Brilliant work!Thanks for sharing!
I tried to train VQGAN v1 and v2 using the FFHQ256 dataset in the original setup without changing the batch size and other Settings in the provided train_vqgan_v1_B16_800K.yml file.
May I ask why PSNR has been only 20 after training more than 140k iters, and what is normal value of the validation's PSNR when training VQGAN?
I would be grateful if you can kindly give a hint! :)