Few-shot video object segmentation with prototype evolution

doi:10.1007/s00521-023-09325-y

	Few-shot video object segmentation with prototype evolution
	Mao, Binjie1,2 ; Liu, Xiyan1,2 ; Shi, Linsu 3; Yu, Jiazhong 3; Li, Fei 3; Xiang, Shiming1,2
发表期刊	NEURAL COMPUTING & APPLICATIONS
ISSN	0941-0643
	2024-01-03
页码	16
通讯作者	Xiang, Shiming(smxiang@nlpr.ia.ac.cn)
摘要	As a challenging task, few-shot video object segmentation attempts to segment objects of novel categories in the video while providing only a few annotated images. Current methods for this task only explore the relationship between support images and target query video ignoring the rich temporal information in the query video itself. To address this problem, we propose a simple yet effective framework named prototype evolution network (PENet) for few-shot video object segmentation in this paper. PENet first adopts a prototype-based structure which efficiently constructs and exploits the correlation between support images and target query video. Then a prototype evolution module is designed to summarize and propagate temporal information through the evolution process of the video prototype. The feature representation adopted by the module is of fixed size and does not increase memory burden as the video frame moves forward. Along with the category prototype extracted from the support set, the global video prototype provides guidance for the current frame segmentation. Additionally, the approach of utilizing the high-level features is introduced as an optional solution that trades a small amount of speed for higher accuracy. Experimental results on the Youtube-VIS dataset of 2019 version and 2021 version demonstrate that our PENet outperforms the previous methods with a sizable margin, validating the superiority of the proposed model.
关键词	Few-shot video segmentation Few-shot segmentation Video object segmentation
DOI	10.1007/s00521-023-09325-y
关键词[WOS]	NETWORK
收录类别	SCI
语种	英语
资助项目	National Key Research and Development Program of China[2018AAA0100400] ; National Key Research and Development Program of China[62071466] ; National Key Research and Development Program of China[62076242] ; National Key Research and Development Program of China[61976208] ; National Natural Science Foundation of China
项目资助者	National Key Research and Development Program of China ; National Key Research and Development Program of China ; National Natural Science Foundation of China
WOS研究方向	Computer Science
WOS类目	Computer Science, Artificial Intelligence
WOS记录号	WOS:001135243500002
出版者	SPRINGER LONDON LTD
引用统计
文献类型	期刊论文
条目标识符	http://ir.ia.ac.cn/handle/173211/54823
专题	多模态人工智能系统全国重点实验室
通讯作者	Xiang, Shiming
作者单位	1.Chinese Acad Sci, Inst Automat, State Key Lab Multimodal Artificial Intelligence S, Beijing 100049, Peoples R China 2.Univ Chinese Acad Sci, Sch Artificial Intelligence, Beijing, Peoples R China 3.China Tower Corp Ltd, Beijing 100195, Peoples R China
第一作者单位	中国科学院自动化研究所
通讯作者单位	中国科学院自动化研究所
推荐引用方式 GB/T 7714	Mao, Binjie,Liu, Xiyan,Shi, Linsu,et al. Few-shot video object segmentation with prototype evolution[J]. NEURAL COMPUTING & APPLICATIONS,2024:16.
APA	Mao, Binjie,Liu, Xiyan,Shi, Linsu,Yu, Jiazhong,Li, Fei,&Xiang, Shiming.(2024).Few-shot video object segmentation with prototype evolution.NEURAL COMPUTING & APPLICATIONS,16.
MLA	Mao, Binjie,et al."Few-shot video object segmentation with prototype evolution".NEURAL COMPUTING & APPLICATIONS (2024):16.