Open RigoYao opened 1 year ago
The implementation is incomplete. In their later papers, the authors do not attempt to implement the mapper in the S space either. If you think about it, it's not necessary. You don't need a mapper in the S space since it has global directions. In the W+ space, the step or the learned residual is different for each input, but that's not the case in the Style Space S. I may be wrong because I'm still wrapping my head around the Style Space stuff, but this is my best guess based on the motivation for this new space.
The error message is as below:
if latent.shape[2] == 512: IndexError: tuple index out of range