gan-metrics

Lots of evaluation metrics of Generative Adversarial Networks in pytorch

Work In Progress...

Requirements

• Python 3.x
• torch 1.x
• torchvision 0.4.x
• numpy
• scipy
• cython
• pot
• pyyaml

Usage

1. Clone the repository and Move the directory.

$ git clone https://github.com/kozistr/gan-metrics
$ cd ./gan-metrics

2. Install the requirements.txt

$ pip3 install -r requirements.txt

3. Configure the config.yml

open the file and edit

4. Run

$ python3 -m metrics

Metrics

Metric	Paper	Code
Inception Score (IS)	arXiv	code
Frechet Inception Distance (FID)		code
Kernel Inception Distance (KID)
SSIM & PSNR
Precision & Recall
Earth Mover's Distance (EMD, a.k.a wasserstein distance)	wiki	code
Perceptual Path Length (PPL)
Learned Perceptual Image Patch Similarity (LPIPS)	arXiv
Reliable Fidelity & Diversity (PRDC)	arXiv
Amazon's Mechanical Turk (AMT)	Amazon

DataSets

You can edit the transform function with your flavor!

• MNIST / FashionMNIST
• CIFAR10 / CIFAR100
• LSUN
• CelebA
• ImageNet
• Custom

References

• GAN-Metrics : https://github.com/xuqiantong/GAN-Metrics
• GAN_Metrics-Tensorflow : https://github.com/taki0112/GAN_Metrics-Tensorflow

Author

Hyeongchan Kim / @kozistr