Misaligned segmentation masks
kazuto1011 opened this issue · comments
Downsampling by PyTorch nearest interpolation in main.py results in misaligned masks. That's going to be replaced with Pillow nearest interpolation. Codes and pre-trained models will be updated sometime soon.
FYI: 1/16 downsampling with various frameworks.
So far this repository has used the right 3 methods.
