The Miami Children’s Museum approached us to create an interactive installation for their new exhibit All About Art. The exhibit focuses on teaching children about the elements of art, like color, form, line, shape, space, and texture, as well as expose visitors to broad movements and styles of art. Our team’s experience as artists ourselves informed the projects we proposed and made us perfect candidates to win the work. We were tasked with creating an installation that was engaging and educational, as well as offered the kids a chance to take a piece of their experience with the exhibit home with them.
Our team has been closely following advancements in the field of machine learning and artificial intelligence. These techniques, applied to the domain of images, has seen great advancements in the last two years, especially in the task of artistic style transfer. This technique, only two years old, features the ability to separate artistic style from content, or subject matter. The ability to separating these semantic axis in images is an incredibly powerful technique, as it allows them to be manipulated independently. Artistic styles can be arbitrarily applied to photos to stylize them to look as if they were painted by Vincent Van Gogh, Pablo Picasso, or the early cave painters at Lascaux.
These techniques were a perfect fit for our contribution to the All About Art exhibit. We created a real-time style transfer photo booth and gallery that allows children to apply six artistic “filters” to a live video of themselves. These filters were algorithmically generated, or learned, from reference images of their corresponding styles: cubism, impressionism, pop art, mosaic, cave painting, and sketch. Kids can experiment with posing for different styles, and when they’re done, they can send themselves their stylized portraits via email. Photos that are taken are also added to the digital gallery wall in the exhibit and persist for a short while for others to see.
In addition to this work, we also created another installation in the same exhibit.
Branger_Briz would like to thank Leon Gatys, Alexander Ecker, Matthias Bethge, Justin Johnson, and Logan Enstrom for their open research in the domain of artistic style transfer.