Train and Validation Split

[stevenwalton]

What was the train and validation split used? I'm using the checkpoint provided and testing with a validation set of the top 10k, similar to Co-Mod-GAN's split (section 5.1). Using this split I am getting a FID of 4.26.

[ForeverFancy]

Hi, as written in the paper, for FFHQ and LSUN Church datasets, we randomly sample 50,000 images from the original datasets as validation sets and calculate FID between the validation sets and 50,000 generated images. For CelebA-HQ, we calculated the FID between 30,000 generated images and all the training samples.

[stevenwalton]

Thanks. Do you have any uncertainty value for these numbers? Or were all your experiments with the same random seed?

[ForeverFancy]

Sorry, we do not uncertainty value of the reported numbers. The random seeds are all set to the default values.

[stevenwalton]

Thanks