 model has achieved remarkable improvements, with an average increase of over 5\%, compared to robust baselines such as Qwen2.5VL-3B and -7B, across more than 15 benchmarks that encompass video conversation, video reasoning, and spatio-temporal perception.

Abstract Audio-Visual Question Answering (AVQA) requires the model to answer questions with complex dynamic audio-visual information. Prior works on this task mainly consider only using single question-answer pairs during training, overlooking the rich semantic associations between questions. In this work, we propose a novel Collective Question-Guided Network (CoQo), which accepts multiple question-answer pairs as input and leverages the reasoning over these questions to assist the model training process. The core module is the proposed Question Guided Transformer (QGT), which uses collective question reasoning to perform question-guided feature extraction. Since multiple question-answer pairs are not always available, especially during inference, our QGT uses a set of learnable tokens to learn the collective information from multiple questions during training. At inference time, these learnable tokens bring additional reasoning information even when only one question is used as input. We employ QGT in both spatial and temporal dimensions to extract question-related features effectively and efficiently. To better capture detailed audio-visual associations, we train the model in a finer level by distinguishing feature pairs of different questions within the same video. Extensive experiments demonstrate that our method can achieve state-of-the-art performance on three AVQA datasets while reducing training time significantly. We also observe strong performances of our method on three VQA benchmarks. Detailed ablation studies further confirm the effectiveness of our proposed collective question reasoning scheme, both quantitatively and qualitatively.

Open-world video recognition is challenging since traditional networks are not generalized well on complex environment variations. Alternatively, foundation models with rich knowledge have recently shown their generalization power. However, how to apply such knowledge has not been fully explored for open-world video recognition. To this end, we propose a generic knowledge transfer pipeline, which progressively exploits and integrates external multimodal knowledge from foundation models to boost open-world video recognition. We name it PCA , based on three stages of P ercept, C hat, and A dapt. First, we perform Percept process to reduce the video domain gap and obtain external visual knowledge. Second, we generate rich linguistic semantics as external textual knowledge in Chat stage. Finally, we blend external multimodal knowledge in Adapt stage, by inserting multimodal knowledge adaptation modules into networks. We conduct extensive experiments on three challenging open-world video benchmarks, i.e., TinyVIRAT, ARID, and QV-Pipe. Our approach achieves state-of-the-art performance on all three datasets.

The task of weakly supervised temporal sentence grounding (WSTSG) aims to detect temporal intervals corresponding to a language description from untrimmed videos with only video-level video-language correspondence. For an anchor sample, most existing approaches generate negative samples either from other videos or within the same video for contrastive learning. However, some training samples are highly similar to the anchor sample, directly regarding them as negative samples leads to difficulties for optimization and ignores the correlations between these similar samples and the anchor sample. To address this, we propose Positive Sample Mining (PSM), a novel framework that mines positive samples from the training set to provide more discriminative supervision. Specifically, for a given anchor sample, we partition the remaining training set into semantically similar and dissimilar subsets based on the similarity of their text queries. To effectively leverage these correlations, we introduce a PSM-guided contrastive loss to ensure that the anchor proposal is closer to similar samples and further from dissimilar ones. Additionally, we design a PSM-guided rank loss to ensure that similar samples are closer to the anchor proposal than to the negative intra-video proposal, aiming to distinguish the anchor proposal and the negative intra-video proposal. Experiments on the WSTSG and grounded VideoQA tasks demonstrate the effectiveness and superiority of our method.

3D perception is a critical problem in autonomous driving. Recently, the Bird’s-Eye-View (BEV) approach has attracted extensive attention, due to low-cost deployment and desirable vision detection capacity. However, the existing models ignore a realistic scenario during the driving procedure, i.e., one or more view cameras may be failed, which largely deteriorates their performance. To tackle this problem, we propose a generic Masked BEV (M-BEV) perception framework, which can effectively improve robustness to this challenging scenario, by random masking and reconstructing camera views in the end-to-end training. More specifically, we develop a novel Masked View Reconstruction (MVR) module in our M-BEV. It mimics various missing cases by randomly masking features of different camera views, then leverages the original features of these views as self-supervision and reconstructs the masked ones with the distinct spatio-temporal context across camera views. Via such a plug-and-play MVR, our M-BEV is capable of learning the missing views from the resting ones, and thus well generalized for robust view recovery and accurate perception in the testing. We perform extensive experiments on the popular NuScenes benchmark, where our framework can significantly boost 3D perception performance of the state-of-the-art models on various missing view cases, e.g., for the absence of back view, our M-BEV promotes the PETRv2 model with 10.3% mAP gain.

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.

Being able to map the activities of others into one's own point of view is a fundamental human skill even from a very early age. Taking a step toward understanding this human ability, we introduce EgoExoLearn, a large-scale dataset that emulates the human demonstration following process, in which individuals record egocentric videos as they execute tasks guided by exocentric-view demonstration videos. Focusing on the potential applications in daily assistance and professional support, EgoExoLearn contains egocentric and demonstration video data spanning 120 hours captured in daily life scenarios and specialized laboratories. Along with the videos we record high-quality gaze data and provide detailed multimodal annotations, formulating a playground for modeling the human ability to bridge asynchronous procedural actions from different viewpoints. To this end, we present benchmarks such as crossview association, cross-view action planning, and crossview referenced skill assessment, along with detailed analysis. We expect EgoExoLearn can serve as an important resource for bridging the actions across views, thus paving the way for creating AI agents capable of seamlessly learning by observing humans in the real world. The dataset and benchmark codes are available at https://github.com/OpenGVLab/EgoExoLearn.

In this work, we present Vlogger, a generic AI systemfor generating a minute-level video blog (i.e., vlog) of user de-scriptions. Different from short videos with a few seconds, vlog often contains a complex storyline with diversified scenes, which is challenging for most existing video generation approaches. To break through this bottleneck, our Vlogger smartly leverages Large Language Model (LLM) as Director and decomposes a long video generation task of vlog into four key stages, where we invoke various foundation models to play the critical roles of vlog profession-als, including (1) Script, (2) Actor, (3) ShowMaker, and (4) Voicer. With such a design of mimicking human beings, our Vlogger can generate vlogs through explainable cooperation of top-down planning and bottom-up shooting. More-over, we introduce a novel video diffusion model, Show-Maker, which serves as a videographer in our Vlogger for generating the video snippet of each shooting scene. By incorporating Script and Actor attentively as textual and visual prompts, it can effectively enhance spatial-temporal coherence in the snippet. Besides, we design a concise mixed training paradigm for ShowMaker, boosting its ca-pacity for both T2V generation and prediction. Finally, the extensive experiments show that our method achieves state-of-the-art performance on zero-shot T2V generation and prediction tasks. More importantly, Vlogger can generate over 5-minute vlogs from open-world descriptions, without loss of video coherence on script and actor.

Recently video generation has achieved substantial progress with realistic results. Nevertheless, existing AI-generated videos are usually very short clips ("shot-level'') depicting a single scene. To deliver a coherent long video ("story-level''), it is desirable to have creative transition and prediction effects across different clips. This paper presents a short-to-long video diffusion model, SEINE, that focuses on generative transition and prediction. The goal is to generate high-quality long videos with smooth and creative transitions between scenes and varying lengths of shot-level videos. Specifically, we propose a random-mask video diffusion model to automatically generate transitions based on textual descriptions. By providing the images of different scenes as inputs, combined with text-based control, our model generates transition videos that ensure coherence and visual quality. Furthermore, the model can be readily extended to various tasks such as image-to-video animation and autoregressive video prediction. To conduct a comprehensive evaluation of this new generative task, we propose three assessing criteria for smooth and creative transition: temporal consistency, semantic similarity, and video-text semantic alignment. Extensive experiments validate the effectiveness of our approach over existing methods for generative transition and prediction, enabling the creation of story-level long videos.

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.

Large Vision-Language Models (LVLMs) show significant strides in general-propose multimodal applications such as visual dialogue and embodied navigation. However, existing multimodal evaluation benchmarks cover a limited number of multimodal tasks testing rudimentary capabilities, falling short in tracking LVLM development. In this study, we present MMT-Bench, a comprehensive benchmark designed to assess LVLMs across massive multimodal tasks requiring expert knowledge and deliberate visual recognition, localization, and reasoning. MMT-Bench comprises $31,325$ meticulously curated multi-choice visual questions from various multimodal scenarios such as vehicle driving and embodied navigation, covering $32$ core meta-tasks and $162$ subtasks in multimodal understanding. Due to its extensive task coverage, MMT-Bench enables the evaluation of LVLMs using a task map, facilitating the discovery of in- and out-of-domain tasks. Evaluation results involving $20$ publicly available LVLMs such as the proprietary GeminiProVision model, underscore the significant challenges posed by MMT-Bench. We anticipate that MMT-Bench will inspire the community to develop next-generation multimodal foundation models aimed at achieving general-purpose multimodal intelligence.

Vision-language foundation models (such as CLIP) have recently shown their power in transfer learning, owing to large-scale image-text pre-training. However, target domain data in the downstream tasks can be highly different from the pre-training phase, which makes it hard for such a single model to generalize well. Alternatively, there exists a wide range of expert models that contain diversified vision and/or language knowledge pre-trained on different modalities, tasks, networks, and datasets. Unfortunately, these models are "isolated agents" with heterogeneous structures, and how to integrate their knowledge for generalizing CLIP-like models has not been fully explored. To bridge this gap, we propose a general and concise TransAgent framework, which transports the knowledge of the isolated agents in a unified manner, and effectively guides CLIP to generalize with multi-source knowledge distillation. With such a distinct framework, we flexibly collaborate with 11 heterogeneous agents to empower vision-language foundation models, without further cost in the inference phase. Finally, our TransAgent achieves state-of-the-art performance on 11 visual recognition datasets. Under the same low-shot setting, it outperforms the popular CoOp with around 10\% on average, and 20\% on EuroSAT which contains large domain shifts.

In this paper, we focus on the weakly supervised video object detection problem, where each training video is only tagged with object labels, without any bounding box annotations of objects. To effectively train object detectors from such weakly-annotated videos, we propose a Progressive Frame-Proposal Mining (PFPM) framework by exploiting discriminative proposals in a coarse-to-fine manner. First, we design a flexible Multi-Level Selection (MLS) scheme, with explicit guidance of video tags. By selecting object-relevant frames and mining important proposals from these frames, the proposed MLS can effectively reduce frame redundancy as well as improve proposal effectiveness to boost weakly-supervised detectors. Moreover, we develop a novel Holistic-View Refinement (HVR) scheme, which can globally evaluate importance of proposals among frames, and thus correctly refine pseudo ground truth boxes for training video detectors in a self-supervised manner. Finally, we evaluate the proposed PFPM on a large-scale benchmark for video object detection, on ImageNet VID, under the setting of weak annotations. The experimental results demonstrate that our PFPM significantly outperforms the state-of-the-art weakly-supervised detectors.

The recent advance of video retrieval has been driven by large-scale visual-language pretraining models. In particular, the state-of-the-art approaches are mainly based on temporal extension of the well-known CLIP model. However, they ignore a critical problem in video retrieval, i.e., the text often refers to a small snippet in the corresponding video. Blindly aggregating all the frames inevitably reduces the discriminative capacity of the final video token to match the text token. Hence, these approaches are limited to retrieve complex videos with diversified contents. To tackle this problem, we propose a concise and novel Attentive Snippet Prompting (ASP) framework, which can dynamically exploit the text-relevant video snippet to boost retrieval. Specifically, our ASP consists of two simple but effective modules, i.e., snippet prompting and video aggregating. Given a pair of text and video, snippet prompting can smartly use cross-modal attention to construct a text-driven visual prompt, namely attentive snippet token, which adaptively describes the relevant video snippet of the text query. Alternatively, video aggregating can summarize all the frame tokens as a video token, for providing the global context. With cooperation of attentive snippet token and global video token, our ASP can effectively learn a robust and text-relevant visual representation for video retrieval. Finally, we evaluate our ASP framework on the widely-used benchmarks, where it simply outperforms a number of recent approaches with a large margin.

Discovering temporal boundary is critical for untrimmed video tasks, such as temporal sentence grounding and action detection. Due to the labor-intensive boundary annotations, the recent studies focus on the weakly-supervised setting, with only sentences or action tags in the training videos. However, how to align temporal boundaries and textual descriptions is problematic in most weakly-supervised approaches. To alleviate this difficulty, we propose a novel Dual Masked Modeling (DM2) framework, which can effectively enhance clip-text alignment to boost temporal boundary discovery, by cross-modal masked modeling in the dual fashion. Specifically, we introduce two coupled reconstruction branches, i.e., Clip-Aware Masked Text Modeling (C-MTM), and Text-Aware Masked Clip Modeling (T-MCM), after generating a temporal proposal of the underlying clip. In C-MTM, we recover the masked text with visual assistance of the clip proposal. In T-MCM, we recover the masked clip proposal with lingual assistance of the text. Via such complementary reconstruction supervision, our DM2 can cooperatively exploit robust matching between the video clip and the referred text, allowing to unify grounding and localization in a concise manner. Finally, we perform extensive experiments on the popular temporal benchmarks, i.e., Charades-STA, ActivityNet Captions, ActivityNet-v1.3 and THUMOS-14. Our DM2 achieves state-of-the-art for both weakly-supervised temporal grounding and localization. Codes and models will be released afterward.

Self-supervised learning has not been extensively investigated in the context of point cloud analysis. Current frameworks are predominantly rely on point cloud reconstruction. Given only 3D coordinates, such approaches tend to learn local geometric structures and contours but struggle to comprehend high-level semantic content. Consequently, they achieve unsatisfactory performance in downstream tasks such as classification, segmentation, etc. To fill this gap, we propose a generic Contour-Perturbed Reconstruction Network (CP-Net), which can effectively guides self-supervised reconstruction to learn semantic content in the point cloud, and thus promote discriminative power of point cloud representation. Initially, we introduce a concise contour-perturbed augmentation module for point cloud reconstruction. With guidance of geometry disentangling, we divide point cloud into contour and content components. Subsequently, we perturb the contour components and preserve the content components on the point cloud. As a result, self supervisor can effectively focus on semantic content, by reconstructing the original point cloud from such perturbed one. Next, we use this perturbed reconstruction as an assistant branch, to guide the learning of basic reconstruction branch via a distinct dual-branch consistency loss. In this case, our CP-Net not only captures structural contour but also learn semantic content for discriminative downstream tasks. Finally, we perform extensive experiments on a number of point cloud benchmarks. Part segmentation results demonstrate that our CP-Net (81.5% of mean Intersection over union) outperforms the previous self-supervised models, and narrows the gap with the fully-supervised methods. For classification, we get a competitive result with the fully-supervised methods on ModelNet40 (92.5% accuracy) and ScanObjectNN (87.9% accuracy). Our code is available athttps://github.com/MingyeXu/cp-net

Scale is the primary factor for building a powerful foundation model that could well generalize to a variety of downstream tasks. However, it is still challenging to train video foundation models with billions of parameters. This paper shows that video masked autoencoder (VideoMAE) is a scalable and general self-supervised pre-trainer for building video foundation models. We scale the VideoMAE in both model and data with a core design. Specifically, we present a dual masking strategy for efficient pre-training, with an encoder operating on a subset of video tokens and a decoder processing another subset of video tokens. Although VideoMAE is very efficient due to high masking ratio in encoder, masking decoder can still further reduce the overall computational cost. This enables the efficient pre-training of billion-level models in video. We also use a progressive training paradigm that involves an initial pre-training on a diverse multi-sourced unlabeled dataset, followed by a post-pre-training on a mixed labeled dataset. Finally, we successfully train a video ViT model with a billion parameters, which achieves a new state-of-the-art performance on the datasets of Kinetics (90.0% on K400 and 89.9% on K600) and Something-Something (68.7% on V1 and 77.0% on V2). In addition, we extensively verify the pre-trained video ViT models on a variety of downstream tasks, demonstrating its effectiveness as a general video representation learner.

Masked Modeling (MM) has demonstrated widespread success in various vision challenges, by reconstructing masked visual patches. Yet, applying MM for large-scale 3D scenes remains an open problem due to the data sparsity and scene complexity. The conventional random masking paradigm used in 2D images often causes a high risk of ambiguity when recovering the masked region of 3D scenes. To this end, we propose a novel informative-preserved reconstruction, which explores local statistics to discover and preserve the representative structured points, effectively enhancing the pretext masking task for 3D scene understanding. Integrated with a progressive reconstruction manner, our method can concentrate on modeling regional geometry and enjoy less ambiguity for masked reconstruction. Besides, such scenes with progressive masking ratios can also serve to self-distill their intrinsic spatial consistency, requiring to learn the consistent representations from unmasked areas. By elegantly combining informative-preserved reconstruction on masked areas and consistency self-distillation from unmasked areas, a unified framework called MM-3DScene is yielded. We conduct comprehensive experiments on a host of downstream tasks. The consistent improvement (e.g., +6.1% [email protected] on object detection and +2.2% mIoU on semantic segmentation) demonstrates the superiority of our approach.

We present a Non-parametric Network for 3D point cloud analysis, Point-NN, which consists of purely non-learnable components: farthest point sampling (FPS), k-nearest neighbors (k-NN), and pooling operations, with trigonometric functions. Surprisingly, it performs well on various 3D tasks, requiring no parameters or training, and even surpasses existing fully trained models. Starting from this basic non-parametric model, we propose two extensions. First, Point-NN can serve as a base architectural framework to construct Parametric Networks by simply inserting linear layers on top. Given the superior non-parametric foundation, the derived Point-PN exhibits a high performance-efficiency tradeoff with only a few learnable parameters. Second, Point-NN can be regarded as a plug-and-play module for the already trained 3D models during inference. Point-NN captures the complementary geometric knowledge and enhances existing methods for different 3D benchmarks without retraining. We hope our work may cast a light on the community for understanding 3D point clouds with non-parametric methods. Code is available at https://github.com/ZrrSkywalker/Point-NN.

Referring Video Object Segmentation (RVOS) is to segment the object instance from a given video, according to the textual description of this object. However, in the open world, the object descriptions are often diversified in contents and flexible in lengths. This leads to the key difficulty in RVOS, i.e., various descriptions of different objects are corresponding to different temporal scales in the video, which is ignored by most existing approaches with single stride of frame sampling. To tackle this problem, we propose a concise Hybrid Temporal-scale Multimodal Learning (HTML) framework, which can effectively align lingual and visual features to discover core object semantics in the video, by learning multimodal interaction hierarchically from different temporal scales. More specifically, we introduce a novel inter-scale multimodal perception module, where the language queries dynamically interact with visual features across temporal scales. It can effectively reduce complex object confusion by passing video context among different scales. Finally, we conduct extensive experiments on the widely used benchmarks, including Ref-Youtube-VOS, Ref-DAVIS17, A2D-Sentences and JHMDB-Sentences, where our HTML achieves state-of-the-art performance on all these datasets.

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.

Open set action recognition (OSAR) is a challenging task that requires a classifier to identify actions that do not belong to any of the classes in its training set. Existing methods employ the Evidential Neural Network (ENN) as an open-set classifier, which is trained in a supervised manner on feature representations from known classes to quantify the predictive uncertainty of human actions. In this paper, we propose a novel framework for OSAR that enriches the discriminative representation from a backbone with a reconstructive one to further improve performance. Our approach involves augmenting the input features with their reconstruction obtained from a reconstruction-based model in unsupervised training on known classes. We then use the correspondence between the two features to learn the open-set classifier, forcing it to associate low correspondence both when the feature is from unknown classes as well as when the input feature and its reconstruction variant are inconsistent with each other. Our experimental results on standard OSAR benchmarks demonstrate that our end-to-end trained model significantly outperforms state-of-the-art methods. Our proposed approach shows the effectiveness of combining discriminative and reconstructive representations for OSAR.

Robustness and generalization are two challenging problems for learning point cloud representation. To tackle these problems, we first design a novel geometry coding model, which can effectively use an invariant eigengraph to group points with similar geometric information, even when such points are far from each other. We also introduce a large-scale point cloud dataset, PCNet184. It consists of 184 categories and 51,915 synthetic objects, which brings new challenges for point cloud classification, and provides a new benchmark to assess point cloud cross-domain generalization. Finally, we perform extensive experiments on point cloud classification, using ModelNet40, ScanObjectNN, and our PCNet184, and segmentation, using ShapeNetPart and S3DIS. Our method achieves comparable performance to state-of-the-art methods on these datasets, for both supervised and unsupervised learning. Code and our dataset are available at https://github.com/MingyeXu/PCNet184 .

It is a challenging task to learn discriminative representation from images and videos, due to large local redundancy and complex global dependency in these visual data. Convolution neural networks (CNNs) and vision transformers (ViTs) have been two dominant frameworks in the past few years. Though CNNs can efficiently decrease local redundancy by convolution within a small neighborhood, the limited receptive field makes it hard to capture global dependency. Alternatively, ViTs can effectively capture long-range dependency via self-attention, while blind similarity comparisons among all the tokens lead to high redundancy. To resolve these problems, we propose a novel Unified transFormer (UniFormer), which can seamlessly integrate the merits of convolution and self-attention in a concise transformer format. Different from the typical transformer blocks, the relation aggregators in our UniFormer block are equipped with local and global token affinity respectively in shallow and deep layers, allowing tackling both redundancy and dependency for efficient and effective representation learning. Finally, we flexibly stack our blocks into a new powerful backbone, and adopt it for various vision tasks from image to video domain, from classification to dense prediction. Without any extra training data, our UniFormer achieves 86.3 top-1 accuracy on ImageNet-1 K classification task. With only ImageNet-1 K pre-training, it can simply achieve state-of-the-art performance in a broad range of downstream tasks. It obtains 82.9/84.8 top-1 accuracy on Kinetics-400/600, 60.9/71.2 top-1 accuracy on Something-Something V1/V2 video classification tasks, 53.8 box AP and 46.4 mask AP on COCO object detection task, 50.8 mIoU on ADE20 K semantic segmentation task, and 77.4 AP on COCO pose estimation task. Moreover, we build an efficient UniFormer with a concise hourglass design of token shrinking and recovering, which achieves 2-4[Formula: see text] higher throughput than the recent lightweight models.

Point cloud completion aims to predict complete shape from its partial observation. Current approaches mainly consist of generation and refinement stages in a coarse-to-fine style. However, the generation stage often lacks robustness to tackle different incomplete variations, while the refinement stage blindly recovers point clouds without the semantic awareness. To tackle these challenges, we unify point cloud Completion by a generic Pretrain-Prompt-Predict paradigm, namely CP3. Inspired by prompting approaches from NLP, we creatively reinterpret point cloud generation and refinement as the prompting and predicting stages, respectively. Then, we introduce a concise self-supervised pretraining stage before prompting. It can effectively increase robustness of point cloud generation, by an Incompletion-Of-Incompletion (IOI) pretext task. Moreover, we develop a novel Semantic Conditional Refinement (SCR) network at the predicting stage. It can discriminatively modulate multi-scale refinement with the guidance of semantics. Finally, extensive experiments demonstrate that our CP3 outperforms the state-of-the-art methods with a large margin. code will be available at https://github.com/MingyeXu/cp3.

Learning spatial-temporal relation among multiple actors is crucial for group activity recognition. Different group activities often show the diversified interactions between actors in the video. Hence, it is often difficult to model complex group activities from a single view of spatial-temporal actor evolution. To tackle this problem, we propose a distinct Dual-path Actor Interaction (Dual-AI) framework, which flexibly arranges spatial and temporal transformers in two complementary orders, enhancing actor relations by integrating merits from different spatio-temporal paths. Moreover, we introduce a novel Multi-scale Actor Contrastive Loss (MAC-Loss) between two interactive paths of Dual-AI. Via self-supervised actor consistency in both frame and video levels, MAC-Loss can effectively distinguish individual actor representations to reduce action confusion among different actors. Consequently, our Dual-AI can boost group activity recognition by fusing such discriminative features of different actors. To evaluate the proposed approach, we conduct extensive experiments on the widely used benchmarks, including Volleyball [21], Collective Activity [II], and NBA datasets [49]. The proposed Dual-AI achieves state-of-the-art performance on all these datasets. It is worth noting the proposed Dual-AI with 50% training data outperforms a number of recent approaches with 100% training data. This confirms the generalization power of Dual-AI for group activity recognition, even under the challenging scenarios of limited supervision.

Cross domain object detection is a realistic and challenging task in the wild. It suffers from performance degradation due to large shift of data distributions and lack of instance-level annotations in the target domain. Existing approaches mainly focus on either of these two difficulties, even though they are closely coupled in cross domain object detection. To solve this problem, we propose a novel Target-perceived Dual-branch Distillation (TDD) framework. By integrating detection branches of both source and target domains in a unified teacher-student learning scheme, it can reduce domain shift and generate reliable supervision effectively. In particular, we first introduce a distinct Target Proposal Perceiver between two domains. It can adaptively enhance source detector to perceive objects in a target image, by leveraging target proposal contexts from iterative cross-attention. Afterwards, we design a concise Dual Branch Self Distillation strategy for model training, which can progressively integrate complementary object knowledge from different domains via self-distillation in two branches. Finally, we conduct extensive experiments on a number of widely-used scenarios in cross domain object detection. The results show that our TDD significantly outperforms the state-of-the-art methods on all the benchmarks. The codes and models will be released afterwards.

Domain adaptive object detection (DAOD) is a promising way to alleviate performance drop of detectors in new scenes. Albeit great effort made in single source domain adaptation, a more generalized task with multiple source domains remains not being well explored, due to knowledge degradation during their combination. To address this issue, we propose a novel approach, namely target-relevant knowledge preservation (TRKP), to unsupervised multi-source DAOD. Specifically, TRKP adopts the teacher-student framework, where the multi-head teacher network is built to extract knowledge from labeled source domains and guide the student network to learn detectors in unlabeled target domain. The teacher network is further equipped with an adversarial multi-source disentanglement (AMSD) module to preserve source domain-specific knowledge and simultaneously perform cross-domain alignment. Besides, a holistic target-relevant mining (HTRM) scheme is developed to re-weight the source images according to the source-target relevance. By this means, the teacher network is enforced to capture target-relevant knowledge, thus benefiting decreasing domain shift when mentoring object detection in the target domain. Extensive experiments are conducted on various widely used benchmarks with new state-of-the-art scores reported, highlighting the effectiveness.

Video understanding is an important problem in computer vision. Currently, the well-studied task in this research is human action recognition, where the clips are manually trimmed from the long videos, and a single class of human action is assumed for each clip. However, we may face more complicated scenarios in the industrial applications. For example, in the real-world urban pipe system, anomaly defects are fine-grained, multi-labeled, domain-relevant. To recognize them correctly, we need to understand the detailed video content. For this reason, we propose to advance research areas of video understanding, with a shift from traditional action recognition to industrial anomaly analysis. In particular, we introduce two high-quality video benchmarks, namely QV-Pipe and CCTV-Pipe, for anomaly inspection in the real-world urban pipe systems. Based on these new datasets, we will host two competitions including (1) Video Defect Classification on QV-Pipe and (2) Temporal Defect Localization on CCTV-Pipe. In this report, we describe the details of these benchmarks, the problem definitions of competition tracks, the evaluation metric, and the result summary. We expect that, this competition would bring new opportunities and challenges for video understanding in smart city and beyond. The details of our VideoPipe challenge can be found in https://videopipe.github.io.

Action recognition has been traditionally treated as a high-level video classification problem. However, such a manner lacks the detailed and semantic understanding of body movement, which is the critical knowledge to explain and infer complex human actions. To fill this gap, we propose to summarize a novel visual knowledge graph from over 15M detailed human annotations, for describing action as the distinct composition of body parts, part movements and interactive objects in videos. Based on it, we design a generic multi-modal Action Knowledge Understanding (AKU) framework, which can progressively infer human actions from body part movements in the videos, with assistance of visual-driven semantic knowledge mining. Finally, we validate AKU on the recent Kinetics-TPS benchmark, which contains body part parsing annotations for detailed understanding of human action in videos. The results show that, our AKU significantly boosts various video backbones with explainable action knowledge in both supervised and few shot settings, and outperforms the recent knowledge-based action recognition framework, e.g., our AKU achieves 83.9% accuracy on Kinetics-TPS while PaStaNet achieves 63.8% accuracy under the same backbone. The codes and models will be released at https://github.com/mayuelala/AKU.

Temporal action localization (TAL) is an important and challenging problem in video understanding. However, most existing TAL benchmarks are built upon the coarse granularity of action classes, which exhibits two major limitations in this task. First, coarse-level actions can make the localization models overfit in high-level context information, and ignore the atomic action details in the video. Second, the coarse action classes often lead to the ambiguous annotations of temporal boundaries, which are inappropriate for temporal action localization. To tackle these problems, we develop a novel large-scale and fine-grained video dataset, coined as FineAction, for temporal action localization. In total, FineAction contains 103K temporal instances of 106 action categories, annotated in 17K untrimmed videos. Compared to the existing TAL datasets, our FineAction takes distinct characteristics of fine action classes with rich diversity, dense annotations of multiple instances, and co-occurring actions of different classes, which introduces new opportunities and challenges for temporal action localization. To benchmark FineAction, we systematically investigate the performance of several popular temporal localization methods on it, and deeply analyze the influence of fine-grained instances in temporal action localization. As a minor contribution, we present a simple baseline approach for handling the fine-grained action detection, which achieves an mAP of 13.17% on our FineAction. We believe that FineAction can advance research of temporal action localization and beyond. The dataset is available at https://deeperaction.github.io/datasets/fineaction.

Motion modeling is crucial in modern action recognition methods. As motion dynamics like moving tempos and action amplitude may vary a lot in different video clips, it poses great challenge on adaptively covering proper motion information. To address this issue, we introduce a Motion Diversification and Selection (MoDS) module to generate diversified spatio-temporal motion features and then select the suitable motion representation dynamically for categorizing the input video. To be specific, we first propose a spatio-temporal motion generation (StMG) module to construct a bank of diversified motion features with varying spatial neighborhood and time range. Then, a dynamic motion selection (DMS) module is leveraged to choose the most discriminative motion feature both spatially and temporally from the feature bank. As a result, our proposed method can make full use of the diversified spatio-temporal motion information, while maintaining computational efficiency at the inference stage. Extensive experiments on five widely-used benchmarks, demonstrate the effectiveness of the method and we achieve state-of-the-art performance on Something-Something V1 & V2 that are of large motion variation.

The end-to-end Human Mesh Recovery (HMR) approach has been successfully used for 3D body reconstruction. However, most HMR-based frameworks reconstruct human body by directly learning mesh parameters from images or videos, while lacking explicit guidance of 3D human pose in visual data. As a result, the generated mesh often exhibits incorrect pose for complex activities. To tackle this problem, we propose to exploit 3D pose to calibrate human mesh. Specifically, we develop two novel Pose Calibration frameworks, i.e., Serial PC-HMR and Parallel PC-HMR. By coupling advanced 3D pose estimators and HMR in a serial or parallel manner, these two frameworks can effectively correct human mesh with guidance of a concise pose calibration module. Furthermore, since the calibration module is designed via non-rigid pose transformation, our PC-HMR frameworks can flexibly tackle bone length variations to alleviate misplacement in the calibrated mesh. Finally, our frameworks are based on generic and complementary integration of data-driven learning and geometrical modeling. Via plug-and-play modules, they can be efficiently adapted for both image/video-based human mesh recovery. Additionally, they have no requirement of extra 3D pose annotations in the testing phase, which releases inference difficulties in practice. We perform extensive experiments on the popular benchmarks, i.e., Human3.6M, 3DPW and SURREAL, where our PC-HMR frameworks achieve the SOTA results.

Self-supervised Multi-view stereo (MVS) with a pretext task of image reconstruction has achieved significant progress recently. However, previous methods are built upon intuitions, lacking comprehensive explanations about the effectiveness of the pretext task in self-supervised MVS. To this end, we propose to estimate epistemic uncertainty in self-supervised MVS, accounting for what the model ignores. Specially, the limitations can be categorized into two types: ambiguious supervision in foreground and invalid supervision in background. To address these issues, we propose a novel Uncertainty reduction Multi-view Stereo (U-MVS) framework for self-supervised learning. To alleviate ambiguous supervision in foreground, we involve extra correspondence prior with a flow-depth consistency loss. The dense 2D correspondence of optical flows is used to regularize the 3D stereo correspondence in MVS. To handle the invalid supervision in background, we use Monte-Carlo Dropout to acquire the uncertainty map and further filter the unreliable supervision signals on invalid regions. Extensive experiments on DTU and Tank&Temples benchmark show that our U-MVS framework1achieves the best performance among unsupervised MVS methods, with competitive performance with its supervised opponents.

Graph Convolution Network (GCN) has been successfully used for 3D human pose estimation in videos. However, it is often built on the fixed human-joint affinity, according to human skeleton. This may reduce adaptation capacity of GCN to tackle complex spatio-temporal pose variations in videos. To alleviate this problem, we propose a novel Dynamical Graph Network (DG-Net), which can dynamically identify human-joint affinity, and estimate 3D pose by adaptively learning spatial/temporal joint relations from videos. Different from traditional graph convolution, we introduce Dynamical Spatial/Temporal Graph convolution (DSG/DTG) to discover spatial/temporal human-joint affinity for each video exemplar, depending on spatial distance/temporal movement similarity between human joints in this video. Hence, they can effectively understand which joints are spatially closer and/or have consistent motion, for reducing depth ambiguity and/or motion uncertainty when lifting 2D pose to 3D pose. We conduct extensive experiments on three popular benchmarks, e.g., Human3.6M, HumanEva-I, and MPI-INF-3DHP, where DG-Net outperforms a number of recent SOTA approaches with fewer input frames and model size.

In this paper, we develop a large-scale visionlanguage fish benchmark, namely WildFish++, for comprehensive studies in multimedia research. Concretely, WildFish++ consists of 2,348 fish categories with 103,034 images in the wild, and 3,817 fish descriptions with 213,858 words. Based on these distinct characteristics, we mainly introduce four challenging research tasks on WildFish++. (1) Fine-Grained Recognition with Comparison Texts. WildFish++ naturally contains subtle difference among fish categories, which leads to fine-grained classification. Most approaches resort to tackle this problem by capturing discriminative regions in the view of each image. However, this paradigm may be still far way from extracting the most distinct features when the context on visual difference is not available. In this case, we propose to introduce comparison fish descriptions, a unique corpus that can directly point out subtle difference between highly-confused species and naturally serve as a kind of valuable context information. With such texts, we creatively elaborate a multi-modal fish network, aiming at incorporating those comparison textual information as prior knowledge and consequently leveraging it to guide CNNs to find subtle yet distinct regions in the context of comparison texts. (2) Open-Set Classification. We often confront with unknown categories in practice, e.g., there may still exist unknown fishes in our planet. Hence, we creatively adapt WildFish++ for a novel open-set classification task, which aims at correctly assigning each test image into the unknown class or one of known classes. More importantly, we investigate a number of practical designs to boost accuracy of deep learning models in open-set scenarios. (3) Cross-Modal Retrieval. WildFish++ not only contains diversified fish images in the wild but also has rich fish descriptions about morphology diagnosis, biology information, etc. Hence, we design a challenging cross-modal retrieval task, which leverages three subtasks such as text-to-text, text-to-image, image-to-text retrieval in a unified end-to-end framework. (4) Automatic Fish Classification. Automatic fish classification is a long-term research in marine biology, while current studies are unsatisfactory due to the lack of large-scale data. In this case, we train a number of CNNs with WildFish++, and use its pre-trained models to boost fish classification on most existing benchmarks of wild fishes. We will release WildFish++ with codes/protocols (https://github.com/PeiqinZhuang/WildFish++). We believe it can promote relevant studies in multimedia and beyond.

Few-shot object detection is a challenging but realistic scenario, where only a few annotated training images are available for training detectors. A popular approach to handle this problem is transfer learning, i.e., fine-tuning a detector pretrained on a source-domain benchmark. However, such transferred detector often fails to recognize new objects in the target domain, due to low data diversity of training samples. To tackle this problem, we propose a novel Context-Transformer within a concise deep transfer framework. Specifically, Context-Transformer can effectively leverage source-domain object knowledge as guidance, and automatically exploit contexts from only a few training images in the target domain. Subsequently, it can adaptively integrate these relational clues to enhance the discriminative power of detector, in order to reduce object confusion in few-shot scenarios. Moreover, Context-Transformer is flexibly embedded in the popular SSD-style detectors, which makes it a plug-and-play module for end-to-end few-shot learning. Finally, we evaluate Context-Transformer on the challenging settings of few-shot detection and incremental few-shot detection. The experimental results show that, our framework outperforms the recent state-of-the-art approaches.

Fine-grained classification is a challenging problem, due to subtle differences among highly-confused categories. Most approaches address this difficulty by learning discriminative representation of individual input image. On the other hand, humans can effectively identify contrastive clues by comparing image pairs. Inspired by this fact, this paper proposes a simple but effective Attentive Pairwise Interaction Network (API-Net), which can progressively recognize a pair of fine-grained images by interaction. Specifically, API-Net first learns a mutual feature vector to capture semantic differences in the input pair. It then compares this mutual vector with individual vectors to generate gates for each input image. These distinct gate vectors inherit mutual context on semantic differences, which allow API-Net to attentively capture contrastive clues by pairwise interaction between two images. Additionally, we train API-Net in an end-to-end manner with a score ranking regularization, which can further generalize API-Net by taking feature priorities into account. We conduct extensive experiments on five popular benchmarks in fine-grained classification. API-Net outperforms the recent SOTA methods, i.e., CUB-200-2011 (90.0%), Aircraft (93.9%), Stanford Cars (95.3%), Stanford Dogs (90.3%), and NABirds (88.1%).

Temporal convolution has been widely used for video classification. However, it is performed on spatio-temporal contexts in a limited view, which often weakens its capacity of learning video representation. To alleviate this problem, we propose a concise and novel SmallBig network, with the cooperation of small and big views. For the current time step, the small view branch is used to learn the core semantics, while the big view branch is used to capture the contextual semantics. Unlike traditional temporal convolution, the big view branch can provide the small view branch with the most activated video features from a broader 3D receptive field. Via aggregating such big-view contexts, the small view branch can learn more robust and discriminative spatio-temporal representations for video classification. Furthermore, we propose to share convolution in the small and big view branch, which improves model compactness as well as alleviates overfitting. As a result, our SmallBigNet achieves a comparable model size like 2D CNNs, while boosting accuracy like 3D CNNs. We conduct extensive experiments on the large-scale video benchmarks, e.g., Kinetics400, Something-Something V1 and V2. Our SmallBig network outperforms a number of recent state-of-the-art approaches, in terms of accuracy and/or efficiency. The codes and models will be available on https://github.com/xhl-video/SmallBigNet.

Practical applications often face a challenging continuous low-shot detection scenario, where a target detection task only has a few annotated training images, and a number of such new tasks come in sequence. To address this challenge, we propose a generic detection scheme via Disentangling-Imprinting-Distilling (DID). DID can leverage delicate transfer insights into the main development flow of deep learning, i.e., architecture design (Disentangling), model initialization (Imprinting), and training methodology (Distilling). This allows DID to be a simple but effective solution for continuous low-shot detection. In addition, DID can integrate the supervision from different detection tasks into a progressive learning procedure. As a result, one can efficiently adapt the previous detector for a new low-shot task, while maintaining the learned detection knowledge in the history. Finally, we evaluate our DID on a number of challenging settings in continuous/incremental low-shot detection. All the results demonstrate that our DID outperforms the recent state-of-the-art approaches. The code and models are available at https://github.com/chenxy99/DID.

Recent development of object detection mainly depends on deep learning with large-scale benchmarks. However, collecting such fully-annotated data is often difficult or expensive for real-world applications, which restricts the power of deep neural networks in practice. Alternatively, humans can detect new objects with little annotation burden, since humans often use the prior knowledge to identify new objects with few elaborately-annotated examples, and subsequently generalize this capacity by exploiting objects from wild images. Inspired by this procedure of learning to detect, we propose a novel Progressive Object Transfer Detection (POTD) framework. Specifically, we make three main contributions in this paper. First, POTD can leverage various object supervision of different domains effectively into a progressive detection procedure. Via such human-like learning, one can boost a target detection task with few annotations. Second, POTD consists of two delicate transfer stages, i.e., Low-Shot Transfer Detection (LSTD), and Weakly-Supervised Transfer Detection (WSTD). In LSTD, we distill the implicit object knowledge of source detector to enhance target detector with few annotations. It can effectively warm up WSTD later on. In WSTD, we design a recurrent object labelling mechanism for learning to annotate weakly-labeled images. More importantly, we exploit the reliable object supervision from LSTD, which can further enhance the robustness of target detector in the WSTD stage. Finally, we perform extensive experiments on a number of challenging detection benchmarks with different settings. The results demonstrate that, our POTD outperforms the recent state-of-the-art approaches. The codes and models are available at https://github.com/Cassie94/LSTD/tree/lstd.

Recent studies witnessed that context features can significantly improve the performance of deep semantic segmentation networks. Current context based segmentation methods differ with each other in how to construct context features and perform differently in practice. This paper firstly introduces three desirable properties of context features in segmentation task. Specially, we find that Global-guided Local Affinity (GLA) can play a vital role in constructing effective context features, while this property has been largely ignored in previous works. Based on this analysis, this paper proposes Adaptive Pyramid Context Network (APCNet) for semantic segmentation. APCNet adaptively constructs multi-scale contextual representations with multiple well-designed Adaptive Context Modules (ACMs). Specifically, each ACM leverages a global image representation as a guidance to estimate the local affinity coefficients for each sub-region, and then calculates a context vector with these affinities. We empirically evaluate our APCNet on three semantic segmentation and scene parsing datasets, including PASCAL VOC 2012, Pascal-Context, and ADE20K dataset. Experimental results show that APCNet achieves state-of-the-art performance on all three benchmarks, and obtains a new record 84.2% on PASCAL VOC 2012 test set without MS COCO pre-trained and any post-processing.

Recent studies have witnessed the successes of using 3D CNNs for video action recognition. However, most 3D models are built upon RGB and optical flow streams, which may not fully exploit pose dynamics, i.e., an important cue of modeling human actions. To fill this gap, we propose a concise Pose-Action 3D Machine (PA3D), which can effectively encode multiple pose modalities within a unified 3D framework, and consequently learn spatio-temporal pose representations for action recognition. More specifically, we introduce a novel temporal pose convolution to aggregate spatial poses over frames. Unlike the classical temporal convolution, our operation can explicitly learn the pose motions that are discriminative to recognize human actions. Extensive experiments on three popular benchmarks (i.e., JHMDB, HMDB, and Charades) show that, PA3D outperforms the recent pose-based approaches. Furthermore, PA3D is highly complementary to the recent 3D CNNs, e.g., I3D. Multi-stream fusion achieves the state-of-the-art performance on all evaluated data sets.

Deep Neural Networks (DNNs) have achieved remarkable successes in large-scale visual recognition. However, they often suffer from overfitting under noisy labels. To alleviate this problem, we propose a conceptually simple but effective MetaCleaner, which can learn to hallucinate a clean representation of an object category, according to a small noisy subset from the same category. Specially, MetaCleaner consists of two flexible submodules. The first submodule, namely Noisy Weighting, can estimate the confidence scores of all the images in the noisy subset, by analyzing their deep features jointly. The second submodule, namely Clean Hallucinating, can generate a clean representation from the noisy subset, by summarizing the noisy images with their confidence scores. Via MetaCleaner, DNNs can strengthen its robustness to noisy labels, as well as enhance its generalization capacity with richer data diversity. Moreover, MetaCleaner can be easily integrated into the standard training procedure of DNNs, which promotes its value for real-life applications. We conduct extensive experiments on two popular benchmarks in noisy-labeled recognition, i.e., Food-101N and Clothing1M. For both datasets, our MetaCleaner significantly outperforms baselines, and achieves the state-of-the-art performance.

Recent advances in object detection are mainly driven by deep learning with large-scale detection benchmarks. However, the fully-annotated training set is often limited for a target detection task, which may deteriorate the performance of deep detectors. To address this challenge, we propose a novel low-shot transfer detector (LSTD) in this paper, where we leverage rich source-domain knowledge to construct an effective target-domain detector with very few training examples. The main contributions are described as follows. First, we design a flexible deep architecture of LSTD to alleviate transfer difficulties in low-shot detection. This architecture can integrate the advantages of both SSD and Faster RCNN in a unified deep framework. Second, we introduce a novel regularized transfer learning framework for low-shot detection, where the transfer knowledge (TK) and background depression (BD) regularizations are proposed to leverage object knowledge respectively from source and target domains, in order to further enhance fine-tuning with a few target images. Finally, we examine our LSTD on a number of challenging low-shot detection experiments, where LSTD outperforms other state-of-the-art approaches. The results demonstrate that LSTD is a preferable deep detector for low-shot scenarios.

Action recognition in still images has been recently promoted by deep learning. However, the success of these deep models heavily depends on huge amount of training images for various action categories, which may not be available in practice. Alternatively, humans can classify new action categories after seeing few images, since we may not only compare appearance similarities between images on hand, but also attempt to recall importance motion cues from relevant action videos in our memory. To mimic this capacity, we propose a novel Hybrid Video Memory (HVM) machine, which can hallucinate temporal features of still images from video memory, in order to boost action recognition with few still images. First, we design a temporal memory module consisting of temporal hallucinating and predicting. Temporal hallucinating can generate temporal features of still images in an unsupervised manner. Hence, it can be flexibly used in realistic scenarios, where image and video categories may not be consistent. Temporal predicting can effectively infer action categories for query image, by integrating temporal features of training images and videos within a domain-adaptation manner. Second, we design a spatial memory module for spatial predicting. As spatial and temporal features are complementary to represent different actions, we apply spatial-temporal prediction fusion to further boost performance. Finally, we design a video selection module to select strongly-relevant videos as memory. In this case, we can balance the number of images and videos to reduce prediction bias as well as preserve computation efficiency. To show the effectiveness, we conduct extensive experiments on three challenging data sets, where our HVM outperforms a number of recent approaches by temporal hallucinating from video memory.

Fish recognition is an important task to understand the marine ecosystem and biodiversity. It is often challenging to identify fish species in the wild, due to the following difficulties. First, most fish benchmarks are small-scale, which may limit the representation power of machine learning models. Second, the number of fish species is huge, and there may still exist unknown categories in our planet. The traditional classifiers often fail to deal with this open-set scenario. Third, certain fish species are highly-confused. It is often hard to figure out the subtle differences, only by the unconstrained images. Motivated by these facts, we introduce a large-scale WildFish benchmark for fish recognition in the wild. Specifically, we make three contributions in this paper. First, WildFish is the largest image data set for wild fish recognition, to our best knowledge. It consists of 1000 fish categories with 54,459 unconstrained images, allowing to train high-capacity models for automatic fish classification. Second, we propose a novel open-set fish classification task for realistic scenarios, and investigate the open-set deep learning framework with a number of practical designs. Third, we propose a novel fine-grained recognition task, with the guidance of pairwise textual descriptions. Via leveraging the comparison knowledge in the sentence, we design a multi-modal fish net to effectively distinguish two confused categories in a pair. Finally, we release WildFish (https://github.com/PeiqinZhuang/WildFish), in order to bring benefit to more research studies in multimedia and beyond.

Recent years have witnessed the popularity of using recurrent neural network (RNN) for action recognition in videos. However, videos are of high dimensionality and contain rich human dynamics with various motion scales, which makes the traditional RNNs difficult to capture complex action information. In this paper, we propose a novel recurrent spatial-temporal attention network (RSTAN) to address this challenge, where we introduce a spatial-temporal attention mechanism to adaptively identify key features from the global video context for every time-step prediction of RNN. More specifically, we make three main contributions from the following aspects. First, we reinforce the classical long short-term memory (LSTM) with a novel spatial-temporal attention module. At each time step, our module can automatically learn a spatial-temporal action representation from all sampled video frames, which is compact and highly relevant to the prediction at the current step. Second, we design an attention-driven appearance-motion fusion strategy to integrate appearance and motion LSTMs into a unified framework, where LSTMs with their spatial-temporal attention modules in two streams can be jointly trained in an end-to-end fashion. Third, we develop actor-attention regularization for RSTAN, which can guide our attention mechanism to focus on the important action regions around actors. We evaluate the proposed RSTAN on the benchmark UCF101, HMDB51 and JHMDB data sets. The experimental results show that, our RSTAN outperforms other recent RNN-based approaches on UCF101 and HMDB51 as well as achieves the state-of-the-art on JHMDB.

Extensive studies have demonstrated that the representations of convolutional neural networks (CNN), which are learned from a large-scale data set in the source domain, can be effectively transferred to a new target domain. However, compared to the source domain, the target domain often has limited data in practice. In this case, overfitting may significantly depress transferability, due to the model redundancy of the intensive CNN structures. To deal with this difficulty, we propose a novel sparse deep transfer learning approach for CNN. There are three main contributions in this work. First, we introduce a Sparse-SourceNet to reduce the redundancy in the source domain. Second, we introduce a Hybrid-TransferNet to improve the generalization ability and the prediction accuracy of transfer learning, by taking advantage of both model sparsity and implicit knowledge. Third, we introduce a Sparse-TargetNet, where we prune our Hybrid-TransferNet to obtain a highly-compact, source-knowledge-integrated CNN in the target domain. To examine the effectiveness of our methods, we perform our sparse deep transfer learning approach on a number of benchmark transfer learning tasks. The results show that, compared to the standard fine-tuning approach, our proposed approach achieves a significant pruning rate on CNN while improves the accuracy of transfer learning.

Recent studies demonstrate the effectiveness of Recurrent Neural Networks (RNNs) for action recognition in videos. However, previous works mainly utilize video-level category as supervision to train RNNs, which may prohibit RNNs to learn complex motion structures along time. In this paper, we propose a recurrent pose-attention network (RPAN) to address this challenge, where we introduce a novel pose-attention mechanism to adaptively learn pose-related features at every time-step action prediction of RNNs. More specifically, we make three main contributions in this paper. Firstly, unlike previous works on pose-related action recognition, our RPAN is an end-to-end recurrent network which can exploit important spatial-temporal evolutions of human pose to assist action recognition in a unified framework. Secondly, instead of learning individual human-joint features separately, our pose-attention mechanism learns robust human-part features by sharing attention parameters partially on the semantically-related human joints. These human-part features are then fed into the human-part pooling layer to construct a highly-discriminative pose-related representation for temporal action modeling. Thirdly, one important byproduct of our RPAN is pose estimation in videos, which can be used for coarse pose annotation in action videos. We evaluate the proposed RPAN quantitatively and qualitatively on two popular benchmarks, i.e., Sub-JHMDB and PennAction. Experimental results show that RPAN outperforms the recent state-of-the-art methods on these challenging datasets.

Traditional feature encoding scheme (e.g., Fisher vector) with local descriptors (e.g., SIFT) and recent convolutional neural networks (CNNs) are two classes of successful methods for image recognition. In this paper, we propose a hybrid representation, which leverages the discriminative capacity of CNNs and the simplicity of descriptor encoding schema for image recognition, with a focus on scene recognition. To this end, we make three main contributions from the following aspects. First, we propose a patch-level and end-to-end architecture to model the appearance of local patches, called PatchNet. PatchNet is essentially a customized network trained in a weakly supervised manner, which uses the image-level supervision to guide the patch-level feature extraction. Second, we present a hybrid visual representation, called VSAD, by utilizing the robust feature representations of PatchNet to describe local patches and exploiting the semantic probabilities of PatchNet to aggregate these local patches into a global representation. Third, based on the proposed VSAD representation, we propose a new state-of-the-art scene recognition approach, which achieves an excellent performance on two standard benchmarks: MIT Indoor67 (86.2%) and SUN397 (73.0%).

Recent studies demonstrate the effectiveness of super vector representation in a number of visual recognition tasks. One popular approach along this line is the Vector of Locally Aggregated Descriptor (VLAD) where the super vector is encoded with a codebook generated by k-means. However, the effectiveness of the codebook is often limited, due to the poor clustering solution, the high dimensionality of visual descriptors and the global PCA for data preprocessing. To circumvent these problems, we propose three approaches for codebook enhancement, (i) partition of data, (ii) partition of feature, and (iii) local PCA. Moreover, all these approaches can be effectively integrated together to further boost the recognition performance. In our experiments, we evaluate our enhancement approaches on two challenging visual tasks, i.e., action recognition (HMDB51) and object recognition (PASCAL VOC2007). The results show that our approaches and the fusion versions significantly outperform the baselines.