A Deep Convolution Generative Adversarial Network for the Production of Images of Human Faces

Abstract

Generative models get huge attention by researchers in different topics of artificial intelligence applications, especially generative adversarial networks (GANs) which have demonstrated good performance in data generation. In this paper, we would like to explore the potential of this class of models in producing human faces images. For that, we will use Deep Convolutional Generative Adversarial Network (DCGAN). Since that, the evaluation of GANs is still difficult even with the existing metrics like Inception Scores (IS), Mode Score (MS), Kernel Inception Distance (KID), Fréchet Inception Distance (FID), Multi-Scale Structural Similarity (MS-SSIM), etc. Thus, the best possible evaluation remains that carried out by human evaluators. This is why we propose a new hybrid measure combining qualitative and quantitative evaluation, we called this measure: Measuring the Quality of the Features of an Image (MEQFI). The images produced with the DCGAN method were trained on three well known datasets from the literature and the results were evaluated with MEQFI.