An empirical study on evaluation metrics of generative adversarial networks
This work addresses the problem of reliable evaluation metrics for GANs, which is crucial for researchers and practitioners in machine learning, though it is incremental as it builds on existing metrics.
The paper tackles the challenge of evaluating generative adversarial networks (GANs) by revisiting sample-based metrics and assessing their ability to distinguish real from generated samples, detect issues like mode dropping, and identify overfitting. Through experiments, it finds that kernel Maximum Mean Discrepancy (MMD) and the 1-Nearest-Neighbor test perform well under suitable conditions, revealing insights into GAN behaviors such as memorization and distribution learning.
Evaluating generative adversarial networks (GANs) is inherently challenging. In this paper, we revisit several representative sample-based evaluation metrics for GANs, and address the problem of how to evaluate the evaluation metrics. We start with a few necessary conditions for metrics to produce meaningful scores, such as distinguishing real from generated samples, identifying mode dropping and mode collapsing, and detecting overfitting. With a series of carefully designed experiments, we comprehensively investigate existing sample-based metrics and identify their strengths and limitations in practical settings. Based on these results, we observe that kernel Maximum Mean Discrepancy (MMD) and the 1-Nearest-Neighbor (1-NN) two-sample test seem to satisfy most of the desirable properties, provided that the distances between samples are computed in a suitable feature space. Our experiments also unveil interesting properties about the behavior of several popular GAN models, such as whether they are memorizing training samples, and how far they are from learning the target distribution.