Joint Token and Feature Alignment Framework for Text-Based Person Search

doi:10.1109/LSP.2022.3217682

CORC > 自动化研究所 > 中国科学院自动化研究所 > 智能感知与计算研究中心

	Joint Token and Feature Alignment Framework for Text-Based Person Search
	Li, Shangze 3; Lu, Andong 3; Huang, Yan 1; Li, Chenglong 2; Wang, Liang 1
刊名	IEEE SIGNAL PROCESSING LETTERS
	2022
卷号	29 页码:2238-2242
关键词	Feature extraction Visualization Representation learning Logic gates Image reconstruction Transformers Training Cross-modal generation feature alignment text-based person search token alignment transformer
ISSN号	1070-9908
DOI	10.1109/LSP.2022.3217682
通讯作者	Li, Chenglong(lcl1314@foxmail.com)
英文摘要	Text-based person search is a challenging cross-modal retrieval task. Existing works reduce the inter-modality and intra-class gaps by aligning local features extracted from image and text modalities, which easily lead to mismatching problems due to the lack of annotation information. Besides, it is sub-optimal to reduce two gaps simultaneously in the same feature space. This work proposes a novel joint token and feature alignment framework to reduce the inter-modality and intra-class gaps progressively. Specifically, we first build a dual-path feature learning network to extract features and conduct feature alignment to reduce the inter-modality gap. Second, we design a text generation module to generate token sequences using visual features, and then token alignment is performed to reduce the intra-class gap. Last, a fusion interaction module is introduced to further eliminate the modality heterogeneity using the strategy of multi-stage feature fusion. Extensive experiments on the CUHK-PEDES dataset demonstrate the effectiveness of our model, which significantly outperforms previous state-of-the-art methods.
资助项目	National Natural Science Foundation of China[61976003] ; National Natural Science Foundation of China[62076003] ; Anhui Provincial Key Research and Development Program[202104d07020008] ; Open Project Program of the National Laboratory of Pattern Recognition (NLPR) ; Gaofeng Discipline Construction Project (Computer Science and Technology)[Z010111016]
WOS研究方向	Engineering
语种	英语
出版者	IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC
WOS记录号	WOS:000880641600004
资助机构	National Natural Science Foundation of China ; Anhui Provincial Key Research and Development Program ; Open Project Program of the National Laboratory of Pattern Recognition (NLPR) ; Gaofeng Discipline Construction Project (Computer Science and Technology)
内容类型	期刊论文
源URL	[http://ir.ia.ac.cn/handle/173211/50678]
专题	自动化研究所_智能感知与计算研究中心
通讯作者	Li, Chenglong
作者单位	1.Chinese Acad Sci, Inst Automat, Natl Lab Pattern Recognit, Beijing 100190, Peoples R China 2.Anhui Univ, Sch Artificial Intelligence, Informat Mat & Intelligent Sensing Lab Anhui Prov, Anhui Prov Key Lab Multimodal Cognit Computat, Hefei 230601, Peoples R China 3.Anhui Univ, Sch Comp Sci & Technol, Anhui Prov Key Lab Multimodal Cognit Computat, Hefei 230601, Peoples R China
推荐引用方式 GB/T 7714	Li, Shangze,Lu, Andong,Huang, Yan,et al. Joint Token and Feature Alignment Framework for Text-Based Person Search[J]. IEEE SIGNAL PROCESSING LETTERS,2022,29:2238-2242.
APA	Li, Shangze,Lu, Andong,Huang, Yan,Li, Chenglong,&Wang, Liang.(2022).Joint Token and Feature Alignment Framework for Text-Based Person Search.IEEE SIGNAL PROCESSING LETTERS,29,2238-2242.
MLA	Li, Shangze,et al."Joint Token and Feature Alignment Framework for Text-Based Person Search".IEEE SIGNAL PROCESSING LETTERS 29(2022):2238-2242.