PiPa: Pixel- and Patch-wise Self-supervised Learning for Domain Adaptative Semantic Segmentation

Benchmark Model Rank Results
image-to-image-translation-on-synthia-to-cityscapesHRDA + PiPa#1mIoU (13 classes): 74.8
semantic-segmentation-on-synthia-to-cityscapesHRDA + PiPa#2Mean IoU: 68.2
image-to-image-translation-on-gtav-to-cityscapes-labelsHRDA + PiPa#3mIoU: 75.6
image-to-image-translation-on-gtav-to-cityscapes-labelsDAFormer + PiPa#6mIoU: 71.7
semantic-segmentation-on-gtav-to-cityscapes-labelsHRDA + PiPa#3mIoU: 75.6
synthetic-to-real-translation-on-synthia-to-cityscapesHRDA+PiPa#3MIoU (16 classes): 68.2MIoU (13 classes): 74.8
unsupervised-domain-adaptation-on-gtav-to-cityscapes-labelsHRDA + PiPa#3mIoU: 75.6
unsupervised-domain-adaptation-on-gtav-to-cityscapes-labelsDAFormer + PiPa#7mIoU: 71.7
unsupervised-domain-adaptation-on-synthia-to-cityscapesHRDA + PiPa#3mIoU (13 classes): 74.8
domain-adaptation-on-synthia-to-cityscapesHRDA+PiPa#5mIoU: 68.2
synthetic-to-real-translation-on-gtav-to-cityscapes-labelsHRDA+PiPa#5mIoU: 75.6
synthetic-to-real-translation-on-gtav-to-cityscapes-labelsDAFormer+PiPa#8mIoU: 71.7
domain-adaptation-on-gta5-to-cityscapesHRDA+PiPa#6mIoU: 75.6