xunhuang1995/AdaIN-style

Training code

Opened this issue · 5 comments

Can you share the training code?

I am reproducing this work in Tensorflow

Has anyone succeeded on re-implementing the training procedure? I am going to give it a shot over the weekend to better understand the algorithm. Also, hopefully I understand this correctly but the only learnable network in the entire pipeline (Figure 2) is the VGG Decoder? The AdaIN is a deterministic function with no learnable parameters as stated in section 5 of the ArXiv paper? Is this correct @xunhuang1995 ? Hmm if this is the case, maybe I can start playing around with loading the DeCoder (as pre-trained network), and continue doing iterations over new style images.

The training code has been updated.