improved-diffusion
improved-diffusion copied to clipboard
How I can get the same FID on CIFAR-10 when sample steps = 100 as the paper?
Hi, thanks for your great work. I test with your checkpoint and the FID with 100 sample steps is much higher than the results in FIG.8 in the main paper (28 vs 10). We have generated 50000 images for this evaluation. Is there something we need to be cautious about calculating FID on this test?