Semantic Concentration for Domain Adaptation

TL;DR

This paper introduces SCDA, a novel domain adaptation method that emphasizes principal features by pair-wise adversarial alignment, improving transfer performance by reducing irrelevant semantic interference.

Contribution

SCDA is a new regularization technique that enhances domain adaptation by focusing on key features through pair-wise adversarial training, adaptable to various existing methods.

Findings

SCDA improves domain adaptation accuracy on benchmark datasets.

It effectively suppresses irrelevant background features.

The method is compatible with multiple DA frameworks.

Abstract

Domain adaptation (DA) paves the way for label annotation and dataset bias issues by the knowledge transfer from a label-rich source domain to a related but unlabeled target domain. A mainstream of DA methods is to align the feature distributions of the two domains. However, the majority of them focus on the entire image features where irrelevant semantic information, e.g., the messy background, is inevitably embedded. Enforcing feature alignments in such case will negatively influence the correct matching of objects and consequently lead to the semantically negative transfer due to the confusion of irrelevant semantics. To tackle this issue, we propose Semantic Concentration for Domain Adaptation (SCDA), which encourages the model to concentrate on the most principal features via the pair-wise adversarial alignment of prediction distributions. Specifically, we train the classifier to class-wisely maximize the prediction distribution divergence of each sample pair, which enables the model to find the region with large differences among the same class of samples. Meanwhile, the feature extractor attempts to minimize that discrepancy, which suppresses the features of dissimilar regions among the same class of samples and accentuates the features of principal parts. As a general method, SCDA can be easily integrated into various DA methods as a regularizer to further boost their performance. Extensive experiments on the cross-domain benchmarks show the efficacy of SCDA.