Blockchain
Ethereum (1)
Description
A GAN feedback loop that generates faces from scratch. This is the third evolutionary stage of my practice to link two custom trained pix2pix GANs together where one has been trained to transform face markers to face images and another one that makes the opposite transformation. Whilst the feedback loop starts from random noise, the single-purpose training of the models very quickly leads to emerging faces which do not remain static due to interpretation errors each of the models introduces into the process.
In 2017 I spent several months on trying to improve the amount of details pix2pix could produce by experimenting with changes to the GAN's architecture, altering the training methods and the hyperparameters. One reason this model's output is in greyscale is that it allowed my to increase the rendering resolution and get a richer variety of texture details.