Hi
Just wondering if it's possible to change something in the settings to make re-extracting from alignments file produce face files that are less zoomed in.
I've been using dfl to detect and generate faces and clean up alignments from videos, but after I generate an alignments file using faceswap and re-extract the faces, I noticed that the resulting faces are too zoomed in (compared to the original face images) and thus, when I use the faces and alignments for training/conversion, the face coverage I have is too low. I already have a few hundred iters on training so I don't really want to redo the training with a higher face coverage if possible. Just using the original face images wouldn't work either due to a size mismatch (512x512 vs 256x256).
Thanks.