🤖 AI Summary
To address domain shift in unsupervised domain adaptation for semantic segmentation—where labeled source and unlabeled target domains exhibit distributional divergence—this paper proposes a hard-sample-aware instance-adaptive pseudo-labeling framework. Our method comprises three key components: (1) an instance-adaptive pseudo-label selector that dynamically identifies high-quality pseudo-labels based on prediction confidence and local consistency; (2) hard-class-aware cross-image augmentation, which selectively mixes features in the embedding space to enhance discrimination of challenging classes; and (3) region-adaptive regularization, the first to extend consistency constraints beyond pseudo-labeled regions to unlabeled areas via dynamic spatial weighting. Evaluated on three standard benchmarks—GTA5→Cityscapes, SYNTHIA→Cityscapes, and Cityscapes→Oxford RobotCar—our approach consistently outperforms state-of-the-art methods. The implementation is publicly available.
📝 Abstract
The divergence between labeled training data and unlabeled testing data is a significant challenge for recent deep learning models. Unsupervised domain adaptation (UDA) attempts to solve such problem. Recent works show that self-training is a powerful approach to UDA. However, existing methods have difficulty in balancing the scalability and performance. In this paper, we propose a hard-aware instance adaptive self-training framework for UDA on the task of semantic segmentation. To effectively improve the quality and diversity of pseudo-labels, we develop a novel pseudo-label generation strategy with an instance adaptive selector. We further enrich the hard class pseudo-labels with inter-image information through a skillfully designed hard-aware pseudo-label augmentation. Besides, we propose the region-adaptive regularization to smooth the pseudo-label region and sharpen the non-pseudo-label region. For the non-pseudo-label region, consistency constraint is also constructed to introduce stronger supervision signals during model optimization. Our method is so concise and efficient that it is easy to be generalized to other UDA methods. Experiments on GTA5 Cityscapes, SYNTHIA Cityscapes, and Cityscapes Oxford RobotCar demonstrate the superior performance of our approach compared with the state-of-the-art methods. Our codes are available at https://github.com/bupt-ai-cz/HIAST.
