Zamen | زامن
Adobe tries to make selfies less embarrassing using AI and machine learning
Sensei, the arm of Adobe that fiddles around with AI and machine learning, has released a trailer for some new selfie-improving features they have in the works. Adobe hasn't announced if or when these features will be included in any of its apps, but the video is still a fun watch.
It shows a dude taking a bad, embarrassing selfie and improving it using tools that apply artificial depth of field, tweak the perspective so it's not clear it was taken from so close to the face, and steal aesthetics from other portraits of other dudes.
That last feature — the ability to copy the style of any given photograph and transfer it to another photograph — was described in detail by a joint research team last week. The researchers, from Adobe and Cornell University, published a paper titled “Deep Photo Style Transfer,” which outlines how the process is more complex and precise than merely applying an Instagram-style filter. The code they used is open source, and you can download it on GitHub if you want.
The selfie our demonstrator has at the end does look better, but I feel it would be irresponsible of me not to point out that the selfie he started with wouldn't have been so awful if he hadn't gone with the classic My Dad move of aiming the camera directly up his chin.
Another way to drastically improve your selfies, beyond relying on admittedly cool-looking Adobe AI, would be to listen to even one thing Kim Kardashian-West has ever said. Repeat after me: never take pictures of yourself from any lower than one foot above your head.