Open imlixinyang opened 4 years ago
I find that FUNIT use a different loss to make the translation image have the same class as the guide image, one is the $L_adv$ and one is the $L_F$, but why don't you use reconstruction of style (or class code) to replace the $L_F$ like MUNIT?
I find that FUNIT use a different loss to make the translation image have the same class as the guide image, one is the $L_adv$ and one is the $L_F$, but why don't you use reconstruction of style (or class code) to replace the $L_F$ like MUNIT?