Open molo32 opened 4 years ago
I met the same problem, have you fixed it?
no :( , if you find a solution let me know, author please help us.
The problem is args.label_nc is not consistent with the label number of the mask. The folder "SEAN/imgs/colormaps" is not the input mask. You should download CelebA-HQ dataset, and note that the input mask is "CelebA-HQ/test/labels" instead of "CelebA-HQ/test/vis". Good luck!
I'm having the same problem, but I am using grayscale images with proper integers for each class. Anyone found a workaround?
Found the culprit, I had some rogue ids in my segmentation maps due to an error when generating them.
I get an error when I run test.py
!python test.py --checkpoints_dir "" --name CelebA-HQ_pretrained --load_size 256 --crop_size 256 --dataset_mode custom --label_dir /content/SEAN/imgs/colormaps --image_dir /content/SEAN/imgs/colormaps --label_nc 19 --no_instance --gpu_ids 0