__        ___  __          ___      
/  \ \  / |__  |__) \  / | |__  |  | 
\__/  \/  |___ |  \  \/  | |___ |/\|

Your objective is to make a self-portrait.

Everyone will work in pairs to mimic a generative adversarial network. For your own self-portrait, you will play the role of discriminator. For your partner’s self-portrait, you will play the role of generator. In other words, your “self-portrait” will in fact be drawn by your partner, according to your rules.

This project’s challenge is to assert your authorship programmatically. Since you are not drawing your own portrait, you have to find other ways to control the process and get a portrait you want to see. You have two levers that you can pull:

  1. The training data you use to establish what an image can look like
  2. The features you use to measure how correct an image is

Do not shy away from the challenge of asserting your authorship. You should have an opinion on what constitutes your self-portrait.

Background information

A generative adversarial network (GAN) is a technology used to generate images. It works by making two machines (i.e. neural networks) compete. One machine is called the generator, and it generates images. The other is called the discriminator, and it reviews them. The discriminator has seen a bunch of images and knows those source images well. Its job is to look at an image and see if it is "real" (meaning it is part of the training data) or "fake" (meaning it was generated from scratch.) The generator tries over and over again to get an image that fools the discriminator. When it does, the network calls that an output.

In this project you will learn how the technology works by pretending to be it. You will be tasked to “think like a computer” and follow procedures—which of course you will design.

This project intends to teach you about computer vision, how images contain information, and how authorship is relocated in an AI-generated ecosystem.

Partners