Bidirectional Adversarial Domain Adaptation with Semantic Consistency

Abstract

Unsupervised domain adaptation (DA) aims to utilize the well-annotated source domain data to recognize the unlabeled target domain data that usually have a large domain shift. Most existing DA methods are developed to align the high-level feature-space distribution between the source and target domains, while neglecting the semantic consistency and low-level pixel-space information. In this paper, we propose a novel bidirectional adversarial domain adaptation (BADA) method to simultaneously adapt the pixel-level and feature-level shifts with semantic consistency. To keep semantic consistency, we propose a soft label-based semantic consistency constraint, which takes advantage of the well-trained source classifier during bidirectional adversarial mappings. Furthermore, the semantic consistency has been first analyzed during the domain adaptation with regard to both qualitative and quantitative evaluation. Systematic experiments on four benchmark datasets show that the proposed BADA achieves the state-of-the-art performance.