Use of generative adversarial networks for the creation and manipulation of facial images in the context of studying false memories and its effects on wrongful conviction cases: implementation of StyleGAN’s generative image modeling and style mixing properties to design an interface for experimentation purposes

l Laboratorio de Sueño y Memoria from Instituto Tecnológico de Buenos Aires (ITBA) studies the formation of false memories, and how these can be reduced or modified, and is in collaboration with the Innocence Project to investigate how these can lead to errors in convictions. From this research, the need arises to carry out experiments with human faces that are similar to each other, and how this similarity can result in the formation of false memories. In this project, we investigate a field of Artificial Intelligence (AI), Deep Learning, which can provide us with a solution to the generation of artificial faces. In particular, we implement a face generation model using a Generative Adversarial Network (GAN), with the aim of generating faces as realistic as possible, so that a human cannot distinguish them from real faces. StyleGAN, a particular implementation of the GAN network, was the chosen architecture, because in addition to producing images with high resolution quality, it presents a model that allows navigation of the latent space and the synthesis of faces, using style mixing properties. Finally, an application called FG-Style was developed and installed on a GPU-based server at ITBA so that the laboratory can have control over the face generation model, and over the generation of faces similar to a selected one, using StyleGAN’s style mixing properties to have a grip over the change of specific features of the generated faces." Proyecto final Ingeniería Informática (grado) - Instituto Tecnológico de Buenos Aires, Buenos Aires, 2021