Abstract
Unsupervised domain adaptation (UDA) aims to bridge the domain shift between the labeled source domain and the unlabeled target domain. However, most existing works perform the global-level feature alignment for semantic segmentation, while the local consistency between the regions has been largely neglected, and these methods are less robust to changing of outdoor environments. Motivated by the above facts, we propose a novel and fully end-to-end trainable approach, called regional contrastive consistency regularization (RCCR) for domain adaptive semantic segmentation. Our core idea is to pull the similar regional features extracted from the same location of different images to be closer, and meanwhile push the features from the different locations of the two images to be separated. We innovatively propose momentum projector heads, where the teacher projector is the exponential moving average of the student. Besides, we present a region-wise contrastive loss with two sampling strategies to realize effective regional consistency. Finally, a memory bank mechanism is designed to learn more robust and stable region-wise features under varying environments. Extensive experiments on two common UDA benchmarks, i.e., GTAV to Cityscapes and SYNTHIA to Cityscapes, demonstrate that our approach outperforms the state-of-the-art methods.
Abstract (translated)
URL
https://arxiv.org/abs/2110.05170
