Learning De-biased prototypes for Few-shot Medical Image Segmentation

Yazhou Zhu, Ziming Cheng, Shidong Wang, Haofeng Zhang

Abstract: Highlights•A De-biasing Masked Average Pooling method is proposed to compute more accurate prototypes.•A Learnable Threshold Generation module is designed to generate adaptive thresholds for classifying background and objects.•The proposed model can achieve SoTA performance on three popular medical image segmentation datasets.