High-resolution Deep Convolutional Generative Adversarial Networks

TL;DR

This paper introduces HDCGAN, a high-resolution GAN architecture with improved convergence stability, and presents a new dataset Curtó & Zarza, augmented with synthetic images, achieving state-of-the-art results in face image generation.

Contribution

The paper proposes HDCGAN, a layered network architecture with techniques for stable high-resolution image synthesis, and introduces Curtó & Zarza, a bias-free face dataset augmented with GAN-generated images.

Findings

HDCGAN achieves state-of-the-art MS-SSIM of 0.1978 on CelebA.

HDCGAN attains a Fréchet Inception Distance of 8.44.

Curtó & Zarza is the first GAN-augmented face dataset.

Abstract

Generative Adversarial Networks (GANs) [Goodfellow et al. 2014] convergence in a high-resolution setting with a computational constrain of GPU memory capacity has been beset with difficulty due to the known lack of convergence rate stability. In order to boost network convergence of DCGAN (Deep Convolutional Generative Adversarial Networks) [Radford et al. 2016] and achieve good-looking high-resolution results we propose a new layered network, HDCGAN, that incorporates current state-of-the-art techniques for this effect. Glasses, a mechanism to arbitrarily improve the final GAN generated results by enlarging the input size by a telescope {\zeta} is also presented. A novel bias-free dataset, Curt\'o & Zarza, containing human faces from different ethnical groups in a wide variety of illumination conditions and image resolutions is introduced. Curt\'o is enhanced with HDCGAN synthetic images, thus being the first GAN augmented dataset of faces. We conduct extensive experiments on CelebA [Liu et al. 2015], CelebA-hq [Karras et al. 2018] and Curt\'o. HDCGAN is the current state-of-the-art in synthetic image generation on CelebA achieving a MS-SSIM of 0.1978 and a FR\'ECHET Inception Distance of 8.44.