Faceswap Forum

After 10+ days of training, I have come to the conclusion that a training run without a preview (including writing to an image) is more stable or completely stable

My second suspect is a combination of loss functions and model training state: start/mid/end.

At the end of the training "Fit training" and "No warp training" are recommended, but should they be combined? I see the following options: 1) "No warp" training with my default face set, then "no warp" training with the target video's face set 2) "No war...

And sometimes i catch the following error: [09:08:25] [#01402] Loss A: 0.05251, Loss B: 0.03235zsh: bus error python faceswap.py train -A /Users/user/Desktop/deepfake/training (faceswap) user@user-MacBook-Pro facusereswap % /Users/user/miniforge3/envs/faceswap/lib/python3.9/multiprocessing/resource_...

identified the problem [20:46:55] [#00254] Loss A: 0.05793, Loss B: 0.03490Fatal Python error: Segmentation fault Thread 0x00000005b76a3000 (most recent call first): File "/Users/user/miniforge3/envs/faceswap/lib/python3.9/threading.py", line 312 in wait File "/Users/user/miniforge3/e...

Hello,
tried different presets for Phaze-A: dny256, dny512, sym384. And it ends up with train.py exit code -11 at iteration N (usually at save interval).

BS: 1-32
M1 Max 64gb