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

Benchmark Model Rank Results
domain-adaptation-on-gta5-to-cityscapesHRDA+PiPa#5mIoU: 75.6
domain-adaptation-on-synthia-to-cityscapesHRDA+PiPa#4mIoU: 68.2
image-to-image-translation-on-gtav-toHRDA + PiPa#2mIoU: 75.6
image-to-image-translation-on-gtav-toDAFormer + PiPa#4mIoU: 71.7
image-to-image-translation-on-synthia-toHRDA + PiPa#1mIoU (13 classes): 74.8
semantic-segmentation-on-gtav-to-cityscapes-1HRDA + PiPa#2mIoU: 75.6
semantic-segmentation-on-synthia-toHRDA + PiPa#1Mean IoU: 68.2
synthetic-to-real-translation-on-gtav-toHRDA+PiPa#3mIoU: 75.6
synthetic-to-real-translation-on-gtav-toDAFormer+PiPa#6mIoU: 71.7
synthetic-to-real-translation-on-synthia-to-1HRDA+PiPa#2MIoU (16 classes): 68.2MIoU (13 classes): 74.8
unsupervised-domain-adaptation-on-gtav-toHRDA + PiPa#2mIoU: 75.6
unsupervised-domain-adaptation-on-gtav-toDAFormer + PiPa#5mIoU: 71.7
unsupervised-domain-adaptation-on-synthia-toHRDA + PiPa#3mIoU (13 classes): 74.8