Hey all,
First I needed a new PC - so being on AMD RYZEN 5 1600 32GB MEM, Gigabyte D450 as MB and sadly for now a Geforce 1050TI (only 4MB VRAM) - it started.
1) Extracting is the Alpha-Omega to make the AI able to train a model. Manual alignment is killing me as it puts the face on everywhere else but the face. (Landmarks warping). Noticed after the latest update.
2) The sorting - if you follow the flow verbatum - it should rename your files back to the original (after deletion and the Remove Frames) - except - it now doesn't anymore. Also after latest update.
Reading through the boards has enlightened a few issues that I noticed I myself had and Faceswap runs smoothly. I took a bad training set just to see how the model would train - and you should - because it gives you an insight to what it actually computes and what is just considered bad data.
I have yet to train a model fully. For AMD users I suggest you always check if you AMD-V or SMV (newer chipsets) is on and that Windows has downloaded virtual environment. I'm on Win10 Home.
As the manual says pay no attention to the 0.XXXXX when model is training. When drops from, example could be 0.18000 to 0.4900 it simply means the AI is still in school and learning. But you should have at least 5000 x 2 images (A->B) - well at least that is my own experience - thus far.
I may have missed some vital info - if so feel free to point it out.
/sergor