Apps that convert your photos into paintings, with styles ranging from Monet to Lichtenstein, have been all the rage recently, with Prisma being one of the most popular offerings. A research team at UC Berkeley has now developed an artificial intelligence algorithm that can do the reverse – and much more.
Using ‘image style transfer’ their AI can convert Monet’s impressionist paintings into a much more realistic image that comes close to photo quality, giving you a good idea of the scene that the French painter was looking at when he put down his easel and started to paint. In addition, the system is capable of changing winter to summer in an image of a Yosemite scene, turn apples into oranges and horses into zebras. Of course it’s also possible to transform regular photos into Monets, Van Goghs or Cezannes.
The research project uses so-called ‘unpaired data’ which, in the words of the project team means that ‘we have knowledge of the set of Monet paintings and of the set of landscape photographs. We can reason about the stylistic differences between those two sets, and thereby imagine what a scene might look like if we were to translate it from one set into another.’
To achieve that the relationships between similar styles had to be coded in a way that can be understood by a computer and then the AI had to be trained using large number of photos from Flickr and other sources. In a final step the quality of the results was checked by both humans and machines before fine-tuning of the algorithms.
On some occasions the results are still far from perfect but overall the AI is impressively good at transferring styles from one image to another. More information is available on the project’s GitHub page.