New Deep Photo Style Transfer Imaging Algorithm Accurately Copies Photo Styles
Ron Perillo / 8 years ago
Researchers from Adobe and Cornell University have come up with a new imaging algorithm called “Deep Photo Style Transfer” that hopefully will find its way to the latest version of Adobe Lightroom as it is quite impressive. The new feature is able to assess the photo style of one picture and apply it accurately to another as if they were taken with the same settings. It is similar to Prisma’s filter creator app that recreates photos as photo-paintings in the style the artist, but the Deep Photo Style Transfer imaging technology is much more impressive since it is actually dealing with photos and not simulating paint strokes.
The new imaging algorithm can match the tones of objects as well as the environment, day or night. Even if the source photo for example was taken during daylight, it can be compared to a reference photo shot during night time or at dusk, and the resulting photo will look as if it were taken during the same time as the reference image. It even includes photos where unnatural filters were applied such as HDR scenes from desaturated sources, etc. Until now, this kind of photo recreation and manipulation is only possible via human retouching and would take a long time to do as the artist has to do a lot of problem solving to render the resulting image correctly.
The research paper explaining the method used by the researchers is available on Arxiv and the code itself is available on GitHub.