- Abstract: A fundamental, and still largely unanswered, question in the context of Generative Adversarial Networks (GANs) is whether GANs are actually able to capture the key characteristics of the datasets they are trained on. The current approaches to examining this issue require significant human supervision, such as visual inspection of sampled images, and often offer only fairly limited scalability. In this paper, we propose new techniques that employ classification-based perspective to evaluate synthetic GAN distributions and their capability to accurately reflect the essential properties of the training data. These techniques require only minimal human supervision and can easily be scaled and adapted to evaluate a variety of state-of-the-art GANs on large, popular datasets. They also indicate that GANs have significant problems in reproducing the more distributional properties of the training dataset. In particular, the diversity of such synthetic data is orders of magnitude smaller than that of the original data.
- TL;DR: We propose new methods for evaluating and quantifying the quality of synthetic GAN distributions from the perspective of classification tasks
- Keywords: Generative adversarial networks, classification, benchmark, mode collapse, diversity