Metrics of GANs

• Fréchet Inception Distance (FID)
• Inception Score (IS)
• Number of statistically-Different Bins (NDB)
• Jensen-Shannon Divergence (JSD)
• Learned Perceptual Image Patch Similarity (LPIPS)

Metric	Usage	Notes
FID	Image Quality	the lower, the better
IS	Image Quality	the higher, the better
NDB	Diversity	the lower, the better
JSD	Diversity	the lower, the better
LPIPS	Diversity	the higher, the better

Configuration

See the environment.yaml. We provide an user-friendly configuring method via Conda system, and you can create a new Conda environment using the command:

conda env create -f environment.yaml

Usage

• IS:

python eval.py --metric is --pred_list <path/to/pred_list> --gpu_id 0 --resize 299

• FID:

python eval.py --metric fid --pred_list <path/to/pred_list> --gt_list <path/to/gt_list> --gpu_id 0 --resize 299

• NBD & JSD:

python eval.py --metric ndb --pred_list <path/to/pred_list> --gt_list <path/to/gt_list> --gpu_id 0 --resize 128

• LPIPS:

python lpips.py --path <path/to/image_folder> --test_list <path/to/test_list>

Example of test list in LPIPS:

a1.png	a2.png
a1.png	a3.png
...