search

facebookresearch / pytorch_GAN_zoo

A mix of GAN implementations including progressive growing
BSD 3-Clause "New" or "Revised" License
1.62k stars 271 forks source link

PGGAN model on CIFAR10 obtains poor result on Inception score and FID #104

Open GongXinyuu opened 4 years ago

GongXinyuu commented 4 years ago

Hi, thanks for your great work. I use the official release tf code of Inception score and FID code to evaluate the PGGAN model on CIFAR10. It turns out the results are not good.

Here's my command to train:

python datasets.py cifar10 ~/dataset/cifar-10-batches-py -o data/cifar10
python train.py PGAN -c config_cifar10.json -n cifar10_eval --no_vis

The result I got is:

FID score: 42.96, Inception score: 6.34299373626709 ± 0.06940167397260666

XenonLamb commented 4 years ago

May I ask how did you get the FID score for the pytorch PGGAN model?

GongXinyuu commented 4 years ago

May I ask how did you get the FID score for the pytorch PGGAN model?

I'm using this repo.