Geometry Score: A Method For Comparing Generative Adversarial Networks.

One of the biggest challenges in the research of generative adversarialnetworks (GANs) is assessing the quality of generated samples and detectingvarious levels of mode collapse. In this work, we construct a novel measure ofperformance of a GAN by comparing geometrical properties of the underlying datamanifold and the generated one, which provides both qualitative andquantitative means for evaluation. Our algorithm can be applied to datasets ofan arbitrary nature and is not limited to visual data. We test the obtainedmetric on various real-life models and datasets and demonstrate that our methodprovides new insights into properties of GANs.