DAFormer: Improving Network Architectures and Training Strategies for Domain-Adaptive Semantic Segmentation

Benchmark Model Rank Results
semantic-segmentation-on-densepassDAFormer#5mIoU: 54.67%
image-to-image-translation-on-synthia-to-cityscapesDAFormer#7mIoU (13 classes): 67.4
semantic-segmentation-on-dark-zurichDAFormer#8mIoU: 53.8
semantic-segmentation-on-gtav-to-cityscapes-labelsDAFormer#10mIoU: 68.3
semantic-segmentation-on-synthia-to-cityscapesDAFormer#10Mean IoU: 60.9
unsupervised-domain-adaptation-on-synthia-to-cityscapesDAFormer#10mIoU (13 classes): 67.4mIoU: 60.9
domain-adaptation-on-cityscapes-to-acdcDAFormer#11mIoU: 55.4
image-to-image-translation-on-gtav-to-cityscapes-labelsDAFormer#11mIoU: 68.3
domain-adaptation-on-gta5-to-cityscapesDAFormer#13mIoU: 68.3
synthetic-to-real-translation-on-synthia-to-cityscapesDAFormer#13MIoU (16 classes): 60.9MIoU (13 classes): 67.4
unsupervised-domain-adaptation-on-gtav-to-cityscapes-labelsDAFormer#14mIoU: 68.3
domain-adaptation-on-synthia-to-cityscapesDAFormer#15mIoU: 60.9
synthetic-to-real-translation-on-gtav-to-cityscapes-labelsDAFormer#15mIoU: 68.3