# Evaluating performance of Generative Adverserial Network?

What is the best way to evaluate performance of Generative Adverserial Network (GAN)? Perhaps measuring the distance between two distributions or maybe something else?