search

Kibeom-Hong / Domain-Aware-Style-Transfer

Official Implementation of Domain-Aware Universal Style Transfer
Other
103 stars 13 forks source link

Training process. #7

Closed sunwoo76 closed 2 years ago

sunwoo76 commented 2 years ago

Hi.

I have a question.

Did you use the total loss or only reconstruction loss when you train your decoder for reconstruction?

Reconstruction and stylization are both trained?

thanks.

Kibeom-Hong commented 2 years ago

Thanks for interesting my research.

I used the total loss for only reconstruction tasks.

Since this work is for arbitrary style transfer, our networks is trained only for reconstruction. (not stylization)

The only stylization is conducted by the transformer.