Hi！One question again. :) The results seem not so good when I apply it to my own data. I think that the images sent to the discriminator should multiply with their associated masks as illustrated in the original paper. Look forward to youre reply!

[DLwbm123]

Hi！One question again. :) The results seem not so good when I apply it to my own data. I think that the images sent to the discriminator should multiply with their associated masks as illustrated in the original paper. Look forward to youre reply!

[yhlleo]

Yes,I tested it, but the results were similar.

[yhlleo]

I have updated the code to support using generated image multiplied by the mask as input for the discriminator. You can set it by changing the parameter --use_mask_for_D.

See the configuration in the scripts/train_uaggan.sh.

[DLwbm123]

Great thanks!