Generative Adversarial Network Architectures For Image Synthesis Using Capsule Networks

TL;DR

This paper introduces GAN architectures utilizing Capsule Networks as critics, leveraging their spatial relationship encoding to improve image synthesis efficiency and quality with fewer training samples and epochs.

Contribution

It presents novel GAN architectures with Capsule Network critics that enhance training efficiency and image quality over traditional CNN-based GANs.

Findings

Capsule-based GANs synthesize images faster and with fewer samples.

Capsule critics improve spatial relationship modeling in generated images.

Lower coverage and diversity issues are analyzed in CNN-based GANs.

Abstract

In this paper, we propose Generative Adversarial Network (GAN) architectures that use Capsule Networks for image-synthesis. Based on the principal of positional-equivariance of features, Capsule Network's ability to encode spatial relationships between the features of the image helps it become a more powerful critic in comparison to Convolutional Neural Networks (CNNs) used in current architectures for image synthesis. Our proposed GAN architectures learn the data manifold much faster and therefore, synthesize visually accurate images in significantly lesser number of training samples and training epochs in comparison to GANs and its variants that use CNNs. Apart from analyzing the quantitative results corresponding the images generated by different architectures, we also explore the reasons for the lower coverage and diversity explored by the GAN architectures that use CNN critics.