Old Work - Landscapes And Nude Portraits (Nude Portraits At The Bottom Of The Page!)
Done in early 2018 - These works are distinct projects; but do not stand very well without each other. I always use them when explaining
my work for the first time; and what a GAN is.
My first real attempt at using a GAN to generate paintings was the below landscapes. I was able to get fairly realistic results; but realistic results are very boring after about 10 minutes.
Employing a fancy new machine and cutting edge algorithm to make the same sort of mediocre landscape paintings we have already been making for the past 500+ years seems a bit cynical.
The further down landscape images below (5th image onwards) are my attempts at generating an image of a more unusal images; but these are still not very interesting.
After I worked with landscapes; I realized that since it was much more interesting when the network did not correctly learn rules; I tried to generate nude portraits - and maximize the "misinterpretation" by the network.
The nude portraits (bottom of this page - after the landscapes) are much more interesting - and they really highlight conceptually exactly why I am interested in GANs - because of misinterpretation. Before GANs; generative art had to have programmed rules, and if rules are programmed there is not any room for the computer to do something completely unexpected (unless the programmer makes a typo or has an error in their logic). With GANs the rules are being 'fed' to the network throughout hundreds or thousands of examples in the dataset (e.g. the network learns these rules through thousands of landscape paintings: ground is on the bottom of an image; trees are in the middle; and sky is on the top).
With the nude portraits - I made sure the network was able to correctly learn 'rules' associated with small and local features of paintings (breasts, folds of fat, etc) - but failed to learn rules concerning the overall structure of the portraits (2 arms, 2 legs, 1 head, proportions, etc).
*The above text is very rough.. I wrote it on the train. I will probably rewrite it soon. Sorry.