Exponential Information Bottleneck Theory Against Intra-Attribute Variations for Pedestrian Attribute Recognition

doi:10.1109/TIFS.2023.3311584

CORC > 自动化研究所 > 中国科学院自动化研究所 > 多模态人工智能系统全国重点实验室

	Exponential Information Bottleneck Theory Against Intra-Attribute Variations for Pedestrian Attribute Recognition
	Wu, Junyi 2,3; Huang, Yan 5; Gao, Min 8; Gao, Zhipeng 3; Zhao, Jianqiang 3; Shi, Jieming 7; Zhang, Anguo 1,4,6,7
刊名	IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY
	2023
卷号	18 页码:5623-5635
关键词	Feature extraction Pedestrians Mutual information Body regions Training Task analysis Semantics Pedestrian attribute recognition intra-attribute variations exponential information bottleneck
ISSN号	1556-6013
DOI	10.1109/TIFS.2023.3311584
通讯作者	Huang, Yan(yan.huang@cripac.ia.ac.cn) ; Zhang, Anguo(anguo.zhang@hotmail.com)
英文摘要	Multi-label pedestrian attribute recognition (PAR) involves assigning multiple attributes to pedestrian images captured by video surveillance cameras. Despite its importance, learning robust attribute-related features for PAR remains a challenge due to the large intra-attribute variations in the image space. These variations, which stem from changes in pedestrian poses, illumination conditions, and background noise, make extracted attribute-related features susceptible to irrelevant information or noise interference. Existing PAR methods rely on body prior extractors or attention mechanisms to locate attribute-correlation regions for extracting robust features. However, these methods may not be robust to intra-attribute variations, which limits their effectiveness. To address this challenge, we propose a novel and flexible PAR framework that leverages the exponential information bottleneck (ExpIB) approach. Our ExpIB-Net uses mutual information compression as the main penalty during the early stage of training, thereby eliminating irrelevant information. As training progresses, the mutual information penalty weakens and the Binary Cross-Entropy Loss (BCELoss) contributes to improving the PAR recognition accuracy. Our method can also be integrated into an attention module to form the AttExpIB-Net, which better handles intra-attribute variations for better performance. Additionally, our model-agnostic ExpIB approach is plug-and-play, requiring no additional computational overhead during inference. Experiments on several challenging PAR datasets show that our method outperforms state-of-the-art approaches.
资助项目	National Natural Science Foundation of China[62306311] ; Fellowship of China Post-Doctoral Science Foundation[62306001] ; International Post-Doctoral Exchange Fellowship Program (Talent-Introduction Program) of China[2022T150698] ; Special Research Assistant Program of the Chinese Academy of Sciences[YJ20210324] ; Public Security Artificial Intelligence Infrastructure Support Platform ; [E2S9180301]
WOS关键词	PERSON REIDENTIFICATION ; ATTENTION NETWORK
WOS研究方向	Computer Science ; Engineering
语种	英语
出版者	IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC
WOS记录号	WOS:001070669000003
资助机构	National Natural Science Foundation of China ; Fellowship of China Post-Doctoral Science Foundation ; International Post-Doctoral Exchange Fellowship Program (Talent-Introduction Program) of China ; Special Research Assistant Program of the Chinese Academy of Sciences ; Public Security Artificial Intelligence Infrastructure Support Platform
内容类型	期刊论文
源URL	[http://ir.ia.ac.cn/handle/173211/53125]
专题	多模态人工智能系统全国重点实验室
通讯作者	Huang, Yan; Zhang, Anguo
作者单位	1.Minist Educ, Res Ctr Autonomous Unmanned Syst Technol, Hefei 230030, Peoples R China 2.Xiamen Meiya Pico Informat Secur Res Inst Co Ltd, Xiamen 361000, Peoples R China 3.Xiamen Meiya Pico Informat Co Ltd, AI Res Ctr, Xiamen 361000, Peoples R China 4.Anhui Univ, Sch Artificial Intelligence, Hefei 230039, Peoples R China 5.Chinese Acad Sci CASIA, Ctr Res Intelligent Percept & Comp CRIPAC, Inst Automat, Natl Lab Pattern Recognit NLPR, Beijing 100190, Peoples R China 6.Anhui Prov Engn Res Ctr Unmanned Syst & Intellige, Hefei 230039, Peoples R China 7.Hong Kong Polytech Univ, Dept Comp, Hong Kong, Peoples R China 8.Fuzhou Univ, Coll Phys & Informat Engn, Fujian Key Lab Intelligent Proc & Wireless Transm, Fuzhou 350002, Peoples R China
推荐引用方式 GB/T 7714	Wu, Junyi,Huang, Yan,Gao, Min,et al. Exponential Information Bottleneck Theory Against Intra-Attribute Variations for Pedestrian Attribute Recognition[J]. IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY,2023,18:5623-5635.
APA	Wu, Junyi.,Huang, Yan.,Gao, Min.,Gao, Zhipeng.,Zhao, Jianqiang.,...&Zhang, Anguo.(2023).Exponential Information Bottleneck Theory Against Intra-Attribute Variations for Pedestrian Attribute Recognition.IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY,18,5623-5635.
MLA	Wu, Junyi,et al."Exponential Information Bottleneck Theory Against Intra-Attribute Variations for Pedestrian Attribute Recognition".IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY 18(2023):5623-5635.