Closed dialuser closed 1 year ago
the d_loss was always zero when I ran the code. I noticed in forward() of LPIPSWithDiscriminator, codebook_loss is never used. Is that related?
It is not related. We don't use VQ loss (see the paper) and to use GAN loss you need to load the checkpoint (that is trained without GAN loss) and execute with different config file that I provided.
the d_loss was always zero when I ran the code. I noticed in forward() of LPIPSWithDiscriminator, codebook_loss is never used. Is that related?