Cross-Domain Grouping and Alignment for Domain Adaptive Semantic Segmentation

Existing techniques to adapt semantic segmentation networks across source and target domains within deep convolutional neural (CNNs) deal with all the samples from two in a global or category-aware manner. They do not consider an inter-class variation domain itself estimated category, providing limitation encode having multi-modal data distribution. To overcome this limitation, we introduce learnable clustering module, novel adaptation framework, called cross-domain grouping alignment. cluster aim maximize alignment without forgetting precise ability on domain, present loss functions, particular, for encouraging consistency orthogonality among clusters. We also so as solve class imbalance problem, which is other of previous methods. Our experiments show that our method consistently boosts performance segmentation, outperforming state-of-the-arts various settings.