Closed zero-nnkn closed 1 year ago
Hi, that is interesting. Unfortunately, I do not remember the exact details of the testing regime, but I would assume that for our work the code found in the folder utils was used. I will have to let you figure the rest out yourself.
Hi, I try to evaluate FID on the cleaned VITON dataset (416 images) with the generated test results that this repo provides. I use pytorch-fid to calculate FID. However, the result is worse than reported in the paper (37.05 and 19.54). Maybe my method is wrong, could you tell me how to reproduce the fid results?