MMGAN: Manifold Matching Generative Adversarial Network for Generating Images

Generative adversarial networks (GANs) are considered as a totally different type of generative models. However, it is well known that GANs are very hard to train. There have been proposed many different techniques in order to stabilize their training procedures. In this paper, we propose a novel training method called manifold matching and a new GAN model called manifold matching GAN (MMGAN). In MMGAN, vector representations extracted from the last layer of the discriminator are used to train the generator. It finds two manifolds representing vector representations of real and fake images. If these two manifolds are matched, it means that real and fake images are identical in the perspective of the discriminator because the manifolds are constructed from the discriminator's last layer. In general, it is much easier to train the discriminator and it becomes more accurate as epoch goes by. This implies that the manifold matching also becomes very accurate as the discriminator is trained. We also use the kernel trick to find better manifolds. We conduct in-depth experiments with three image datasets and several state-of-the-art GAN models. Our experiments demonstrate the efficacy of the proposed MMGAN model.