GAN Metrics

Inception Score

using Inception V3 get 1000 output class vector
based on assumption

  1. a real photo have higher probably belong to 1 class
  2. a good generative model should output different class

Fréchet Inception Distance (FID)

use feature instead of final output

Precision, Recall and F1 Score

Learned Perceptual Image Patch Similarity (LPIPS)

The Unreasonable Effectiveness of Deep Features as a Perceptual Metric(CVPR 2018) to measure the average feature distances between generated samples. Higher LPIPS scores indicate better diversity among the generated images.