论文理解：Self-Ensembing with GAN-based Data Augmentation for Domain Adaption in Sematic Segmentation