BEGAN: Boundary Equilibrium Generative Adversarial Networks

TL;DR

BEGAN introduces a novel equilibrium enforcing method with Wasserstein-based loss for auto-encoder GANs, achieving high-quality, stable, and diverse image generation at high resolutions with simple models.

Contribution

It presents a new equilibrium enforcing technique and convergence measure, improving training stability and image quality in auto-encoder GANs.

Findings

Achieves high visual quality at high resolutions

Provides a stable and fast training process

Balances image diversity and quality effectively

Abstract

We propose a new equilibrium enforcing method paired with a loss derived from the Wasserstein distance for training auto-encoder based Generative Adversarial Networks. This method balances the generator and discriminator during training. Additionally, it provides a new approximate convergence measure, fast and stable training and high visual quality. We also derive a way of controlling the trade-off between image diversity and visual quality. We focus on the image generation task, setting a new milestone in visual quality, even at higher resolutions. This is achieved while using a relatively simple model architecture and a standard training procedure.