ProxyMix: Proxy-based Mixup training with label refinery for source-free domain adaptation

Yuhe Ding, Lijun Sheng, Jian Liang, Aihua Zheng, Ran He

Abstract: Highlights•We propose a simple method, ProxyMix, for source-free domain adaptation.•We exploit the weights of the source model to construct a proxy domain.•We propose a new frequency-weighted aggregation strategy for pseudo-labeling.