Faceswap Forum

As I understand, I do an extraction simply to load up the alignments file with all the frames I skipped during training. Then I use the manual tool to fix alignments and do masks just like with training, but there are THOUSANDS of frames.

If that's what has to be done, so be it, but I really don't want to go through a thousand frames if there's something I don't know about how the tool works. Is it really necessary to fix each frame that's out of alignment? Or does the training take care of some of those somehow?

You only need to adjust faces that aren't accurate or are missing. If they're fine then you move on to the next.

Our face detector has the most problems with things like obstructions covering most of the face and the face turning away from the camera. So look especially for those frames to make sure they're correct.

Personally, my preferred way of finding bad frames is a test convert. I run a convert process and remember where in the video there were problems, then I find those in the manual tool to fix them. Others do look at every single frame, if they want a perfect result with absolutely no flaws.

Fair enough. I find that I can get through it quicker by toggling the face alignments view in the manual tool where I can see the overlay on all the frames. It makes it easier to pick out the bad patches in a hurry. At least I know to focus on the bad areas and hopefully that will help

Great tool all around and props to everyone that's working on it!

I do hope someday, the tool advances to modeling a 3d head geometry with full shape and structure though. If it does, it could more easily handle the times when someone is in full profile or greater or looking down and up. But what we have is so awesome for a home tool (and free to boot!). Thank you!