taki0112 / GAN_Metrics-Tensorflow

Simple Tensorflow implementation of metrics for GAN evaluation (Inception score, Frechet-Inception distance, Kernel-Inception distance)

GAN_Metrics-Tensorflow

Simple Tensorflow implementation of metrics for GAN evaluation

• Inception score
• Frechet-Inception distance
• Kernel-Inception distance

Summary

Name	Description	Performance score
Inception score	KL-Divergence between conditional and marginal label distributions over generated data.	Higher is better.
Frechet-Inception distance	Wasserstein-2 distance between multi-variate Gaussians fitted to data embedded into a feature space.	Lower is better.
Kernel-Inception distance	Measures the dissimilarity between two probability distributions Pr and Pg using samples drawn independently from each distribution.	Lower is better.

Usage

├── real_source 
    ├── aaa.png
    ├── bbb.jpg
├── real_target 
    ├── ccc.png
    ├── ddd.jpg
├── fake 
    ├── ccc_fake.png
    ├── ddd_fake.jpg
├── main.py
├── inception_score.py
└── frechet_kernel_Inception_distance.py

> python main.py

Reference

• Pros and Cons of GAN Evaluation Measures
• Inception score
• Frechet-Inception score

Author

Junho Kim