Could you show some detail about the training loss implementation about pngan HOT 1 CLOSED

Hi, thanks for your attention,
The loss functions are detailed in Eq. (2), (6), (12), and (13) in our paper.

We use a pre-trained denoiser RID-Net as a regularizer to conduct image domain alignment in Eq. (2)
The loss functions of Generator and Discriminator are given in Eq. (6)
Eq. (12) describes perceptual loss
Eq. (13) is the overall training objective which is a weighted sum of Eq. (2), (6), (12)

Admittedly, the training process is very sophisticated. We are going to do more research and extend our paper to a journal version. I appreciate your kind re-implementation of our PNGAN. I will mention your work in this repo. Thank you very much.

from pngan.