Adversarial Domain Adaptation with Semantic Consistency for Cross-Domain Image Classification

• DOI: 10.1145/3357384.3357918
• Publication Date: 2019-11-04T14:11:35Z
• Authors: Manliang Cao; Xiangdong Zhou; Yiming Xu; Yue Pang; Bo Yao
• Keywords: Discriminative model; Domain Adaptation; Contextual image classification

In the cross-domain image classification scenario, domain adaption aims to address challenge of transferring knowledge obtained from source target that is regarded as similar but different domain. To get more reliable invariant representations, recent methods start consider class-level distribution alignment across and domains by adaptively assigning pseudo labels. However, these approaches are vulnerable error accumulation hence unable preserve category consistency. Because accuracy labels cannot be guaranteed explicitly. this paper, we propose Adversarial Domain Adaptation with Semantic Consistency (ADASC) model align discriminative features progressively effectively, via exploiting relations between domains. Specifically, simultaneously alleviate negative influence false pseudo-target features, introduce an Adaptive Centroid Alignment (ACA) strategy a Class Discriminative Constraint (CDC) step complement each other iteratively alternatively in end-to-end framework. Extensive experiments conducted on several unsupervised adaptation datasets, results show ADASC outperforms state-of-the-art methods.