Semantically Meaningful Class Prototype Learning for One-Shot Image Segmentation

One-shot semantic image segmentation aims to segment the object regions for novel class with only one annotated image. Recent works adopt episodic training strategy mimic expected situation at testing time. However, these existing approaches simulate test conditions too strictly during process, and thus cannot make full use of given label information. Besides, mainly focus on foreground-background target setting. They utilize binary mask labels training. In this paper, we propose leverage multi-class information It will encourage network generate more semantically meaningful features each category. After integrating cues into query features, then a pyramid feature fusion module mine fused final classifier. Furthermore, take advantage support image-mask pair, self-prototype guidance branch segmentation. can constrain generating compact robust prototype class. For inference, Specifically, prediction extract pseudo-prototype combine it initial prototype. Then guide Extensive experiments demonstrate superiority our proposed approach. The source codes models have been made available https://github.com/NUST-Machine-Intelligence-Laboratory/SMCP .