Official Implementation of PiPa: Pixel- and Patch-wise Self-supervised Learning for Domain Adaptative Semantic Segmentation
Official Implementation of PiPa: Pixel- and Patch-wise Self-supervised Learning for Domain Adaptative Semantic Segmentation
The reproduced results are relatively low, with the mIoU for Human, Rider, and Sidewalk being only 37.67, 18.03, and 37.01 respectively. What could be the reason for this? I would be very grateful to receive your response.
Loss problem
how to run this code on CPU only
Google drive link to download the [MiT weights] is not working
out of memory
patch loss + pixel loss
Have you used imagenet feature distance?