Kristen Stewart — the actress best known for “Twilight” — has co-written a paper on machine learning.
The paper outlines the use of neural style transfer in Stewart’s directorial debut, “Come Swim”, which is about to premiere at Sundance Film Festival. Neural style transfer turns normal images into impressionist art, and is used by popular photo app Prisma.
The paper, first spotted by Quartz, is co-bylined with Adobe research engineer Bhautik J. Joshi and producer David Shapiro. It was published yesterday on ArXiv, a repository run by Cornell University for scientific papers that are not yet peer-reviewed.
“Come Swim” describes itself as “half realist, half impressionist portraits” of one man’s day. A teaser video for the short film is shot in a surreal style.
According to the paper, the project was based on an impressionistic painting of Stewart’s, which shows a man waking up.
The paper outlines the repetitive technical process of trying to replicate the feel of the painting on-screen. Here’s a sample:
“The painting itself evokes the thoughts an individual has in the first moments of waking (fading in-between dreams and reality), and this theme is explored in the introductory and final scenes where this technique is applied. This directly drove the look of the shot, leading us to map the emotions we wanted to evoke to parameters in the algorithm as well as making use of more conventional techniques in the 2D compositing stage.”
The response to the paper has been positive from the academic community, if slightly bemused.
I used to tell Twilight jokes in my Intro to AI class. I take it all back.
— Mark Riedl ???? (@mark_riedl) January 19, 2017