A Capsule Conditional Generative Adversarial Network

Generative Adversarial Networks (GAN) is currently one of the most popular research topics in machine learning. It mainly uses two competing networks, a generator network and a discriminator network to learn from the distribution of real samples. The complexity of the real sample distribution leads to many problems in the stability of the generative adversarial network in the training process. The Deep Convolutional Generative Adversarial Networks (DCGAN) were introduced to solve some problem. The performance quality of the results has been significantly improved by DCGAN, but it still sometimes fails to produce realistic data. Due to the deficiencies above DCGAN, this paper involves the capsule network structure into the conditional generation adversarial network, and proposes the CapsuleCGAN (Capsule Conditional Generative Adverserial Networks) architecture. Through the experiment on the MNIST dataset, comparing CGAN and DCGAN, the results show that the proposed CapsuleCGAN architecture can be successfully trained. From the experimental results, not only the high-quality images can be generated, but also in generating maps as diversity. Finally, we put some forward suggestions and ideas for improvement in the future work.