Abstract
Generative Adversarial Networks (GANs) convergence in a high-resolutionsetting with a computational constrain of GPU memory capacity (from 12GB to 24GB) has been beset with difficulty due to the known lack of convergence ratestability. In order to boost network convergence of DCGAN (Deep ConvolutionalGenerative Adversarial Networks) and achieve good-looking high-resolutionresults we propose a new layered network structure, HDCGAN, that incorporatescurrent state-of-the-art techniques for this effect. A novel dataset, Curt\'oZarza (CZ), containing human faces from different ethnical groups in a widevariety of illumination conditions and image resolutions is introduced. CZ isenhanced with HDCGAN synthetic images, thus being the first GAN augmented facedataset. We conduct extensive experiments on CelebA and CZ.