Deep Photo Style Transfer

Adobe and a team at Cornell University have reportedly devised a way to transfer photo styles from one image to another.

The project will essentially make it possible to mimic the colors and lighting in a photo from a reference image. The applications of the capability are numerous; say for example a designer wants to modify a photo of a house against a grey background at night so that it has a blue sky during the day. Pretty cool.

The researchers used deep learning methods to capture lighting and color cues from the reference image, and developed a way to constrain the transformation so the changes applied in a natural way to the target photo without coloring outside the lines. While not "suitable" for photorealistic style tranfer, according to the researchers, the approach "successfully suppresses distortion and yields satisfying photorealistic style transfers in a broad variety of scenarios, including transfer of the time of day, weather, season, and artistic edits.

The technology may ultimately be integrated into photo editing tool like Adobe's Photoshop or Lightroom apps. For now, designers may want to dig into the research paper and explore the code available in this GitHub repository.