Unsupervised Contrastive Domain Adaptation for Semantic Segmentation

TL;DR

This paper presents an unsupervised contrastive learning method for domain adaptation in semantic segmentation, improving feature alignment across domains and enhancing performance on challenging datasets.

Contribution

It introduces a novel contrastive learning framework with in-domain and cross-domain pairs, plus a label expansion technique for better adaptation.

Findings

Achieves 60.2% mIoU on Cityscapes with GTA5 training data

Outperforms state-of-the-art domain adaptation methods

Effectively discovers hard class samples during adaptation

Abstract

Semantic segmentation models struggle to generalize in the presence of domain shift. In this paper, we introduce contrastive learning for feature alignment in cross-domain adaptation. We assemble both in-domain contrastive pairs and cross-domain contrastive pairs to learn discriminative features that align across domains. Based on the resulting well-aligned feature representations we introduce a label expansion approach that is able to discover samples from hard classes during the adaptation process to further boost performance. The proposed approach consistently outperforms state-of-the-art methods for domain adaptation. It achieves 60.2% mIoU on the Cityscapes dataset when training on the synthetic GTA5 dataset together with unlabeled Cityscapes images.