Paper Info
Title
CRNet: Collaborative Refinement Network for Self-Supervised Video Object Segmentation
Abstract
Most self-supervised based methods just rely on a point-to-point correspondence strategy to propagate masks through a video sequence. However, the pixel level matching is not sufficient and often results in noise. To ease the problem, we propose our collaborative refinement network (CRNet) for self-supervised video object segmentation. Our collaborative refinement network consists of two modules, i.e., memory retrieval module and collaborative refinement module. The memory retrieval module is used to perform point-to-point correspondence and produce a propagated mask for a query frame. The collaborative refinement module is designed to aggregate the reference & query information and learn the collaborative relationship among them implicitly to refine the output of the memory retrieval module. The whole model is trained from unlabeled video data without any human annotation in a self-supervised manner. Extensive experiments conducted on DAVIS-17 and YouTube-VOS demonstrate that our CRNet surpasses the state-of-the-art self-supervised methods and narrows the gap with the fully supervised methods.
Year
DOI
Venue
2022
10.1109/MIPR54900.2022.00037
2022 IEEE 5th International Conference on Multimedia Information Processing and Retrieval (MIPR)
Keywords
DocType
ISSN
collaborative refinement network,CRNet,self-supervised video object segmentation,memory retrieval module,collaborative refinement module,collaborative relationship,unlabeled video data,self-supervised manner,state-of-the-art self-supervised methods,fully supervised methods,point-to-point correspondence strategy,video sequence,YouTube-VOS,DAVIS-17
Conference
2770-4327
ISBN
Citations 
PageRank 
978-1-6654-9549-3
0
0.34
References 
Authors
2
6
Name
Order
Citations
PageRank
Dexiang Hong101.01
Guorong Li200.34
Bineng Zhong310.69
Zhenjun Han417616.40
Li Su500.34
Qingming Huang600.34