Which director is making the most interesting use of technology? Right now, that’d be Kristen Stewart, and she has the academic receipts to prove it. Along with a research engineer from Adobe and produce David Shapiro, Stewart co-wrote “Bringing Impressionism to Life with Neural Style Transfer in Come Swim,” a short research paper that was posted to arXiv last week.
So, what’s Stewart have to say about machine learning? The paper explains how the technique of style transfer was implemented in her short film, Come Swim. You can dive into the technical details by reading the paper, but in general, style transfer involves applying a style (impressionism, in this case) to another piece. Neural style transfer, the subject of the paper, involves using machine learning to automate that process.
Creatively, the idea is brilliant and could become a big deal in movies. Using this tech, Stewart was able to apply an impressionist style to her live-action short film, creating a surreal effect. In this case, scenes from the movie were recreated (redrawn, says the paper) in the impressionist style.
Automation shouldn’t be taken as a lack of creativity. While the redrawing is being done by a computer, the paper recounts the process of adjusting settings to achieve just the right effect. With a little fine-tuning, the team was able to bring an impressionist painting to life in the form of a short film. It’s a super cool idea, and we can’t wait to see who’s going to use it next!
Via The Verge