search

geyuying / PF-AFN

Official code for "Parser-Free Virtual Try-on via Distilling Appearance Flows", CVPR 2021.
545 stars 138 forks source link

Which dataset should we use to calculate FID score? #73

Open mesllo opened 2 years ago

mesllo commented 2 years ago

I read on some resources that the FID score should be calculated using the dataset of generated outputs and the dataset that was used to train the generator. This feels a bit counter-intuitive since also have the testing dataset (test_pairs) as well in our case. Which datasets should we use to get the FID score? It seems that comparing with the training dataset (which is much larger) I usually get a better FID score for CP-VTON.

TianJiaJiaFirst commented 1 year ago

I meet the same question. Have you sovled this ? @mesllo

xxxxl888 commented 1 year ago

I read on some resources that the FID score should be calculated using the dataset of generated outputs and the dataset that was used to train the generator. This feels a bit counter-intuitive since also have the testing dataset (test_pairs) as well in our case. Which datasets should we use to get the FID score? It seems that comparing with the training dataset (which is much larger) I usually get a better FID score for CP-VTON.

Hello, I would like to ask how you calculate the FID score? Are you using the try-on results generated from the test and the images of the person inputted? My FID score is particularly high.