Overcoming learning bias via Prototypical Feature Compensation for source-free domain adaptation

Zicheng Pan, Xiaohan Yu, Weichuan Zhang, Yongsheng Gao

Abstract: Highlights•Develop a prototypical feature compensation method for source-free domain adaptation.•Extract representative source domain features to compensate target domain features.•Minimize feature discrepancy between source and target domain data.•Demonstrate effective correction of learning bias in experiments.•Achieve state-of-the-art performance on Office-31, Office-Home, and DomainNet-126.