Steroid Filter; Adobe Algorithm Copies Photo Styles

Technically called “Deep Photo Style Transfer”, researchers from Adobe and Cornell University have teamed up to develop an algorithm that lifts the photo style from one image to another.

The software is developed on pre-existing technology used in Prisma app which copies painting styles however; the software’s algorithm produces satisfying photorealistic style transfers in a wide range of scenarios including transfer of weather, season, artistic edits and time of day more than 80% of the time.

Here are some examples of the algorithm in action: "A" is the reference image, "B" is the image to be edited and "C" is the output

Example I

Example II

Example II

This level of accuracy and faithfulness was achievable by;

(a)    Constraining transformation to be locally affine in colour space from input to output in order to prevent distortions.

(b)   Introduction of an optional guidance to the style transfer process to avoid content mismatch. This guide is based on semantic segmentation of the inputs.

Hopefully in the nearest future, a refined version of this algorithm will find its way into Lightroom and or Photoshop but do not bank on it for now.

Dive into the tech details behind the algorithm here, also get view the code and more samples via GitHub