Taking a Day Off from B&W Art

Back to the art of B&W tomorrow. Today I’d like to share something that I found fascinating – from many perspectives, just one of which is photography.

Scientists at the University of Tubingen in Germany, used an artificial intelligence based neural network to create artistic images when provided with a photo and an example painting to learn a style from. They presented this paper on August 26th. Five “learned-by-example” painting are shown below.

Don’t peek at the text beneath these examples until you decide how many of the five famous artists (B through F) you recognize through their paintings. OK – now you can look. I didn’t recognize B.

To see these images LARGE, click on the link above to the paper, go to page 5, it’s a PDF page so use your PDF reader’s size option to enlarge it.

neural net art

I spent six months (many years ago) on a sabbatical doing artificial intelligence research. I love art and study books on art (including a reference on all of the paintings in the Louvre) to guide my photography. These two pieces of trivia only begin to describe why I found this paper and its results fascinating.

Tomorrow – Luminosity Tips

Enter your email address to follow this blog and receive notifications of new posts by email.

2 thoughts on “Taking a Day Off from B&W Art

  1. I only recognised C. Does this make me an art philistine? I have a strange feeling that it does. On the subject of artificial intelligence, I think the human race will rue the day that AI enters the battlefield paradigm.

    • I think I got 4 only because of all the research I did to prepare the abstract photography presentation I did last year. E & F were natural outgrowths of that – especially F. The Scream (D) also was recognizable because it was a “style-landmark” when created and appears in art literature often because of that.

      I have a solution to AI on the battlefield – arm everything with paint and brushes (but no palette knives). My research was in an area called Genetic Algorithms. Now in terms of robots taking over, this is the scary one since – like the name suggests – devices designed with this approach evolve. Survival of the fittest is how I looked at it – poor performers (algorithm-wise) dropped out of the gene-pool and the most fit moved on to breed even better performers. I demonstrated this in a large number of applications and the results were amazing – even scary. Example – backing an 18-wheeler into a very tight space: it learned to do this in less than 5 “generations”. Similarly a famous AI problem of keeping a broom balanced on a moving wagon which it learned in only 1/2 the number of trials as did the best previous conventional approach. Now – translate that into battlefield devices. ;(

Comments are closed.