Giters

mseitzer / pytorch-fid

Compute FID scores with PyTorch.

Repository from Github https://github.commseitzer/pytorch-fidRepository from Github https://github.commseitzer/pytorch-fid

Dataset sizes

JesseWiers opened this issue · comments

commented

As I understand it both the synthetic dataset and the real one need to consist of at least 2048 images to use FID. However, does it matter if the real data set has a lot more images compared to the sampled (synthetic) dataset? Since sampling takes a while I only want to sample 2048 images whereas I have a lot of real images