Ensembles of Generative Adversarial Networks for Disconnected Data

TL;DR

This paper demonstrates that disconnected data distributions in computer vision cannot be perfectly modeled by a single continuous GAN, proposing ensemble methods as a more effective alternative with practical advantages.

Contribution

It introduces a regularized optimization framework linking single GANs, ensembles, and mixture models, showing ensembles outperform single models on disconnected data.

Findings

Ensembles outperform single continuous GANs on disconnected datasets.

The proposed regularization links various generative models efficiently.

Optimal performance is achievable through hyperparameter tuning.

Abstract

Most current computer vision datasets are composed of disconnected sets, such as images from different classes. We prove that distributions of this type of data cannot be represented with a continuous generative network without error. They can be represented in two ways: With an ensemble of networks or with a single network with truncated latent space. We show that ensembles are more desirable than truncated distributions in practice. We construct a regularized optimization problem that establishes the relationship between a single continuous GAN, an ensemble of GANs, conditional GANs, and Gaussian Mixture GANs. This regularization can be computed efficiently, and we show empirically that our framework has a performance sweet spot which can be found with hyperparameter tuning. This ensemble framework allows better performance than a single continuous GAN or cGAN while maintaining fewer total parameters.