检索结果-内蒙古大学图书馆

LVAgent: Long Video Understanding by Multi-Round Dynamical Collaboration of MLLM Agents

学校读者我要写书评

暂无评论

arXiv 2025年

作者： Chen, Boyu Yue, Zhengrong Chen, Siran Wang, Zikang Liu, Yang Li, Peng Wang, Yali Shenzhen Key Lab of Computer Vision and Pattern Recognition Shenzhen Institutes of Advanced Technology Chinese Academy of Sciences China School of Artificial Intelligence University of Chinese Academy of Sciences China Tsinghua University Beijing China Dept. of Comp. Sci. & Tech. Institute for AI Tsinghua University Beijing China Shanghai Artificial Intelligence Laboratory China Shanghai Jiao Tong University China

Existing Multimodal Large Language Models (MLLMs) encounter significant challenges in modeling the temporal context within long videos. Currently, mainstream Agent-based methods use external tools (e.g., search engine, memory banks, OCR, retrieval models) 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 methodology 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, improving the coverage of critical temporal segments while maintaining computational efficiency. 3) Action: Agents answer long video-related 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 (including GPT-4o) and open-source models (including 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, on the LongVideoBench dataset, LVAgent improves accuracy by up to 14.3% compared with SOTA. © 2025, CC BY-NC-SA.

关键词： Open systems

Periodic Action Temporal Localization Method Based on Two-Path Architecture for Product Counting in Sewing Video 15th

学校读者我要写书评

暂无评论

Periodic Action Temporal Localization Method Based on Two-Pa...

15th International Conference on Intelligent Computing, ICIC 2019

作者： Huang, Jin-Long Zhang, Hong-Bo Du, Ji-Xiang Zheng, Jian-Feng Peng, Xiao-Xiao Department of Computer Science and Technology Huaqiao University Xiamen China Xiamen Key Laboratory of Computer Vision and Pattern Recognition Huaqiao University Xiamen China

ISBN: (纸本)9783030267650

Automatically product counting in the handmade process plays a vital role in the manufacturing industry, especially at the sewing industry. Nevertheless, there is currently a few methods to count the product number in the hand sewing process from surveillance video automatically. Due to the sewing procedure is a cyclical action, the product counting in hand sewing process is regarded as periodic action temporal localization and counting problem. In this paper, in order to solve this problem, we propose a novel two-path method, based on pose estimation and region-based convolutional neural network. The pose estimation method is used to obtain the trajectory information of human joint points, and the periodic action is located by detecting the periodic changes in joint trajectory. An effective two-threshold method is proposed to locate each action and count the number of periodic action from the trajectory information. To more accurately localization, we use a convolutional neural network to predict whether the workbench is empty or not. We fuse the results of joint trajectory and the status of the workbench to adjust the final periodic action localization and counting the number. To verify the proposed method, we built a new video database collected in the real sewing industry. The experimental results show that the proposed method is effective and constructive at periodic action localization and counting in the video for the sewing industry. © 2019, Springer Nature Switzerland AG.

关键词： Neural networks

FDDH: Fast discriminative discrete hashing for large-scale cross-modal retrieval

学校读者我要写书评

暂无评论

arXiv 2021年

作者： Liu, Xin Wang, Xingzhi Cheung, Yiu-Ming Department of Computer Science Huaqiao University Xiamen Key Laboratory of Computer Vision and Pattern Recognition Fujian Key Laboratory of Big Data Intelligence and Security Xiamen361021 China School of Electronics and Information Technology Sun Yat-sen University Guangzhou510006 China Department of Computer Science Hong Kong Baptist University Hong Kong Hong Kong

Cross-modal hashing, favored for its effectiveness and efficiency, has received wide attention to facilitating efficient retrieval across different modalities. Nevertheless, most existing methods do not sufficiently exploit the discriminative power of semantic information when learning the hash codes, while often involving time-consuming training procedure for handling the large-scale dataset. To tackle these issues, we formulate the learning of similarity-preserving hash codes in terms of orthogonally rotating the semantic data so as to minimize the quantization loss of mapping such data to hamming space, and propose an efficient Fast Discriminative Discrete Hashing (FDDH) approach for large-scale cross-modal retrieval. More specifically, FDDH introduces an orthogonal basis to regress the targeted hash codes of training examples to their corresponding semantic labels, and utilizes ϵ-dragging technique to provide provable large semantic margins. Accordingly, the discriminative power of semantic information can be explicitly captured and maximized. Moreover, an orthogonal transformation scheme is further proposed to map the nonlinear embedding data into the semantic subspace, which can well guarantee the semantic consistency between the data feature and its semantic representation. Consequently, an efficient closedform solution is derived for discriminative hash code learning, which is very computationally efficient. In addition, an effective and stable online learning strategy is presented for optimizing modality-specific projection functions, featuring adaptivity to different training sizes and streaming data. The proposed FDDH approach theoretically approximates the bi-Lipschitz continuity, runs sufficiently fast, and also significantly improves the retrieval performance over the state-of-the-art methods. The source code is released at: https://***/starxliu/FDDH. © 2021, CC0.

关键词： Hash functions

Low-Resolution Action recognition for Tiny Actions Challenge

学校读者我要写书评

暂无评论

arXiv 2022年

作者： Chen, Boyu Qiao, Yu Wang, Yali ShenZhen Key Lab of Computer Vision and Pattern Recognition Shenzhen Institute of Advanced Technology Chinese Academy of Sciences China University of Chinese Academy of Sciences China Shanghai AI Laboratory Shanghai China SIAT Branch Shenzhen Institute of Artificial Intelligence and Robotics for Society China

Tiny Actions Challenge focuses on understanding human activities in real-world surveillance. Basically, there are two main difficulties for activity recognition in this scenario. First, human activities are often recorded at a distance, and appear in a small resolution without much discriminative clue. Second, these activities are naturally distributed in a long-tailed way. It is hard to alleviate data bias for such heavy category imbalance. To tackle these problems, we propose a comprehensive recognition solution in this paper. First, we train video backbones with data balance, in order to alleviate overfitting in the challenge benchmark. Second, we design a dual-resolution distillation framework, which can effectively guide low-resolution action recognition by super-resolution knowledge. Finally, we apply model ensemble with post-processing, which can further boost performance on the long-tailed categories. Our solution ranks Top-1 on the leaderboard. Copyright © 2022, The Authors. All rights reserved.

关键词： Distillation

UDC-UNet: Under-Display Camera Image Restoration via U-shape Dynamic Network

学校读者我要写书评

暂无评论

arXiv 2022年

作者： Liu, Xina Hu, Jinfan Chen, Xiangyu Dong, Chao Shenzhen Key Lab of Computer Vision and Pattern Recognition SIAT-SenseTime Joint Lab Shenzhen Institutes of Advanced Technology Chinese Academy of Sciences Shanghai China University of Chinese Academy of Sciences Shanghai China University of Macau Shanghai China Shanghai AI Laboratory Shanghai China

Under-Display Camera (UDC) has been widely exploited to help smartphones realize full-screen displays. However, as the screen could inevitably affect the light propagation process, the images captured by the UDC system usually contain flare, haze, blur, and noise. Particularly, flare and blur in UDC images could severely deteriorate the user experience in high dynamic range (HDR) scenes. In this paper, we propose a new deep model, namely UDC-UNet, to address the UDC image restoration problem with an estimated PSF in HDR scenes. Our network consists of three parts, including a U-shape base network to utilize multi-scale information, a condition branch to perform spatially variant modulation, and a kernel branch to leverage the prior knowledge of the PSF. According to the characteristics of HDR data, we additionally design a tone mapping loss to stabilize network optimization and achieve better visual quality. Experimental results show that the proposed UDC-UNet outperforms the state-of-the-art methods in quantitative and qualitative comparisons. Our approach won second place in the UDC image restoration track of the MIPI challenge. Codes and models are available at https://***/J-FHu/UDCUNet. © 2022, CC BY.

关键词： Image reconstruction

Digging into Uncertainty in Self-supervised Multi-view Stereo

学校读者我要写书评

暂无评论

Digging into Uncertainty in Self-supervised Multi-view Stere...

International Conference on computer vision (ICCV)

作者： Hongbin Xu Zhipeng Zhou Yali Wang Wenxiong Kang Baigui Sun Hao Li Yu Qiao ShenZhen Key Lab of Computer Vision and Pattern Recognition Shenzhen Institute of Advanced Technology Chinese Academy of Sciences South China University of Technology Alibaba Group Pazhou Laboratory Shanghai AI Laboratory

ISBN: (纸本)9781665428132

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 framework 1 achieves the best performance among unsupervised MVS methods, with competitive performance with its supervised opponents.

关键词： Optical losses Optical filters computer vision Uncertainty Three-dimensional displays Monte Carlo methods Benchmark testing

Digging into uncertainty in self-supervised multi-view stereo

学校读者我要写书评

暂无评论

arXiv 2021年

作者： Xu, Hongbin Zhou, Zhipeng Wang, Yali Kang, Wenxiong Sun, Baigui Li, Hao Qiao, Yu ShenZhen Key Lab of Computer Vision and Pattern Recognition Shenzhen Institute of Advanced Technology Chinese Academy of Sciences South China University of Technology Shanghai AI Laboratory Alibaba Group Pazhou Laboratory

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 (UMVS) 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 framework achieves the best performance among unsupervised MVS methods, with competitive performance with its supervised opponents. © 2021, CC BY.

关键词： Image reconstruction

NTIRE 2023 Image Shadow Removal Challenge Report

学校读者我要写书评

暂无评论

NTIRE 2023 Image Shadow Removal Challenge Report

2023 IEEE/CVF Conference on computer vision and pattern recognition Workshops, CVPRW 2023

作者： Vasluianu, Florin-Alexandru Seizinger, Tim Timofte, Radu Cui, Shuhao Huang, Junshi Tian, Shuman Fan, Mingyuan Zhang, Jiaqi Zhu, Li Wei, Xiaoming Wei, Xiaolin Luo, Ziwei Gustafsson, Fredrik K. Zhao, Zheng Sjölund, Jens Schön, Thomas B. Dong, Xiaoyi Zhang, Xi Sheryl Li, Chenghua Leng, Cong Yeo, Woon-Ha Oh, Wang-Taek Lee, Yeo-Reum Ryu, Han-Cheol Luo, Jinting Jiang, Chengzhi Han, Mingyan Wu, Qi Lin, Wenjie Yu, Lei Li, Xinpeng Jiang, Ting Fan, Haoqiang Liu, Shuaicheng Xu, Shuning Song, Binbin Chen, Xiangyu Zhang, Shile Zhou, Jiantao Zhang, Zhao Zhao, Suiyi Zheng, Huan Gao, Yangcheng Wei, Yanyan Wang, Bo Ren, Jiahuan Luo, Yan Kondo, Yuki Miyata, Riku Yasue, Fuma Naruki, Taito Ukita, Norimichi Chang, Hua-En Yang, Hao-Hsiang Chen, Yi-Chung Chiang, Yuan-Chun Huang, Zhi-Kai Chen, Wei-Ting Chen, I-Hsiang Hsieh, Chia-Hsuan Kuo, Sy-Yen Xianwei, Li Fu, Huiyuan Liu, Chunlin Ma, Huadong Fu, Binglan He, Huiming Wang, Mengjia She, Wenxuan Liu, Yu Nathan, Sabari Kansal, Priya Zhang, Zhongjian Yang, Huabin Wang, Yan Zhang, Yanru Phutke, Shruti S. Kulkarni, Ashutosh Khan, Md Raqib Murala, Subrahmanyam Vipparthi, Santosh Kumar Ye, Heng Liu, Zixi Yang, Xingyi Liu, Songhua Wu, Yinwei Jing, Yongcheng Yu, Qianhao Zheng, Naishan Huang, Jie Long, Yuhang Yao, Mingde Zhao, Feng Zhao, Bowen Ye, Nan Shen, Ning Cao, Yanpeng Xiong, Tong Xia, Weiran Li, Dingwen Xia, Shuchen Computer Vision Lab Ifi Caidas University of Würzburg Germany Computer Vision Lab Eth Zürich Switzerland Meituan Group China Department of Information Technology Uppsala University Sweden Institute of Automation Chinese Academy of Sciences Beijing China Nanjing China Maicro Nanjing China Department of Artificial Intelligence Convergence Sahmyook University Seoul Korea Republic of Megvii Technology China University of Electronic Science and Technology of China China University of Macau China China Toyota Technological Institute Japan Graduate Institute of Electronics Engineering National Taiwan University Taiwan Department of Electrical Engineering National Taiwan University Taiwan Graduate Institute of Communication Engineering National Taiwan University Taiwan ServiceNow United States Beijing University of Post and Teleconmunication Beijing China Key Laboratory of Intelligent Perception and Image Understanding of Ministry of Education China Couger Inc. Computer Vision and Pattern Recognition Lab Indian Institute of Technology Ropar Punjab Rupnagar India Research Institute Singapore National University of Singapore Singapore Research Institute Singapore University of Sydney Australia Brain-Inspired Vision Laboratory Information Science and Technology Institution University of Science and Technology of China China State Key Laboratory of Fluid Power and Mechatronic Systems School of Mechanical Engineering Zhejiang University Hangzhou310027 China Key Laboratory of Advanced Manufacturing Technology of Zhejiang Province School of Mechanical Engineering Zhejiang University Hangzhou310027 China South China University of Technology China

ISBN: (纸本)9798350302493

This work reviews the results of the NTIRE 2023 Challenge on Image Shadow Removal. The described set of solutions were proposed for a novel dataset, which captures a wide range of object-light interactions. It consists of 1200 roughly pixel aligned pairs of real shadow free and shadow affected images, captured in a controlled environment. The data was captured in a white-box setup, using professional equipment for lights and data acquisition sensors. The challenge had a number of 144 participants registered, out of which 19 teams were compared in the final ranking. The proposed solutions extend the work on shadow removal, improving over the performance level describing state-of-the-art methods. © 2023 IEEE.

关键词： Data acquisition

Learning dynamical human-joint affinity for 3D pose estimation in videos

学校读者我要写书评

暂无评论

arXiv 2021年

作者： Zhang, Junhao Wang, Yali Zhou, Zhipeng Luan, Tianyu Wang, Zhe Qiao, Yu ShenZhen Key Lab of Computer Vision and Pattern Recognition Shenzhen Institute of Advanced Technology Chinese Academy of Sciences University of California Irvine United States Shanghai AI Laboratory Shanghai China

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. Copyright © 2021, The Authors. All rights reserved.

关键词： Convolution