vae enc_dec configuration questions
treasuraid opened this issue · 0 comments
treasuraid commented
I had a question on model configurations with paper.
In paper, f=8 thus the for super-res model's latent map has 6464 resolution map with image size 512512.
Thus, I was trying with the vae with number of layer is 3 thus to make sure they have 64*64 resolution. However I think it may different with paper's implementation detail that made vae consisit of 4 layers and 256 dim with finetuning or 2 layers and 128 dim without finetuning decoder.
How can I set configuration of Vae in code to match with paper's?