YuanZYF opened this issue 5 years ago · 1 comments
If input_img is a natural scene image, how can I train?
U-net with image-net pretrained backbones should work well for natural scene images.