Elle O’Brien

Generative Adversarial Network Self-Portrait (2019). This video is a walk through latent space, a montage of images generated by a Generative Adversarial Network trained on thousands of images of my own face. It’s a self-portrait rendered by a neural network. The portrait is the set of all possible faces the neural network could generate- although only an infinitely small fraction of those faces can be seen in a finite amount of time. To make this image, I used transfer learning with Nvidia’s StyleGAN architecture. The model was initialized with weights from their released model for generating faces (collected from the photo-sharing website Flickr), and then retrained on a custom dataset of approximately 7000 images of my own face.