Everyone knows, the way you properly evaluate a — ahem — selfie, is by counting the number of likes it gets on social media. It’s a foundational system in our current society. Andrej Karpathy is a computer science PhD student at Stanford and he has developed a neural network that can automatically tell a good selfie from a bad one.
He fed roughly two million selfies into the software and it analyzed their attributes and correlated them against the number of likes the photos got.
Looking at the results, it seems like the best way to take a “good” selfie is to be a white woman with blonde hair. It does provide some other more useful insight, though, like cutting off the top of a person’s head actually increases its chances of getting a higher ranking. The bot is also very partial to lots of filters and other rather annoying elements like fake Instagram photo borders.
If you want to see what the bot thinks of your own selfie, you can use their bot on Twitter. I fed it one of my selfies and I got a 49%, which makes me below average. There’s no official numbers on how beards usually perform, but my guess is that it’s not very high.
While you may have already checked out on this article based on the simple fact that it’s about selfies, this kind of technology does raise a rather interesting point. Computers are already surprisingly capable when it comes to identifying faces and other objects in photos. Now, by tying in social networks and other user information, the computers are actually making quantitative judgments about photos.
Read more about the project here.