Kristen Stewart is known on the Internet as the blank-faced leading woman named Bella Swan in the top-grossing Twilight Saga films. But perhaps that will change now as the celebrity who wrote about artificial intelligence on a scientific paper – something that not so many Hollywood superstars do.
Recently, she took a different career path and started being a writer and director for the short film Come Swim, where a machine learning technique known as “style transfer” was used to create an impressionistic visual style. For the laymen, the said technique is defined to be applying the aesthetic of one image or video to another.
She did this project with special effects engineer Bhautik J Joshi and producer David Shapiro. In line for the work in the film, Stewart has co-written a paper and published it on arXiv, a popular online repository for non-peer reviewed work.
Entitled “Bringing Impressionism to Life with Neural Style Transfer in Come Swim” which was also published through the Cornell University Library without the peer review, the paper details a case study on how to use the technique in a film.
It explores a “recently-developed technique that uses neural networks to artistically redraw an image in the style of a source style image”.
The paper describes the film as a “poetic, impressionistic portrait of a heartbroken man underwater”. This can be seen from the painting of Stewart showing a “man rousing from sleep”, which expresses the film’s aesthetic.
The trio used existing neural networks to imitate the painting-like effect, by transferring the style of the painting in a test frame and fine-tuning it with “blocks of color and texture”. The process of tuning is repeatedly done to get the desired effects and applied it to the film.
It’s surprising for someone like Kristern Stewart to be involved in artificial intelligence but it’s good that these machine learning tools are slowly getting into the limelight. Thanks to open-source AI frameworks like Tensor Flow and Keras, anyone could try these tools and techniques when it comes to machine learning.
Source: Telegraph UK
