DeepFaceLab icon indicating copy to clipboard operation
DeepFaceLab copied to clipboard

RTM WF faceset

Open CHRISTOPHERMOSLEY1975 opened this issue 3 years ago • 1 comments

My data_src face size is 768x768 and RTM WF Faceset.zip (https://disk.yandex.ru/d/7i5XTKIKVg5UUg/DeepFaceLab/Facesets) are different size (768x768 and 224x224) for create my own RTM model must change RTM WF Faceset.zip faces size to 768x768 or use without changing image size?

CHRISTOPHERMOSLEY1975 avatar May 10 '22 17:05 CHRISTOPHERMOSLEY1975

I haven't trained such particular model, but I think you don't have to change the resolution, especially if your target will be 224x224 (i.e. not higher than the lower resolution images in your set.) There are preprocessings of the images on each iteration anyway, in order to align the landmarks, to apply warps etc.

Actually I think if the target resolution would be 224x, you could rather resize (reextract) your faceset to lower resolution for that training, because that may increase the performance in this domain (reading and preprocessing the images), however I don't know how big the difference would be. You could test it with training partial facesets only with 224x224 and only with 768x - e.g.: place them in different directories, edit the batch file.

Twenkid avatar May 11 '22 19:05 Twenkid

Did you ever find the answer? If so, would you mind sharing it and closing this issue?

joolstorrentecalo avatar Jun 08 '23 23:06 joolstorrentecalo