BLPSeg: Balance the Label Preference in Scribble-Supervised Semantic Segmentation

doi:10.1109/TIP.2023.3301342

	BLPSeg: Balance the Label Preference in Scribble-Supervised Semantic Segmentation
	Wang, Yude 1; Zhang, Jie 1; Kan, Meina 1; Shan, Shiguang 1,2,3; Chen, Xilin 1
	2023
发表期刊	IEEE TRANSACTIONS ON IMAGE PROCESSING
ISSN	1057-7149
卷号	32 页码:4921-4934
摘要	Scribble-supervised semantic segmentation is an appealing weakly supervised technique with low labeling cost. Existing approaches mainly consider diffusing the labeled region of scribble by low-level feature similarity to narrow the supervision gap between scribble labels and mask labels. In this study, we observe an annotation bias between scribble and object mask, i.e., label workers tend to scribble on the spacious region instead of corners. This label preference makes the model learn well on those frequently labeled regions but poor on rarely labeled pixels. Therefore, we propose BLPSeg to balance the label preference for complete segmentation. Specifically, the BLPSeg first predicts an annotation probability map to evaluate the rarity of labels on each image, then utilizes a novel BLP loss to balance the model training by up-weighting those rare annotations. Additionally, to further alleviate the impact of label preference, we design a local aggregation module (LAM) to propagate supervision from labeled to unlabeled regions in gradient backpropagation. We conduct extensive experiments to illustrate the effectiveness of our BLPSeg. Our single-stage method even outperforms other advanced multi-stage methods and achieves state-of-the-art performance.
关键词	Annotations Semantic segmentation Training Task analysis Head Semantics Costs Scribble-supervised weakly supervised semantic segmentation
DOI	10.1109/TIP.2023.3301342
收录类别	SCI
语种	英语
资助项目	National Key Research and Development Program of China[2021YFC3310100] ; National Natural Science Foundation of China[62176251] ; National Natural Science Foundation of China[62122074] ; National Natural Science Foundation of China[61976219]
WOS研究方向	Computer Science ; Engineering
WOS类目	Computer Science, Artificial Intelligence ; Engineering, Electrical & Electronic
WOS记录号	WOS:001067700800002
出版者	IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC
引用统计	被引频次：2[WOS] [WOS记录] [WOS相关记录]
文献类型	期刊论文
条目标识符	http://119.78.100.204/handle/2XEOYT63/21152
专题	中国科学院计算技术研究所期刊论文_英文
通讯作者	Zhang, Jie
作者单位	1.Chinese Acad Sci, Inst Comp Technol, Key Lab Intelligent Informat Proc, CAS, Beijing 100190, Peoples R China 2.Univ Chinese Acad Sci, Beijing 100049, Peoples R China 3.Peng Cheng Natl Lab, Shenzhen 518055, Peoples R China
推荐引用方式 GB/T 7714	Wang, Yude,Zhang, Jie,Kan, Meina,et al. BLPSeg: Balance the Label Preference in Scribble-Supervised Semantic Segmentation[J]. IEEE TRANSACTIONS ON IMAGE PROCESSING,2023,32:4921-4934.
APA	Wang, Yude,Zhang, Jie,Kan, Meina,Shan, Shiguang,&Chen, Xilin.(2023).BLPSeg: Balance the Label Preference in Scribble-Supervised Semantic Segmentation.IEEE TRANSACTIONS ON IMAGE PROCESSING,32,4921-4934.
MLA	Wang, Yude,et al."BLPSeg: Balance the Label Preference in Scribble-Supervised Semantic Segmentation".IEEE TRANSACTIONS ON IMAGE PROCESSING 32(2023):4921-4934.