Domain adaptation for semantic segmentation of road scenes via two-stage alignment of traffic elements

Yuan Gao, Yaochen Li, Hao Liao, Tenweng Zhang, Chao Qiu

Abstract: Highlights•A two-stage self-training framework alleviates domain shift at various levels.•A strategy aligns traffic elements, differentiating dynamic and static features.•Proposed method adaptively refines pseudo labels using uncertainty estimation.•Proposed method achieves better performance on the challenging benchmark datasets.