Skip to content

README.md

Latest commit

 

History

History
26 lines (17 loc) · 572 Bytes

README.md

File metadata and controls

26 lines (17 loc) · 572 Bytes
Raw

FID

Real activation statistics

In order to compute FID, we first need to compute the activation statistics over all real images.

Run

python compute_fid_stats.py --dataset cifar10
python compute_fid_stats.py --dataset imagenet64

to compute and save the mean and std of real activation on CIFAR10 and 64x64 ImageNet.

Compute FID with a trained model

python fid_model.py --dataset cifar10 --ckpt={the path to your model}

Compute FID from generated samples

python fid_zip.py {the path to your zip file} --dataset cifar10