search

xc-csc101 / RAFSwap

64 stars 2 forks source link

save the soft mask #4

Open Cameltr opened 1 year ago

Cameltr commented 1 year ago

Thank you for your excellent work! I want to ask how to save the soft mask? I have made some tries, but it seems different from your result in paper? The following mask image is the test result of using image in your github repository. mask

johntrimble commented 1 year ago

Yeah, I think something is wrong. Been doing some tests with inference_celeba.py:

1388_1388 1388_14024 1388_28068

That's target, source, result, and mask. It looks like there is more than just mask that isn't right. It almost looks like the source is being inverted and that's it.

johntrimble commented 1 year ago

Err, wanted to follow up here. My input masks for the source and target were bad. Fixing that didn't make the output mask any better, but at least the result looks more like it should:

1388_1388 1388_14024 1388_28068 1388_28132