Generative Adversarial Networks Based Approach for Artificial Face Dataset Generation in Acne Disease Cases

Abstract

Deep-Learning based approaches in dermatology face a significant problem regarding the availability of free open datasets. Recently, Generative Adversarial Networks (GANs) were successfully employed to generate artificial images through a combination of two models: Generator and Discriminator. In this work, we propose using StyleGAN2 to generate realistic artificial faces presenting acne diseases. The model uses a collection of authentic face images gathered from multiple sources regardless of the acquisition conditions such as resolution, pose, Etc. Results show that the model can produce an unlimited number of artificial faces of acne diseases. The biomedical community can take advantage of such a dataset to evaluate the performance of some specific algorithms.