search

zyainfal / One-Shot-Face-Swapping-on-Megapixels

One Shot Face Swapping on Megapixels.
Other
314 stars 40 forks source link

Dataset version + masks #18

Closed alessiabertugli closed 2 years ago

alessiabertugli commented 2 years ago

Hi,

Where can I find CelebaHQ dataset with the corresponding masks used for the training process (the same version you used)? I saw that the images already swapped were shared, but not the original dataset with which the train was made.

Thanks for your help

zyainfal commented 2 years ago

If your problem is about the inference, you can change the indexes in face_part_ids according to the new version. https://github.com/zyainfal/One-Shot-Face-Swapping-on-Megapixels/blob/edfb9c85120fb9fd32386dca1465a70694b1207c/inference/inference.py#L15 Besides, the provided masks are in 512x512, we simply resize it to 1024x1024 for inference.

And if your problem is about training, you can ignore the dataset version, because we don't need masks for training.

zyainfal commented 2 years ago

Well, one more thing for training, you still need FFHQ and more HD training samples (by running StyleGAN version 1) as well. And if you have real HD samples should be much better.