I've been stalking the forum since a week or so, small introduction here.
After I figured out that my laptop was basically doing nothing much all day while I'm working I thought I'd make it do something else that the protein folding that it has been doing for the last 4 years.
Wanted to look into face swapping for a while, and discovering a GUI for it made it so much easier I thought I'd give it a go.
First video I had in front of me when I installed face swap was "Useless" by Depeche Mode and I figured "why not" and started collecting images of the singer and did some quick video's of myself on my phone just to see how the whole process works.
After 80K iterations I did a side by side
This immediately got a copyright strike so I embedded different audio (ripped out the voice with AI) and at least it is now viewable everywhere apart from the US and Canada. I did an upload on Vimeo too
I also did a small compilation of earlier swaps after approx l2, 24 and 40K iterations to show how much the model improves over time (I hope the privacy settings on this don't mess it up)
This model obviously lacks the resolution I need for the closer up sections of the clip.
My poor old laptop only has a GTX970 mobile, and I'm getting about 7 iterations per minute it seems.
I'll kick off a higher res learning model tonight, I started one a few days ago but I'm only getting about 4 iterations per minute, making a 100K iteration a bit of a time investment
Does anyone know some software to see what my gfx card is actually doing? Windows pretends as if it is doing not much at all and my VRAM is only using 0.6GB out of 6 available
I am using batch size 32 now instead of 16, is there any advantage in using larger batch sizes? I don't really fully understand the use of batch size.
Thanks, and great to see others people work, very cool to see people share their knowledge.