cleardusk/3DDFA_V2

How did you train the backbone?

VictorHManrique opened this issue · 1 comments

I understand that you used MobileNet-V1 for the backbone AND you used pre-trained weights for it.

How did you obtained those pre-trained weights? The last layer of the architecture of MobileNet-V1 has 62 classes and you showed the mathematical procedure, but how about the programming procedure?

I'm currently working on changing the backbone of the model with a better lightweight model as part of my master's thesis, but I don't understand the process to obtained those pre-trained weights.

Hi @VictorHManrique , any updates on this ?