Running Deepfacelab 2.0, SAEHD, NVIDIA RTZ 2060 (6gb vram).
90% of my render looks great. There are a few quick shots (semi-blurry, but still visible/recognizable) that are training very slowly I had to re-mask and add some frames after maybe 400k iterations, and since then those frames are lagging behind the rest (this is my first DeepFake, I'll know better to xseg train all frames first next time).
My question - is it possible to temporarily move the dst frames that have modeled well to another folder, and only train the ones that aren't generating as fast? Or will this destroy the model? Should I just continue with the current settings and wait it out?
dropped learning rate enabled (cpu)
Random warp off
Batch 6
90% of my render looks great. There are a few quick shots (semi-blurry, but still visible/recognizable) that are training very slowly I had to re-mask and add some frames after maybe 400k iterations, and since then those frames are lagging behind the rest (this is my first DeepFake, I'll know better to xseg train all frames first next time).
My question - is it possible to temporarily move the dst frames that have modeled well to another folder, and only train the ones that aren't generating as fast? Or will this destroy the model? Should I just continue with the current settings and wait it out?
dropped learning rate enabled (cpu)
Random warp off
Batch 6