Skip to Content

Who drew this? A computer… or Van Gogh?

A photo of the Neckarfront in Tubingen, Germany is shown ¨
in the style of B. J.M.W. Turner,  C. Vincent van Gogh, D. Edvard Munch.A photo of the Neckarfront in Tubingen, Germany is shown ¨
in the style of B. J.M.W. Turner,  C. Vincent van Gogh, D. Edvard Munch.
A photo of the Neckarfront in Tubingen, Germany is shown ¨ in the style of B. J.M.W. Turner, C. Vincent van Gogh, D. Edvard Munch.Image sourced from A Neural Algorithm of Artistic Style.

German researchers have created a machine-learning algorithm that can take a photograph and make it look like a drawing by Vincent van Gogh or Pablo Picasso. In addition to benefiting art lovers worldwide (think of the new Instagram filters this could unleash), this will help computer scientists figure out how to teach computers better ways to classify images. In short, computers will get better at “seeing.”

Companies ranging from Google and Facebook to the U.S. department of defense are all trying to make advances in computer vision, which will help usher in innovations as significant as driverless cars and as mundane as better social-networking software. Getting computers to help us parse and categorize visual data could help humans deal with the increasing flood of information, as well as helping us analyze the information we already have.

The paper—based on work from researchers at the University of Tubingen in Germany, with some help from a neuroscientist at Baylor College of Medicine in Houston—discusses how the scientists taught a neural network to render a photograph of a row of houses in each artist’s style. The researchers discovered that they had to strike a balance between optimizing for the content of the photograph (the row of houses) or the artistic style. If they let the computer put too much weight on one they ended up with images that were heavily distorted, although very stylized.

But in doing so, the authors discovered that the computer was able to separate the content of an image, such as a picture of a house or starry night from the style in which it was rendered. The researchers believe this is the first time that separation has been demonstrated. They also believe it could pave the way for computers that really do see more like humans and less like computers. And that would be truly game changing.

For more on the future of AI in the technology world see our Fortune video: