Ever thought the work of the masters didn’t look realistic enough to you? Researchers at UC Berkeley have created a way to make classic paintings look frighteningly close to photographs.
The researchers used a process called “image-to-image translation,” which reconstructs paintings as pictures, among other things. According to the article’s abstract:
We present an approach for learning to translate an image from a source domain X to a target domain Y in the absence of paired examples.
That’s way too many science-y words for me, but the proof is in the pretty pictures, and they don’t come prettier than this Monet-turned-rural landscape.
The scientists were able to turn photos into paintings by the same process, but we already have Prisma for that. Turning your Instagram pictures into Cezanne paintings is old news.
The researchers also used the process to alter photos to different seasons, or even different objects in the photos. You could, in theory, turn a selfie of you holding an orange in summer into a selfie of you holding an apple in winter … by Van Gogh.