Unsupervised cross domain semantic segmentation with mutual refinement and information distillation

Dexin Ren, Shidong Wang, Zheng Zhang, Wankou Yang, Mingwu Ren, Haofeng Zhang

Abstract: Highlights•A Mutual Refinement module is proposed to interact information for cross domain semantic segmentation.•An Information Distillation module is employed to simplify the inference network with performance maintenance.•Curriculum Learning is also utilized to progressively train the model with confidence-based pseudo-labels.•The proposed method can achieve SOTA performance on three publicly available datasets.