Some initial experiments using generative adversarial networks to transfer stylistic characteristics from one image to another.

These test images come from my initial experiments working with GANs or Generative Adversarial Networks. The objectives for this process were quite simple, in that I sought to learn about a technique, understand a little about how parameters modify the output and finally I wanted to document some of the results. It was especially important to keep track of things visually because of the nature of the process at hand. GANs can produce a lot of images, so taking the time to organise output and indeed to look carefully at these was vital.

As a basic data set, plants and especially flowers seemed a good start for exploring the terrain. There are two good personal reasons for this. Plants have a vast variety of complex structures and a homogeneous range of colours. Colour can be a demanding variable in art, plants and flowers just seem to naturally select a harmonious colour group to work with.

These first test images were a stepping stone to later work. What became quickly apparent with such a process, was that the parameters were vast, the outputs many and the danger of just playing around very strong. Therefore, a strict methodology was needed to keep track of the whole process and try to understand the creative possibilites.