CapsuleGAN: Generative Adversarial Capsule Network

TL;DR

CapsuleGAN introduces capsule networks as discriminators in GANs, leading to improved image data modeling on MNIST and CIFAR-10, with better performance in generative and semi-supervised tasks.

Contribution

This work is the first to integrate capsule networks into GAN discriminators, providing design guidelines and demonstrating superior performance over CNN-based GANs.

Findings

CapsuleGAN outperforms convolutional-GAN on MNIST and CIFAR-10.

CapsuleGAN achieves better results on the generative adversarial metric.

CapsuleGAN improves semi-supervised image classification.

Abstract

We present Generative Adversarial Capsule Network (CapsuleGAN), a framework that uses capsule networks (CapsNets) instead of the standard convolutional neural networks (CNNs) as discriminators within the generative adversarial network (GAN) setting, while modeling image data. We provide guidelines for designing CapsNet discriminators and the updated GAN objective function, which incorporates the CapsNet margin loss, for training CapsuleGAN models. We show that CapsuleGAN outperforms convolutional-GAN at modeling image data distribution on MNIST and CIFAR-10 datasets, evaluated on the generative adversarial metric and at semi-supervised image classification.