Efficient spatiotemporal context modeling for action recognition

doi:10.1016/j.neucom.2023.126289

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

	Efficient spatiotemporal context modeling for action recognition
	Cao, Congqi 2,3,4; Lu, Yue 2; Zhang, Yifan 1,5,6; Jiang, Dongmei 2; Zhang, Yanning 2
刊名	NEUROCOMPUTING
	2023-08-07
卷号	545 页码:13
关键词	Action recognition Long -range context modeling Spatiotemporal feature map Attention module Relation
ISSN号	0925-2312
DOI	10.1016/j.neucom.2023.126289
通讯作者	Cao, Congqi(congqi.cao@nwpu.edu.cn)
英文摘要	Contextual information is essential in action recognition. However, local operations have difficulty in modeling two distant elements, and directly computing the dense relations between any two points brings huge computation and memory burden. Inspired by the recurrent 2D criss-cross attention (RCCA-2D) in image segmentation, we propose a recurrent 3D criss-cross attention (RCCA-3D) that factorizes the global relation map into sparse relation maps to model long-range spatiotemporal context with minor costs for video-based action recognition. Specifically, we first propose a 3D criss-cross attention (CCA-3D) module. Compared with the CCA-2D which only works in space, it can capture the spatiotemporal relationship between the points in the same line along the direction of width, height and time. However, only replacing the two CCA-2Ds in the RCCA-2D with our CCA-3Ds cannot model the spatiotemporal context in videos. Therefore, we further duplicate the CCA-3D with a recurrent mechanism to transmit the relation between the points in a line to a plane and finally to the whole spatiotemporal space. To make the RCCA-3D adaptive for action recognition, we propose a novel recurrent structure rather than directly extending the original 2D structure to 3D. In the experiments, we make a thorough analysis of different structures of RCCA-3D, verifying the proposed structure is more suitable for action recognition. We also compare our RCCA-3D with the non-local attention, showing that the RCCA-3D requires 25% fewer parameters and 30% fewer FLOPs with even higher accuracy. Finally, equipped with our RCCA-3D, 3 networks achieve better and leading performance on 5 RGB-based and skeleton-based datasets.
资助项目	National Natural Science Foundation of China[U19B2037] ; National Natural Science Foundation of China[62273347] ; National Natural Science Foundation of China[61906155] ; National Key R&D Program of China[2020AAA0106900] ; Key R&D Project in Shaanxi Province[2023-YBGY-240] ; Young Talent Fund of Association for Science and Technology in Shaanxi, China[20220117]
WOS研究方向	Computer Science
语种	英语
出版者	ELSEVIER
WOS记录号	WOS:001000901900001
资助机构	National Natural Science Foundation of China ; National Key R&D Program of China ; Key R&D Project in Shaanxi Province ; Young Talent Fund of Association for Science and Technology in Shaanxi, China
内容类型	期刊论文
源URL	[http://ir.ia.ac.cn/handle/173211/53432]
专题	自动化研究所_智能感知与计算研究中心
通讯作者	Cao, Congqi
作者单位	1.Chinese Acad Sci, Inst Automat, Natl Lab Pattern Recognit, Beijing, Peoples R China 2.Northwestern Polytech Univ, Sch Comp Sci, Xian, Peoples R China 3.Northwestern Polytech Univ, Natl Engn Lab Integrated Aerosp Ground Ocean Big D, Xian 710129, Peoples R China 4.Northwestern Polytech Univ, Sch Comp Sci, Shaanxi Prov Key Lab Speech & Image Informat Proc, Xian 710129, Peoples R China 5.Chinese Acad Sci, Inst Automat, Natl Lab Pattern Recognit, Beijing 100190, Peoples R China 6.Univ Chinese Acad Sci, Beijing 100190, Peoples R China
推荐引用方式 GB/T 7714	Cao, Congqi,Lu, Yue,Zhang, Yifan,et al. Efficient spatiotemporal context modeling for action recognition[J]. NEUROCOMPUTING,2023,545:13.
APA	Cao, Congqi,Lu, Yue,Zhang, Yifan,Jiang, Dongmei,&Zhang, Yanning.(2023).Efficient spatiotemporal context modeling for action recognition.NEUROCOMPUTING,545,13.
MLA	Cao, Congqi,et al."Efficient spatiotemporal context modeling for action recognition".NEUROCOMPUTING 545(2023):13.