Intriguing Properties of Modern GANs

arXiv:2402.14098v1 Announce Type: new
Abstract: Modern GANs achieve remarkable performance in terms of generating realistic and diverse samples. This has led many to believe that ``GANs capture the training data manifold''. In this work we show that this interpretation is wrong. We empirically show that the manifold learned by modern GANs does not fit the training distribution: specifically the manifold does not pass through the training examples and passes closer to out-of-distribution images than to in-distribution images. We also investigate …

abstract arxiv cs.cv cs.lg data diverse gans interpretation manifold modern performance samples show terms training training data type work