A team of developers named Gatys, Ecker, and Bethge recently developed an implementation of a technique known as a “style transfer,” which involves taking a specific pattern and “applying” it to a piece of video, such that the available surfaces in the video take on the texture of the original pattern. It’s kind of like a face swap only more ambitious.
A few months have passed, and a clever individual named Joshi Bhautik has tried to apply the technique as a way of mashing up great art and classic cinema. Specifically, he took a painting by Pablo Picasso, one of his “Les Femmes d’Alger” (Women of Algiers) series, which looks like this:
... and used it as the base image for a deep neural net-based style transfer on Stanley Kubrick’s masterpiece 2001: A Space Odyssey, which is full of striking images to say the least.
Once you do that, the generally stately, slow cinematography of the movie becomes a shimmering kaleidoscope, as seen in the following image:
The method has the peculiar effect of turning the entire movie into a version of the phantasmagorical, psychedelic journey Dave Bowman goes on for several minutes at the end of the movie, a sequence MAD magazine once compared to “crashing through the brand-new 105-story Jupiter Museum of Op Art.”
Here’s Bhautik’s description of what this is:
‘2001: A Space Odyssey’ rendered in the style of Picasso using deep neural network based style transfer. The cubist style had mixed results in the transfer; you can see that big empty blocks of colour didn’t map coherently between the frames. I’m working on a solution for that :]
Bhautik was good enough to upload a 90-second clip with short sequences from many different parts of the movie (including the Jupiter Museum of Op Art), but I’m not sure if this represents an excerpt from the entire movie or if only that much footage was processed.