If I I'm reusing a B face, and I'm uninterested in swapping the A onto the B face, am I wasting my time trying to get the A face to the absolute loss minimum?
Further Explanation:
I use a lot of the same actors faces for my B face, and now have them down to a reasonable loss value. It occured to me as I'm working with a new A face model, that it might not be necessary to run training for a million iterations as I'm not interested in swapping my A face onto my B face. I just need my machine to understand the basic structure of the A face and how my B face can fit onto it.
Last edited by MaxHunter on Wed Apr 12, 2023 6:01 pm, edited 1 time in total.
When I’ve dome that before, it’s more of a case where the new face A is “good enough” than looking for perfection.
I’m sure it’ll probably be better if both sides are fully trained to perfection, but like all things it’s a balancing act between wants and resources.
Of course the proper way is to have the model trained from the ground up on two faces but sometimes you just want to make something quick for fun instead of caring about identity loss/bleed.