This page provides information about Computer Generated Art projects in the Research Group of Alan Blair at UNSW, Sydney.
In this project, abstract art is generated through adversarial training between a genetic program (HERCL) artist and a CNN (LeNet) critic, based on photographs of ten famous landmarks. The artist acts as a function from pixel values x,y to intensity values R,G,B.
In this project, abstract art emerges through adversarial training (or coevolution) between a genetic program artist (generator) and a deep learning neural network critic (discriminator) based on images from the popular MNIST and CIFAR-10 datasets. The artist again acts as a function from pixel values x,y to intensity values (black and white or R,G,B).
In this project, a genetic program (HERCL) is used for both the artist and the critic. The artist creates an image by issuing a sequence of drawing commands in the form of turtle graphics. The critic makes its decision based on a set of statistical features extracted from the images via OpenCV.
"Imagination is a good servant, and a bad master. The simplest explanation is always the most likely."- Hercule Poirot