"High quality sources" vs. "training only uses 64x64 image"

Want to understand the training process better? Got tips for which model to use and when? This is the place for you

Forum rules

Read the FAQs and search the forum before posting a new topic.

This forum is for discussing tips and understanding the process involved with Training a Faceswap model.

If you have found a bug are having issues with the Training process not working, then you should post in the Training Support forum.

Please mark any answers that fixed your problems so others can find the solutions.

Post Reply
User avatar
Posts: 19
Joined: Mon Mar 22, 2021 4:24 pm
Been thanked: 1 time

"High quality sources" vs. "training only uses 64x64 image"

Post by Replicon »

Random thought/question that just popped into my mind:

The recommendation is always that using high quality images/video sources for training is important...

... but since models seem to use small images (many only doing 64x64), does it mean the source quality isn't all that important, as long as the face is as detailed as it can be in that size?

To put it another way: If I use a 1080p video of an interview, and the faces take up say 512x512 pixels on the actual video frames, but the training model only does 64x64, will there be much of a difference between that and using a 100x100 pixel source video with a head close-up where the face takes up say 80x80 on the video frame?

This all sounds kind of academic, but I guess it might help answer the question, "if I have a really high quality video, should I still keep smaller faces in the training data, as long as they're bigger than 64x64 pixels in the video?"

Post Reply