Bi-classifier with neighborhood aggregation for unsupervised domain adaptation

Shuai Lü, Xinyu Zhang, Zongze Li, Jingyao Li, Meng Kang

Abstract: Highlights•We propose a novel framework based on bi-classifier.•We propose the neighborhood aggregation to get high-quality pseudo-labels.•We propose a self-supervised loss with two pseudo-labels based on weights.•We propose the cross consistency regularization strategy.•BCNA achieves state-of-the-art performance on three popular datasets.