Xiaohan Wang

Wenhao Wu, Xiaohan Wang, Haipeng Luo et al. · amazon-science

Vision-language models (VLMs) pre-trained on large-scale image-text pairs have demonstrated impressive transferability on various visual tasks. Transferring knowledge from such powerful VLMs is a promising direction for building effective video recognition models. However, current exploration in this field is still limited. We believe that the greatest value of pre-trained VLMs lies in building a bridge between visual and textual domains. In this paper, we propose a novel framework called BIKE, which utilizes the cross-modal bridge to explore bidirectional knowledge: i) We introduce the Video Attribute Association mechanism, which leverages the Video-to-Text knowledge to generate textual auxiliary attributes for complementing video recognition. ii) We also present a Temporal Concept Spotting mechanism that uses the Text-to-Video expertise to capture temporal saliency in a parameter-free manner, leading to enhanced video representation. Extensive studies on six popular video datasets, including Kinetics-400 & 600, UCF-101, HMDB-51, ActivityNet and Charades, show that our method achieves state-of-the-art performance in various recognition scenarios, such as general, zero-shot, and few-shot video recognition. Our best model achieves a state-of-the-art accuracy of 88.6% on the challenging Kinetics-400 using the released CLIP model. The code is available at https://github.com/whwu95/BIKE .

Xiaolong Shen, Zongxin Yang, Xiaohan Wang et al.

Video-based 3D human pose and shape estimations are evaluated by intra-frame accuracy and inter-frame smoothness. Although these two metrics are responsible for different ranges of temporal consistency, existing state-of-the-art methods treat them as a unified problem and use monotonous modeling structures (e.g., RNN or attention-based block) to design their networks. However, using a single kind of modeling structure is difficult to balance the learning of short-term and long-term temporal correlations, and may bias the network to one of them, leading to undesirable predictions like global location shift, temporal inconsistency, and insufficient local details. To solve these problems, we propose to structurally decouple the modeling of long-term and short-term correlations in an end-to-end framework, Global-to-Local Transformer (GLoT). First, a global transformer is introduced with a Masked Pose and Shape Estimation strategy for long-term modeling. The strategy stimulates the global transformer to learn more inter-frame correlations by randomly masking the features of several frames. Second, a local transformer is responsible for exploiting local details on the human mesh and interacting with the global transformer by leveraging cross-attention. Moreover, a Hierarchical Spatial Correlation Regressor is further introduced to refine intra-frame estimations by decoupled global-local representation and implicit kinematic constraints. Our GLoT surpasses previous state-of-the-art methods with the lowest model parameters on popular benchmarks, i.e., 3DPW, MPI-INF-3DHP, and Human3.6M. Codes are available at https://github.com/sxl142/GLoT.

Shuai Zhao, Linchao Zhu, Xiaohan Wang et al.

Recently, large-scale pre-training methods like CLIP have made great progress in multi-modal research such as text-video retrieval. In CLIP, transformers are vital for modeling complex multi-modal relations. However, in the vision transformer of CLIP, the essential visual tokenization process, which produces discrete visual token sequences, generates many homogeneous tokens due to the redundancy nature of consecutive and similar frames in videos. This significantly increases computation costs and hinders the deployment of video retrieval models in web applications. In this paper, to reduce the number of redundant video tokens, we design a multi-segment token clustering algorithm to find the most representative tokens and drop the non-essential ones. As the frame redundancy occurs mostly in consecutive frames, we divide videos into multiple segments and conduct segment-level clustering. Center tokens from each segment are later concatenated into a new sequence, while their original spatial-temporal relations are well maintained. We instantiate two clustering algorithms to efficiently find deterministic medoids and iteratively partition groups in high dimensional space. Through this token clustering and center selection procedure, we successfully reduce computation costs by removing redundant visual tokens. This method further enhances segment-level semantic alignment between video and text representations, enforcing the spatio-temporal interactions of tokens from within-segment frames. Our method, coined as CenterCLIP, surpasses existing state-of-the-art by a large margin on typical text-video benchmarks, while reducing the training memory cost by 35\% and accelerating the inference speed by 14\% at the best case. The code is available at \href{https://github.com/mzhaoshuai/CenterCLIP}{https://github.com/mzhaoshuai/CenterCLIP}.

Shuai Zhao, Linchao Zhu, Xiaohan Wang et al.

Self-supervised learning makes significant progress in pre-training large models, but struggles with small models. Mainstream solutions to this problem rely mainly on knowledge distillation, which involves a two-stage procedure: first training a large teacher model and then distilling it to improve the generalization ability of smaller ones. In this work, we introduce another one-stage solution to obtain pre-trained small models without the need for extra teachers, namely, slimmable networks for contrastive self-supervised learning (SlimCLR). A slimmable network consists of a full network and several weight-sharing sub-networks, which can be pre-trained once to obtain various networks, including small ones with low computation costs. However, interference between weight-sharing networks leads to severe performance degradation in self-supervised cases, as evidenced by gradient magnitude imbalance and gradient direction divergence. The former indicates that a small proportion of parameters produce dominant gradients during backpropagation, while the main parameters may not be fully optimized. The latter shows that the gradient direction is disordered, and the optimization process is unstable. To address these issues, we introduce three techniques to make the main parameters produce dominant gradients and sub-networks have consistent outputs. These techniques include slow start training of sub-networks, online distillation, and loss re-weighting according to model sizes. Furthermore, theoretical results are presented to demonstrate that a single slimmable linear layer is sub-optimal during linear evaluation. Thus a switchable linear probe layer is applied during linear evaluation. We instantiate SlimCLR with typical contrastive learning frameworks and achieve better performance than previous arts with fewer parameters and FLOPs. The code is at https://github.com/mzhaoshuai/SlimCLR.

Xin Xie, Zhoubo Li, Xiaohan Wang et al.

Knowledge Graphs (KGs) often have two characteristics: heterogeneous graph structure and text-rich entity/relation information. Text-based KG embeddings can represent entities by encoding descriptions with pre-trained language models, but no open-sourced library is specifically designed for KGs with PLMs at present. In this paper, we present LambdaKG, a library for KGE that equips with many pre-trained language models (e.g., BERT, BART, T5, GPT-3), and supports various tasks (e.g., knowledge graph completion, question answering, recommendation, and knowledge probing). LambdaKG is publicly open-sourced at https://github.com/zjunlp/PromptKG/tree/main/lambdaKG, with a demo video at http://deepke.zjukg.cn/lambdakg.mp4 and long-term maintenance.

Shengyu Mao, Xiaohan Wang, Mengru Wang et al.

This paper introduces an innovative task focused on editing the personality traits of Large Language Models (LLMs). This task seeks to adjust the models' responses to opinion-related questions on specified topics since an individual's personality often manifests in the form of their expressed opinions, thereby showcasing different personality traits. Specifically, we construct PersonalityEdit, a new benchmark dataset to address this task. Drawing on the theory in Social Psychology, we isolate three representative traits, namely Neuroticism, Extraversion, and Agreeableness, as the foundation for our benchmark. We then gather data using GPT-4, generating responses that align with a specified topic and embody the targeted personality trait. We conduct comprehensive experiments involving various baselines and discuss the representation of personality behavior in LLMs. Our findings uncover potential challenges of the proposed task, illustrating several remaining issues. We anticipate that our work can stimulate further annotation in model editing and personality-related research. Code is available at https://github.com/zjunlp/EasyEdit.

Rui Liu, Xiaohan Wang, Wenguan Wang et al.

Vision-language navigation (VLN), which entails an agent to navigate 3D environments following human instructions, has shown great advances. However, current agents are built upon panoramic observations, which hinders their ability to perceive 3D scene geometry and easily leads to ambiguous selection of panoramic view. To address these limitations, we present a BEV Scene Graph (BSG), which leverages multi-step BEV representations to encode scene layouts and geometric cues of indoor environment under the supervision of 3D detection. During navigation, BSG builds a local BEV representation at each step and maintains a BEV-based global scene map, which stores and organizes all the online collected local BEV representations according to their topological relations. Based on BSG, the agent predicts a local BEV grid-level decision score and a global graph-level decision score, combined with a sub-view selection score on panoramic views, for more accurate action prediction. Our approach significantly outperforms state-of-the-art methods on REVERIE, R2R, and R4R, showing the potential of BEV perception in VLN.

Xiaohan Wang, Wenguan Wang, Jiayi Shao et al.

Recently, visual-language navigation (VLN) -- entailing robot agents to follow navigation instructions -- has shown great advance. However, existing literature put most emphasis on interpreting instructions into actions, only delivering "dumb" wayfinding agents. In this article, we devise LANA, a language-capable navigation agent which is able to not only execute human-written navigation commands, but also provide route descriptions to humans. This is achieved by simultaneously learning instruction following and generation with only one single model. More specifically, two encoders, respectively for route and language encoding, are built and shared by two decoders, respectively, for action prediction and instruction generation, so as to exploit cross-task knowledge and capture task-specific characteristics. Throughout pretraining and fine-tuning, both instruction following and generation are set as optimization objectives. We empirically verify that, compared with recent advanced task-specific solutions, LANA attains better performances on both instruction following and route description, with nearly half complexity. In addition, endowed with language generation capability, LANA can explain to humans its behaviors and assist human's wayfinding. This work is expected to foster future efforts towards building more trustworthy and socially-intelligent navigation robots.

Tuo Feng, Wenguan Wang, Xiaohan Wang et al.

Point cloud analysis (such as 3D segmentation and detection) is a challenging task, because of not only the irregular geometries of many millions of unordered points, but also the great variations caused by depth, viewpoint, occlusion, etc. Current studies put much focus on the adaption of neural networks to the complex geometries of point clouds, but are blind to a fundamental question: how to learn an appropriate point embedding space that is aware of both discriminative semantics and challenging variations? As a response, we propose a clustering based supervised learning scheme for point cloud analysis. Unlike current de-facto, scene-wise training paradigm, our algorithm conducts within-class clustering on the point embedding space for automatically discovering subclass patterns which are latent yet representative across scenes. The mined patterns are, in turn, used to repaint the embedding space, so as to respect the underlying distribution of the entire training dataset and improve the robustness to the variations. Our algorithm is principled and readily pluggable to modern point cloud segmentation networks during training, without extra overhead during testing. With various 3D network architectures (i.e., voxel-based, point-based, Transformer-based, automatically searched), our algorithm shows notable improvements on famous point cloud segmentation datasets (i.e.,2.0-2.6% on single-scan and 2.0-2.2% multi-scan of SemanticKITTI, 1.8-1.9% on S3DIS, in terms of mIoU). Our algorithm also demonstrates utility in 3D detection, showing 2.0-3.4% mAP gains on KITTI.

Haitian Zeng, Xiaohan Wang, Wenguan Wang et al.

We introduce a novel speaker model \textsc{Kefa} for navigation instruction generation. The existing speaker models in Vision-and-Language Navigation suffer from the large domain gap of vision features between different environments and insufficient temporal grounding capability. To address the challenges, we propose a Knowledge Refinement Module to enhance the feature representation with external knowledge facts, and an Adaptive Temporal Alignment method to enforce fine-grained alignment between the generated instructions and the observation sequences. Moreover, we propose a new metric SPICE-D for navigation instruction evaluation, which is aware of the correctness of direction phrases. The experimental results on R2R and UrbanWalk datasets show that the proposed KEFA speaker achieves state-of-the-art instruction generation performance for both indoor and outdoor scenes.

Jiayi Shao, Xiaohan Wang, Ruijie Quan et al.

This report presents ReLER submission to two tracks in the Ego4D Episodic Memory Benchmark in CVPR 2023, including Natural Language Queries and Moment Queries. This solution inherits from our proposed Action Sensitivity Learning framework (ASL) to better capture discrepant information of frames. Further, we incorporate a series of stronger video features and fusion strategies. Our method achieves an average mAP of 29.34, ranking 1st in Moment Queries Challenge, and garners 19.79 mean R1, ranking 2nd in Natural Language Queries Challenge. Our code will be released.

Yucheng Suo, Zhedong Zheng, Xiaohan Wang et al.

Sign language is the window for people differently-abled to express their feelings as well as emotions. However, it remains challenging for people to learn sign language in a short time. To address this real-world challenge, in this work, we study the motion transfer system, which can transfer the user photo to the sign language video of specific words. In particular, the appearance content of the output video comes from the provided user image, while the motion of the video is extracted from the specified tutorial video. We observe two primary limitations in adopting the state-of-the-art motion transfer methods to sign language generation:(1) Existing motion transfer works ignore the prior geometrical knowledge of the human body. (2) The previous image animation methods only take image pairs as input in the training stage, which could not fully exploit the temporal information within videos. In an attempt to address the above-mentioned limitations, we propose Structure-aware Temporal Consistency Network (STCNet) to jointly optimize the prior structure of human with the temporal consistency for sign language video generation. There are two main contributions in this paper. (1) We harness a fine-grained skeleton detector to provide prior knowledge of the body keypoints. In this way, we ensure the keypoint movement in a valid range and make the model become more explainable and robust. (2) We introduce two cycle-consistency losses, i.e., short-term cycle loss and long-term cycle loss, which are conducted to assure the continuity of the generated video. We optimize the two losses and keypoint detector network in an end-to-end manner.

Jiahao Li, Zongxin Yang, Xiaohan Wang et al.

In this study, we focus on the problem of 3D human mesh recovery from a single image under obscured conditions. Most state-of-the-art methods aim to improve 2D alignment technologies, such as spatial averaging and 2D joint sampling. However, they tend to neglect the crucial aspect of 3D alignment by improving 3D representations. Furthermore, recent methods struggle to separate the target human from occlusion or background in crowded scenes as they optimize the 3D space of target human with 3D joint coordinates as local supervision. To address these issues, a desirable method would involve a framework for fusing 2D and 3D features and a strategy for optimizing the 3D space globally. Therefore, this paper presents 3D JOint contrastive learning with TRansformers (JOTR) framework for handling occluded 3D human mesh recovery. Our method includes an encoder-decoder transformer architecture to fuse 2D and 3D representations for achieving 2D$\&$3D aligned results in a coarse-to-fine manner and a novel 3D joint contrastive learning approach for adding explicitly global supervision for the 3D feature space. The contrastive learning approach includes two contrastive losses: joint-to-joint contrast for enhancing the similarity of semantically similar voxels (i.e., human joints), and joint-to-non-joint contrast for ensuring discrimination from others (e.g., occlusions and background). Qualitative and quantitative analyses demonstrate that our method outperforms state-of-the-art competitors on both occlusion-specific and standard benchmarks, significantly improving the reconstruction of occluded humans.

Naiyuan Liu, Xiaohan Wang, Xiaobo Li et al.

In this report, we present the ReLER@ZJU-Alibaba submission to the Ego4D Natural Language Queries (NLQ) Challenge in CVPR 2022. Given a video clip and a text query, the goal of this challenge is to locate a temporal moment of the video clip where the answer to the query can be obtained. To tackle this task, we propose a multi-scale cross-modal transformer and a video frame-level contrastive loss to fully uncover the correlation between language queries and video clips. Besides, we propose two data augmentation strategies to increase the diversity of training samples. The experimental results demonstrate the effectiveness of our method. The final submission ranked first on the leaderboard.

Orr Zohar, Xiaohan Wang, Yonatan Bitton et al.

The performance of Large Vision Language Models (LVLMs) is dependent on the size and quality of their training datasets. Existing video instruction tuning datasets lack diversity as they are derived by prompting large language models with video captions to generate question-answer pairs, and are therefore mostly descriptive. Meanwhile, many labeled video datasets with diverse labels and supervision exist - however, we find that their integration into LVLMs is non-trivial. Herein, we present Video Self-Training with augmented Reasoning (Video-STaR), the first video self-training approach. Video-STaR allows the utilization of any labeled video dataset for video instruction tuning. In Video-STaR, an LVLM cycles between instruction generation and finetuning, which we show (I) improves general video understanding and (II) adapts LVLMs to novel downstream tasks with existing supervision. During generation, an LVLM is prompted to propose an answer. The answers are then filtered only to those that contain the original video labels, and the LVLM is then re-trained on the generated dataset. By only training on generated answers that contain the correct video labels, Video-STaR utilizes these existing video labels as weak supervision for video instruction tuning. Our results demonstrate that Video-STaR-enhanced LVLMs exhibit improved performance in (I) general video QA, where TempCompass performance improved by 10%, and (II) on downstream tasks, where Video-STaR improved Kinetics700-QA accuracy by 20% and action quality assessment on FineDiving by 15%.

Xiaohan Wang, Dian Li, Yilin Zhao et al.

Utilizing tools with Large Language Models (LLMs) is essential for grounding AI agents in real-world applications. The prevailing approach involves few-shot prompting with demonstrations or fine-tuning with expert annotations. However, mere in-context demonstrations may fail to cover sufficient knowledge for complex tools and tasks. Training on solution paths is also hindered by the high cost of expert annotations and generalizing to new tools. A core challenge of generalizable tool use lies in understanding the "meta", or fundamental natures of tools that are transferable across tasks, such as causality and constraints. In this paper, we present MetaTool, a novel tool learning methodology designed to generalize across any reusable toolset. Our approach incorporates a self-supervised augmentation technique derived from a series of meta-tasks. This involves predicting masked elements in the tool execution process. The self-supervised procedure enables scalable generation of high-quality QA data, which is handy for supervising tool understanding. By incorporating meta-task data into task-oriented training, our method significantly enhances the performance of open-source LLMs, achieving results comparable to ChatGPT in both tool-based planning and chatting scenarios. Through large-scale instruction tuning, the MetaTool model demonstrates impressive zero-shot generalizability on new tasks.

Yuebing Liang, Yichao Liu, Xiaohan Wang et al.

Public events, such as concerts and sports games, can be major attractors for large crowds, leading to irregular surges in travel demand. Accurate human mobility prediction for public events is thus crucial for event planning as well as traffic or crowd management. While rich textual descriptions about public events are commonly available from online sources, it is challenging to encode such information in statistical or machine learning models. Existing methods are generally limited in incorporating textual information, handling data sparsity, or providing rationales for their predictions. To address these challenges, we introduce a framework for human mobility prediction under public events (LLM-MPE) based on Large Language Models (LLMs), leveraging their unprecedented ability to process textual data, learn from minimal examples, and generate human-readable explanations. Specifically, LLM-MPE first transforms raw, unstructured event descriptions from online sources into a standardized format, and then segments historical mobility data into regular and event-related components. A prompting strategy is designed to direct LLMs in making and rationalizing demand predictions considering historical mobility and event features. A case study is conducted for Barclays Center in New York City, based on publicly available event information and taxi trip data. Results show that LLM-MPE surpasses traditional models, particularly on event days, with textual data significantly enhancing its accuracy. Furthermore, LLM-MPE offers interpretable insights into its predictions. Despite the great potential of LLMs, we also identify key challenges including misinformation and high costs that remain barriers to their broader adoption in large-scale human mobility analysis.

Hao Chen, Zhexin Hu, Jiajun Chai et al.

Training LLMs to invoke tools and leverage retrieved information necessitates high-quality, diverse data. However, existing pipelines for synthetic data generation often rely on tens of thousands of real API calls to enhance generalization, incurring prohibitive costs while lacking multi-hop reasoning and self-reflection. To address these limitations, we introduce ToolForge, an automated synthesis framework that achieves strong real-world tool-calling performance by constructing only a small number of virtual tools, eliminating the need for real API calls. ToolForge leverages a (question, golden context, answer) triple to synthesize large-scale tool-learning data specifically designed for multi-hop search scenarios, further enriching the generated data through multi-hop reasoning and self-reflection mechanisms. To ensure data fidelity, we employ a Multi-Layer Validation Framework that integrates both rule-based and model-based assessments. Empirical results show that a model with only 8B parameters, when trained on our synthesized data, outperforms GPT-4o on multiple benchmarks. Our code and dataset are publicly available at https://github.com/Buycar-arb/ToolForge .

Xiaohan Wang, Xiaoyan Yang, Yuqi Zhu et al.

Large Language Models (LLMs) like GPT-4, MedPaLM-2, and Med-Gemini achieve performance competitively with human experts across various medical benchmarks. However, they still face challenges in making professional diagnoses akin to physicians, particularly in efficiently gathering patient information and reasoning the final diagnosis. To this end, we introduce the RuleAlign framework, designed to align LLMs with specific diagnostic rules. We develop a medical dialogue dataset comprising rule-based communications between patients and physicians and design an alignment learning approach through preference learning. Experimental results demonstrate the effectiveness of the proposed approach. We hope that our work can serve as an inspiration for exploring the potential of LLMs as AI physicians.

Jiayi Shao, Xiaohan Wang, Yi Yang

In this report, we present the ReLER@ZJU1 submission to the Ego4D Moment Queries Challenge in ECCV 2022. In this task, the goal is to retrieve and localize all instances of possible activities in egocentric videos. Ego4D dataset is challenging for the temporal action localization task as the temporal duration of the videos is quite long and each video contains multiple action instances with fine-grained action classes. To address these problems, we utilize a multi-scale transformer to classify different action categories and predict the boundary of each instance. Moreover, in order to better capture the long-term temporal dependencies in the long videos, we propose a segment-level recurrence mechanism. Compared with directly feeding all video features to the transformer encoder, the proposed segment-level recurrence mechanism alleviates the optimization difficulties and achieves better performance. The final submission achieved Recall@1,tIoU=0.5 score of 37.24, average mAP score of 17.67 and took 3-rd place on the leaderboard.

Liping Wang, Jiawei Li, Lifan Zhao et al.

Predicting stock prices presents a challenging research problem due to the inherent volatility and non-linear nature of the stock market. In recent years, knowledge-enhanced stock price prediction methods have shown groundbreaking results by utilizing external knowledge to understand the stock market. Despite the importance of these methods, there is a scarcity of scholarly works that systematically synthesize previous studies from the perspective of external knowledge types. Specifically, the external knowledge can be modeled in different data structures, which we group into non-graph-based formats and graph-based formats: 1) non-graph-based knowledge captures contextual information and multimedia descriptions specifically associated with an individual stock; 2) graph-based knowledge captures interconnected and interdependent information in the stock market. This survey paper aims to provide a systematic and comprehensive description of methods for acquiring external knowledge from various unstructured data sources and then incorporating it into stock price prediction models. We also explore fusion methods for combining external knowledge with historical price features. Moreover, this paper includes a compilation of relevant datasets and delves into potential future research directions in this domain.

Siyu Xia, Zekun Xu, Jiajun Chai et al.

Large Language Models (LLMs) based agents have demonstrated remarkable potential in autonomous task-solving across complex, open-ended environments. A promising approach for improving the reasoning capabilities of LLM agents is to better utilize prior experiences in guiding current decisions. However, LLMs acquire experience either through implicit memory via training, which suffers from catastrophic forgetting and limited interpretability, or explicit memory via prompting, which lacks adaptability. In this paper, we introduce a novel agent-centric, trainable, multi-layered graph memory framework and evaluate how context memory enhances the ability of LLMs to utilize parametric information. The graph abstracts raw agent trajectories into structured decision paths in a state machine and further distills them into high-level, human-interpretable strategic meta-cognition. In order to make memory adaptable, we propose a reinforcement-based weight optimization procedure that estimates the empirical utility of each meta-cognition based on reward feedback from downstream tasks. These optimized strategies are then dynamically integrated into the LLM agent's training loop through meta-cognitive prompting. Empirically, the learnable graph memory delivers robust generalization, improves LLM agents' strategic reasoning performance, and provides consistent benefits during Reinforcement Learning (RL) training.

James Burgess, Xiaohan Wang, Yuhui Zhang et al. · stanford

How do two individuals differ when performing the same action? In this work, we introduce Video Action Differencing (VidDiff), the novel task of identifying subtle differences between videos of the same action, which has many applications, such as coaching and skill learning. To enable development on this new task, we first create VidDiffBench, a benchmark dataset containing 549 video pairs, with human annotations of 4,469 fine-grained action differences and 2,075 localization timestamps indicating where these differences occur. Our experiments demonstrate that VidDiffBench poses a significant challenge for state-of-the-art large multimodal models (LMMs), such as GPT-4o and Qwen2-VL. By analyzing failure cases of LMMs on VidDiffBench, we highlight two key challenges for this task: localizing relevant sub-actions over two videos and fine-grained frame comparison. To overcome these, we propose the VidDiff method, an agentic workflow that breaks the task into three stages: action difference proposal, keyframe localization, and frame differencing, each stage utilizing specialized foundation models. To encourage future research in this new task, we release the benchmark at https://huggingface.co/datasets/jmhb/VidDiffBench and code at http://jmhb0.github.io/viddiff.

Josiah Aklilu, Xiaohan Wang, Serena Yeung-Levy

Precise action localization in untrimmed video is vital for fields such as professional sports and minimally invasive surgery, where the delineation of particular motions in recordings can dramatically enhance analysis. But in many cases, large scale datasets with video-label pairs for localization are unavailable, limiting the opportunity to fine-tune video-understanding models. Recent developments in large vision-language models (LVLM) address this need with impressive zero-shot capabilities in a variety of video understanding tasks. However, the adaptation of LVLMs, with their powerful visual question answering capabilities, to zero-shot localization in long-form video is still relatively unexplored. To this end, we introduce a true Zero-shot Action Localization method (ZEAL). Specifically, we leverage the built-in action knowledge of a large language model (LLM) to inflate actions into detailed descriptions of the archetypal start and end of the action. These descriptions serve as queries to LVLM for generating frame-level confidence scores which can be aggregated to produce localization outputs. The simplicity and flexibility of our method lends it amenable to more capable LVLMs as they are developed, and we demonstrate remarkable results in zero-shot action localization on a challenging benchmark, without any training. Our code is publicly available at $\href{https://github.com/josaklil-ai/zeal}{github.com/josaklil-ai/zeal}$.

Xiaohan Wang, Yuhui Zhang, Orr Zohar et al. · stanford

Long-form video understanding represents a significant challenge within computer vision, demanding a model capable of reasoning over long multi-modal sequences. Motivated by the human cognitive process for long-form video understanding, we emphasize interactive reasoning and planning over the ability to process lengthy visual inputs. We introduce a novel agent-based system, VideoAgent, that employs a large language model as a central agent to iteratively identify and compile crucial information to answer a question, with vision-language foundation models serving as tools to translate and retrieve visual information. Evaluated on the challenging EgoSchema and NExT-QA benchmarks, VideoAgent achieves 54.1% and 71.3% zero-shot accuracy with only 8.4 and 8.2 frames used on average. These results demonstrate superior effectiveness and efficiency of our method over the current state-of-the-art methods, highlighting the potential of agent-based approaches in advancing long-form video understanding.

Xiangpeng Yang, Linchao Zhu, Xiaohan Wang et al.

Text-video retrieval is a critical multi-modal task to find the most relevant video for a text query. Although pretrained models like CLIP have demonstrated impressive potential in this area, the rising cost of fully finetuning these models due to increasing model size continues to pose a problem. To address this challenge, prompt tuning has emerged as an alternative. However, existing works still face two problems when adapting pretrained image-text models to downstream video-text tasks: (1) The visual encoder could only encode frame-level features and failed to extract global-level general video information. (2) Equipping the visual and text encoder with separated prompts failed to mitigate the visual-text modality gap. To this end, we propose DGL, a cross-modal Dynamic prompt tuning method with Global-Local video attention. In contrast to previous prompt tuning methods, we employ the shared latent space to generate local-level text and frame prompts that encourage inter-modal interaction. Furthermore, we propose modeling video in a global-local attention mechanism to capture global video information from the perspective of prompt tuning. Extensive experiments reveal that when only 0.67% parameters are tuned, our cross-modal prompt tuning strategy DGL outperforms or is comparable to fully finetuning methods on MSR-VTT, VATEX, LSMDC, and ActivityNet datasets. Code will be available at https://github.com/knightyxp/DGL

Shuai Zhao, Xiaohan Wang, Linchao Zhu et al.

One fascinating aspect of pre-trained vision-language models~(VLMs) learning under language supervision is their impressive zero-shot generalization capability. However, this ability is hindered by distribution shifts between the training and testing data. Previous test time adaptation~(TTA) methods for VLMs in zero-shot classification rely on minimizing the entropy of model outputs, tending to be stuck in incorrect model predictions. In this work, we propose TTA with feedback to rectify the model output and prevent the model from becoming blindly confident. Specifically, a CLIP model is adopted as the reward model during TTA and provides feedback for the VLM. Given a single test sample, the VLM is forced to maximize the CLIP reward between the input and sampled results from the VLM output distribution. The proposed \textit{reinforcement learning with CLIP feedback~(RLCF)} framework is highly flexible and universal. Beyond the classification task, with task-specific sampling strategies and a proper reward baseline choice, RLCF can be easily extended to not only discrimination tasks like retrieval but also generalization tasks like image captioning, improving the zero-shot generalization capacity of VLMs. According to the characteristics of these VL tasks, we build different fully TTA pipelines with RLCF to improve the zero-shot generalization ability of various VLMs. Extensive experiments along with promising empirical results demonstrate the effectiveness of RLCF. The code is available at https://github.com/mzhaoshuai/RLCF.

Yuqi Zhu, Xiaohan Wang, Jing Chen et al.

This paper presents an exhaustive quantitative and qualitative evaluation of Large Language Models (LLMs) for Knowledge Graph (KG) construction and reasoning. We engage in experiments across eight diverse datasets, focusing on four representative tasks encompassing entity and relation extraction, event extraction, link prediction, and question-answering, thereby thoroughly exploring LLMs' performance in the domain of construction and inference. Empirically, our findings suggest that LLMs, represented by GPT-4, are more suited as inference assistants rather than few-shot information extractors. Specifically, while GPT-4 exhibits good performance in tasks related to KG construction, it excels further in reasoning tasks, surpassing fine-tuned models in certain cases. Moreover, our investigation extends to the potential generalization ability of LLMs for information extraction, leading to the proposition of a Virtual Knowledge Extraction task and the development of the corresponding VINE dataset. Based on these empirical findings, we further propose AutoKG, a multi-agent-based approach employing LLMs and external sources for KG construction and reasoning. We anticipate that this research can provide invaluable insights for future undertakings in the field of knowledge graphs. The code and datasets are in https://github.com/zjunlp/AutoKG.

Kezhou Lin, Xiaohan Wang, Linchao Zhu et al.

In this paper, we tackle the problem of sign language translation (SLT) without gloss annotations. Although intermediate representation like gloss has been proven effective, gloss annotations are hard to acquire, especially in large quantities. This limits the domain coverage of translation datasets, thus handicapping real-world applications. To mitigate this problem, we design the Gloss-Free End-to-end sign language translation framework (GloFE). Our method improves the performance of SLT in the gloss-free setting by exploiting the shared underlying semantics of signs and the corresponding spoken translation. Common concepts are extracted from the text and used as a weak form of intermediate representation. The global embedding of these concepts is used as a query for cross-attention to find the corresponding information within the learned visual features. In a contrastive manner, we encourage the similarity of query results between samples containing such concepts and decrease those that do not. We obtained state-of-the-art results on large-scale datasets, including OpenASL and How2Sign. The code and model will be available at https://github.com/HenryLittle/GloFE.

Xin Xu, Yuqi Zhu, Xiaohan Wang et al.

Scaling language models have revolutionized widespread NLP tasks, yet little comprehensively explored few-shot relation extraction with large language models. In this paper, we investigate principal methodologies, in-context learning and data generation, for few-shot relation extraction via GPT-3.5 through exhaustive experiments. To enhance few-shot performance, we further propose task-related instructions and schema-constrained data generation. We observe that in-context learning can achieve performance on par with previous prompt learning approaches, and data generation with the large language model can boost previous solutions to obtain new state-of-the-art few-shot results on four widely-studied relation extraction datasets. We hope our work can inspire future research for the capabilities of large language models in few-shot relation extraction. Code is available in https://github.com/zjunlp/DeepKE/tree/main/example/llm.

Shuai Bai, Zhedong Zheng, Xiaohan Wang et al.

Vehicle search is one basic task for the efficient traffic management in terms of the AI City. Most existing practices focus on the image-based vehicle matching, including vehicle re-identification and vehicle tracking. In this paper, we apply one new modality, i.e., the language description, to search the vehicle of interest and explore the potential of this task in the real-world scenario. The natural language-based vehicle search poses one new challenge of fine-grained understanding of both vision and language modalities. To connect language and vision, we propose to jointly train the state-of-the-art vision models with the transformer-based language model in an end-to-end manner. Except for the network structure design and the training strategy, several optimization objectives are also re-visited in this work. The qualitative and quantitative experiments verify the effectiveness of the proposed method. Our proposed method has achieved the 1st place on the 5th AI City Challenge, yielding competitive performance 18.69% MRR accuracy on the private test set. We hope this work can pave the way for the future study on using language description effectively and efficiently for real-world vehicle retrieval systems. The code will be available at https://github.com/ShuaiBai623/AIC2021-T5-CLV.

Zhedong Zheng, Nenggan Zheng, Yi Yang

People live in a 3D world. However, existing works on person re-identification (re-id) mostly consider the semantic representation learning in a 2D space, intrinsically limiting the understanding of people. In this work, we address this limitation by exploring the prior knowledge of the 3D body structure. Specifically, we project 2D images to a 3D space and introduce a novel parameter-efficient Omni-scale Graph Network (OG-Net) to learn the pedestrian representation directly from 3D point clouds. OG-Net effectively exploits the local information provided by sparse 3D points and takes advantage of the structure and appearance information in a coherent manner. With the help of 3D geometry information, we can learn a new type of deep re-id feature free from noisy variants, such as scale and viewpoint. To our knowledge, we are among the first attempts to conduct person re-identification in the 3D space. We demonstrate through extensive experiments that the proposed method (1) eases the matching difficulty in the traditional 2D space, (2) exploits the complementary information of 2D appearance and 3D structure, (3) achieves competitive results with limited parameters on four large-scale person re-id datasets, and (4) has good scalability to unseen datasets. Our code, models and generated 3D human data are publicly available at https://github.com/layumi/person-reid-3d .

Lisa Dunlap, Yuhui Zhang, Xiaohan Wang et al. · stanford

How do two sets of images differ? Discerning set-level differences is crucial for understanding model behaviors and analyzing datasets, yet manually sifting through thousands of images is impractical. To aid in this discovery process, we explore the task of automatically describing the differences between two $\textbf{sets}$ of images, which we term Set Difference Captioning. This task takes in image sets $D_A$ and $D_B$, and outputs a description that is more often true on $D_A$ than $D_B$. We outline a two-stage approach that first proposes candidate difference descriptions from image sets and then re-ranks the candidates by checking how well they can differentiate the two sets. We introduce VisDiff, which first captions the images and prompts a language model to propose candidate descriptions, then re-ranks these descriptions using CLIP. To evaluate VisDiff, we collect VisDiffBench, a dataset with 187 paired image sets with ground truth difference descriptions. We apply VisDiff to various domains, such as comparing datasets (e.g., ImageNet vs. ImageNetV2), comparing classification models (e.g., zero-shot CLIP vs. supervised ResNet), summarizing model failure modes (supervised ResNet), characterizing differences between generative models (e.g., StableDiffusionV1 and V2), and discovering what makes images memorable. Using VisDiff, we are able to find interesting and previously unknown differences in datasets and models, demonstrating its utility in revealing nuanced insights.

Rui Li, Xiaohan Wang, Yuhui Zhang et al. · stanford

Despite significant advancements in video large multimodal models (video-LMMs), achieving effective temporal grounding in long-form videos remains a challenge for existing models. To address this limitation, we propose Temporal Preference Optimization (TPO), a novel post-training framework designed to enhance the temporal grounding capabilities of video-LMMs through preference learning. TPO adopts a self-training approach that enables models to differentiate between well-grounded and less accurate temporal responses by leveraging curated preference datasets at two granularities: localized temporal grounding, which focuses on specific video segments, and comprehensive temporal grounding, which captures extended temporal dependencies across entire video sequences. By optimizing on these preference datasets, TPO significantly enhances temporal understanding while reducing reliance on manually annotated data. Extensive experiments on three long-form video understanding benchmarks--LongVideoBench, MLVU, and Video-MME--demonstrate the effectiveness of TPO across two state-of-the-art video-LMMs. Notably, LLaVA-Video-TPO establishes itself as the leading 7B model on the Video-MME benchmark, underscoring the potential of TPO as a scalable and efficient solution for advancing temporal reasoning in long-form video understanding. Project page: https://ruili33.github.io/tpo_website.

Xiaohan Wang, Shengyu Mao, Ningyu Zhang et al.

Recently, there has been a growing interest in knowledge editing for Large Language Models (LLMs). Current approaches and evaluations merely explore the instance-level editing, while whether LLMs possess the capability to modify concepts remains unclear. This paper pioneers the investigation of editing conceptual knowledge for LLMs, by constructing a novel benchmark dataset ConceptEdit and establishing a suite of new metrics for evaluation. The experimental results reveal that, although existing editing methods can efficiently modify concept-level definition to some extent, they also have the potential to distort the related instantial knowledge in LLMs, leading to poor performance. We anticipate this can inspire further progress in better understanding LLMs. Our project homepage is available at https://zjunlp.github.io/project/ConceptEdit.

Mark Endo, Xiaohan Wang, Serena Yeung-Levy

Recent works on accelerating Vision-Language Models achieve strong performance across a variety of vision-language tasks despite highly compressing visual information. In this work, we examine the popular acceleration approach of early pruning of visual tokens inside the language model. Surprisingly, we find that while strong performance is maintained across many tasks, it exhibits drastically different behavior for a subset of vision-centric tasks such as localization. Upon further investigation, we uncover a core issue with the acceleration approach where most tokens towards the top of the image are pruned away. Yet, on many benchmarks aiming to evaluate vision-centric capabilities, strong performance persists with the flawed pruning strategy, highlighting these benchmarks' limited ability to assess fine-grained visual capabilities. Based on these findings, we propose FEATHER (Fast and Effective Acceleration wiTH Ensemble cRiteria), a straightforward approach that resolves the discovered early-layer pruning issue and further enhances the preservation of relevant tokens via multistage pruning with early uniform sampling to ensure broad image coverage. With comparable computational savings, we find that FEATHER achieves more than 5x performance improvement on the vision-centric localization benchmarks compared to the original acceleration approach.

Devanish N. Kamtam, Joseph B. Shrager, Satya Deepya Malla et al.

Background: We evaluate SAM 2 for surgical scene understanding by examining its semantic segmentation capabilities for organs/tissues both in zero-shot scenarios and after fine-tuning. Methods: We utilized five public datasets to evaluate and fine-tune SAM 2 for segmenting anatomical tissues in surgical videos/images. Fine-tuning was applied to the image encoder and mask decoder. We limited training subsets from 50 to 400 samples per class to better model real-world constraints with data acquisition. The impact of dataset size on fine-tuning performance was evaluated with weighted mean Dice coefficient (WMDC), and the results were also compared against previously reported state-of-the-art (SOTA) results. Results: SurgiSAM 2, a fine-tuned SAM 2 model, demonstrated significant improvements in segmentation performance, achieving a 17.9% relative WMDC gain compared to the baseline SAM 2. Increasing prompt points from 1 to 10 and training data scale from 50/class to 400/class enhanced performance; the best WMDC of 0.92 on the validation subset was achieved with 10 prompt points and 400 samples per class. On the test subset, this model outperformed prior SOTA methods in 24/30 (80%) of the classes with a WMDC of 0.91 using 10-point prompts. Notably, SurgiSAM 2 generalized effectively to unseen organ classes, achieving SOTA on 7/9 (77.8%) of them. Conclusion: SAM 2 achieves remarkable zero-shot and fine-tuned performance for surgical scene segmentation, surpassing prior SOTA models across several organ classes of diverse datasets. This suggests immense potential for enabling automated/semi-automated annotation pipelines, thereby decreasing the burden of annotations facilitating several surgical applications.

Luis Wiedmann, Orr Zohar, Amir Mahla et al.

The advancement of vision-language models (VLMs) is hampered by a fragmented landscape of inconsistent and contaminated public datasets. We introduce FineVision, a meticulously collected, curated, and unified corpus of 24 million samples - the largest open resource of its kind. We unify more than 200 sources into 185 subsets via a semi-automated, human-in-the-loop pipeline: automation performs bulk ingestion and schema mapping, while reviewers audit mappings and spot-check outputs to verify faithful consumption of annotations, appropriate formatting and diversity, and safety; issues trigger targeted fixes and re-runs. The workflow further applies rigorous de-duplication within and across sources and decontamination against 66 public benchmarks. FineVision also encompasses agentic/GUI tasks with a unified action space; reviewers validate schemas and inspect a sample of trajectories to confirm executable fidelity. Models trained on FineVision consistently outperform those trained on existing open mixtures across a broad evaluation suite, underscoring the benefits of scale, data hygiene, and balanced automation with human oversight. We release the corpus and curation tools to accelerate data-centric VLM research.

Zihan Lin, Xiaohan Wang, Jie Cao et al.

Large language models (LLMs) transcend passive generation and act as goal-directed agents by invoking external tools. Reinforcement learning (RL) offers a principled framework for optimizing these emergent tool-use policies, yet the prevailing paradigm relies exclusively on sparse outcome rewards and lacks consideration of the particularity of tool-use tasks, inflating policy-gradient variance and resulting in inefficient training. To better understand and address these challenges, we first establish a theoretical link between policy entropy and training stability of tool-use tasks, which reveals that structured, low-entropy tokens are primary determinants of rewards. Motivated by this insight, we propose \textbf{Res}haped \textbf{T}oken-level policy gradients (\textbf{ResT}) for tool-use tasks. ResT reshapes the policy gradient through entropy-informed token reweighting, progressively upweighting reasoning tokens as training proceeds. This entropy-aware scheme enables a smooth shift from structural correctness to semantic reasoning and stabilizes convergence in multi-turn tool-use tasks. Evaluation on BFCL and API-Bank shows that ResT achieves state-of-the-art results, outperforming prior methods by up to $8.76\%$. When fine-tuned on a 4B base LLM, ResT further surpasses GPT-4o by $4.11\%$ on single-turn tasks and $1.50\%$ on multi-turn base tasks.

Xiaohan Wang, Yu Zhang, Guibin Jiang et al.

Marketing optimization, commonly formulated as an online budget allocation problem, has emerged as a pivotal factor in driving user growth. Most existing research addresses this problem by following the principle of 'first predict then optimize' for each individual, which presents challenges related to large-scale counterfactual prediction and solving complexity trade-offs. Note that the practical data quality is uncontrollable, and the solving scale tends to be tens of millions. Therefore, the existing approaches make the robust budget allocation non-trivial, especially in industrial scenarios with considerable data noise. To this end, this paper proposes a novel approach that solves the problem from the cluster perspective. Specifically, we propose a multi-task representation network to learn the inherent attributes of individuals and project the original features into high-dimension hidden representations through the first two layers of the trained network. Then, we divide these hidden representations into $K$ groups through partitioning-based clustering, thus reformulating the problem as an integer stochastic programming problem under different total budgets. Finally, we distill the representation module and clustering model into a multi-category model to facilitate online deployment. Offline experiments validate the effectiveness and superiority of our approach compared to six state-of-the-art marketing optimization algorithms. Online A/B tests on the Meituan platform indicate that the approach outperforms the online algorithm by 0.53% and 0.65%, considering order volume (OV) and gross merchandise volume (GMV), respectively.

Xiaohan Wang, Zhangtao Cheng, Ting Zhong et al.

Weight Averaging (WA) has emerged as a powerful technique for enhancing generalization by promoting convergence to a flat loss landscape, which correlates with stronger out-of-distribution performance. However, applying WA directly to multi-modal domain generalization (MMDG) is challenging: differences in optimization speed across modalities lead WA to overfit to faster-converging ones in early stages, suppressing the contribution of slower yet complementary modalities, thereby hindering effective modality fusion and skewing the loss surface toward sharper, less generalizable minima. To address this issue, we propose MBCD, a unified collaborative distillation framework that retains WA's flatness-inducing advantages while overcoming its shortcomings in multi-modal contexts. MBCD begins with adaptive modality dropout in the student model to curb early-stage bias toward dominant modalities. A gradient consistency constraint then aligns learning signals between uni-modal branches and the fused representation, encouraging coordinated and smoother optimization. Finally, a WA-based teacher conducts cross-modal distillation by transferring fused knowledge to each uni-modal branch, which strengthens cross-modal interactions and steer convergence toward flatter solutions. Extensive experiments on MMDG benchmarks show that MBCD consistently outperforms existing methods, achieving superior accuracy and robustness across diverse unseen domains.

Elaine Sui, Xiaohan Wang, Serena Yeung-Levy

Advancements in vision-language models (VLMs) have propelled the field of computer vision, particularly in the zero-shot learning setting. Despite their promise, the effectiveness of these models often diminishes due to domain shifts in test environments. To address this, we introduce the Test-Time Prototype Shifting (TPS) framework, a pioneering approach designed to adapt VLMs to test datasets using unlabeled test inputs. Our method is based on the notion of modulating per-class prototypes in the shared embedding space. By pre-computing and caching prototypes generated with the pre-trained text encoder, TPS not only facilitates optimization-free prototype reuse for subsequent predictions but also enables seamless integration with current advancements in prompt engineering. At test-time, TPS dynamically learns shift vectors for each prototype based solely on the given test sample, effectively bridging the domain gap and enhancing classification accuracy. A notable aspect of our framework is its significantly reduced memory and computational demands when compared to conventional text-prompt tuning methods. Extensive evaluations across 15 image classification datasets involving natural distribution shifts and cross-dataset generalization, as well as in context-dependent visual reasoning, demonstrate TPS's superior performance, achieving state-of-the-art results while reducing resource requirements.

Yunhong Lou, Linchao Zhu, Yaxiong Wang et al.

We present DiverseMotion, a new approach for synthesizing high-quality human motions conditioned on textual descriptions while preserving motion diversity.Despite the recent significant process in text-based human motion generation,existing methods often prioritize fitting training motions at the expense of action diversity. Consequently, striking a balance between motion quality and diversity remains an unresolved challenge. This problem is compounded by two key factors: 1) the lack of diversity in motion-caption pairs in existing benchmarks and 2) the unilateral and biased semantic understanding of the text prompt, focusing primarily on the verb component while neglecting the nuanced distinctions indicated by other words.In response to the first issue, we construct a large-scale Wild Motion-Caption dataset (WMC) to extend the restricted action boundary of existing well-annotated datasets, enabling the learning of diverse motions through a more extensive range of actions. To this end, a motion BLIP is trained upon a pretrained vision-language model, then we automatically generate diverse motion captions for the collected motion sequences. As a result, we finally build a dataset comprising 8,888 motions coupled with 141k text.To comprehensively understand the text command, we propose a Hierarchical Semantic Aggregation (HSA) module to capture the fine-grained semantics.Finally,we involve the above two designs into an effective Motion Discrete Diffusion (MDD) framework to strike a balance between motion quality and diversity. Extensive experiments on HumanML3D and KIT-ML show that our DiverseMotion achieves the state-of-the-art motion quality and competitive motion diversity. Dataset, code, and pretrained models will be released to reproduce all of our results.

Wenjie Zhuo, Yifan Sun, Xiaohan Wang et al.

This paper presents a whitening-based contrastive learning method for sentence embedding learning (WhitenedCSE), which combines contrastive learning with a novel shuffled group whitening. Generally, contrastive learning pulls distortions of a single sample (i.e., positive samples) close and push negative samples far away, correspondingly facilitating the alignment and uniformity in the feature space. A popular alternative to the "pushing'' operation is whitening the feature space, which scatters all the samples for uniformity. Since the whitening and the contrastive learning have large redundancy w.r.t. the uniformity, they are usually used separately and do not easily work together. For the first time, this paper integrates whitening into the contrastive learning scheme and facilitates two benefits. 1) Better uniformity. We find that these two approaches are not totally redundant but actually have some complementarity due to different uniformity mechanism. 2) Better alignment. We randomly divide the feature into multiple groups along the channel axis and perform whitening independently within each group. By shuffling the group division, we derive multiple distortions of a single sample and thus increase the positive sample diversity. Consequently, using multiple positive samples with enhanced diversity further improves contrastive learning due to better alignment. Extensive experiments on seven semantic textual similarity tasks show our method achieves consistent improvement over the contrastive learning baseline and sets new states of the art, e.g., 78.78\% (+2.53\% based on BERT\ba) Spearman correlation on STS tasks.

Jiayi Shao, Xiaohan Wang, Ruijie Quan et al.

Temporal action localization (TAL), which involves recognizing and locating action instances, is a challenging task in video understanding. Most existing approaches directly predict action classes and regress offsets to boundaries, while overlooking the discrepant importance of each frame. In this paper, we propose an Action Sensitivity Learning framework (ASL) to tackle this task, which aims to assess the value of each frame and then leverage the generated action sensitivity to recalibrate the training procedure. We first introduce a lightweight Action Sensitivity Evaluator to learn the action sensitivity at the class level and instance level, respectively. The outputs of the two branches are combined to reweight the gradient of the two sub-tasks. Moreover, based on the action sensitivity of each frame, we design an Action Sensitive Contrastive Loss to enhance features, where the action-aware frames are sampled as positive pairs to push away the action-irrelevant frames. The extensive studies on various action localization benchmarks (i.e., MultiThumos, Charades, Ego4D-Moment Queries v1.0, Epic-Kitchens 100, Thumos14 and ActivityNet1.3) show that ASL surpasses the state-of-the-art in terms of average-mAP under multiple types of scenarios, e.g., single-labeled, densely-labeled and egocentric.

Xiang Chen, Jintian Zhang, Xiaohan Wang et al.

Current Multimodal Knowledge Graph Construction (MKGC) models struggle with the real-world dynamism of continuously emerging entities and relations, often succumbing to catastrophic forgetting-loss of previously acquired knowledge. This study introduces benchmarks aimed at fostering the development of the continual MKGC domain. We further introduce MSPT framework, designed to surmount the shortcomings of existing MKGC approaches during multimedia data processing. MSPT harmonizes the retention of learned knowledge (stability) and the integration of new data (plasticity), outperforming current continual learning and multimodal methods. Our results confirm MSPT's superior performance in evolving knowledge environments, showcasing its capacity to navigate balance between stability and plasticity.

Xu Chen, Yahong Han, Xiaohan Wang et al.

Reducing redundancy is crucial for improving the efficiency of video recognition models. An effective approach is to select informative content from the holistic video, yielding a popular family of dynamic video recognition methods. However, existing dynamic methods focus on either temporal or spatial selection independently while neglecting a reality that the redundancies are usually spatial and temporal, simultaneously. Moreover, their selected content is usually cropped with fixed shapes, while the realistic distribution of informative content can be much more diverse. With these two insights, this paper proposes to integrate temporal and spatial selection into an Action Keypoint Network (AK-Net). From different frames and positions, AK-Net selects some informative points scattered in arbitrary-shaped regions as a set of action keypoints and then transforms the video recognition into point cloud classification. AK-Net has two steps, i.e., the keypoint selection and the point cloud classification. First, it inputs the video into a baseline network and outputs a feature map from an intermediate layer. We view each pixel on this feature map as a spatial-temporal point and select some informative keypoints using self-attention. Second, AK-Net devises a ranking criterion to arrange the keypoints into an ordered 1D sequence. Consequentially, AK-Net brings two-fold benefits for efficiency: The keypoint selection step collects informative content within arbitrary shapes and increases the efficiency for modeling spatial-temporal dependencies, while the point cloud classification step further reduces the computational cost by compacting the convolutional kernels. Experimental results show that AK-Net can consistently improve the efficiency and performance of baseline methods on several video recognition benchmarks.

Chao Sun, Zhedong Zheng, Xiaohan Wang et al.

The manual annotation for large-scale point clouds costs a lot of time and is usually unavailable in harsh real-world scenarios. Inspired by the great success of the pre-training and fine-tuning paradigm in both vision and language tasks, we argue that pre-training is one potential solution for obtaining a scalable model to 3D point cloud downstream tasks as well. In this paper, we, therefore, explore a new self-supervised learning method, called Mixing and Disentangling (MD), for 3D point cloud representation learning. As the name implies, we mix two input shapes and demand the model learning to separate the inputs from the mixed shape. We leverage this reconstruction task as the pretext optimization objective for self-supervised learning. There are two primary advantages: 1) Compared to prevailing image datasets, eg, ImageNet, point cloud datasets are de facto small. The mixing process can provide a much larger online training sample pool. 2) On the other hand, the disentangling process motivates the model to mine the geometric prior knowledge, eg, key points. To verify the effectiveness of the proposed pretext task, we build one baseline network, which is composed of one encoder and one decoder. During pre-training, we mix two original shapes and obtain the geometry-aware embedding from the encoder, then an instance-adaptive decoder is applied to recover the original shapes from the embedding. Albeit simple, the pre-trained encoder can capture the key points of an unseen point cloud and surpasses the encoder trained from scratch on downstream tasks. The proposed method has improved the empirical performance on both ModelNet-40 and ShapeNet-Part datasets in terms of point cloud classification and segmentation tasks. We further conduct ablation studies to explore the effect of each component and verify the generalization of our proposed strategy by harnessing different backbones.

Kezhou Lin, Xiaohan Wang, Zhedong Zheng et al.

Obtaining viewer responses from videos can be useful for creators and streaming platforms to analyze the video performance and improve the future user experience. In this report, we present our method for 2021 Evoked Expression from Videos Challenge. In particular, our model utilizes both audio and image modalities as inputs to predict emotion changes of viewers. To model long-range emotion changes, we use a GRU-based model to predict one sparse signal with 1Hz. We observe that the emotion changes are smooth. Therefore, the final dense prediction is obtained via linear interpolating the signal, which is robust to the prediction fluctuation. Albeit simple, the proposed method has achieved pearson's correlation score of 0.04430 on the final private test set.

Xiaohan Wang, Linchao Zhu, Yi Yang

Text-video retrieval is a challenging task that aims to search relevant video contents based on natural language descriptions. The key to this problem is to measure text-video similarities in a joint embedding space. However, most existing methods only consider the global cross-modal similarity and overlook the local details. Some works incorporate the local comparisons through cross-modal local matching and reasoning. These complex operations introduce tremendous computation. In this paper, we design an efficient global-local alignment method. The multi-modal video sequences and text features are adaptively aggregated with a set of shared semantic centers. The local cross-modal similarities are computed between the video feature and text feature within the same center. This design enables the meticulous local comparison and reduces the computational cost of the interaction between each text-video pair. Moreover, a global alignment method is proposed to provide a global cross-modal measurement that is complementary to the local perspective. The global aggregated visual features also provide additional supervision, which is indispensable to the optimization of the learnable semantic centers. We achieve consistent improvements on three standard text-video retrieval benchmarks and outperform the state-of-the-art by a clear margin.