Artificial Intelligence

This AI Can Turn a Photo of a Face into a 3D Image

Have you ever noticed that it's sometimes hard to recognize somebody if you only know them from a photograph? Maybe it's because looking at a still image is a lot different than looking at a living person. But sometimes a person just looks completely different from different angles. Turns out, figuring out what a 3D face looks like only from a photograph has long been a puzzle for computers as well — but a new AI might have solved the problem, and you can try it out yourself.

AI Can See Clearly Now

What makes it so hard to derive a three-dimensional image from a two-dimensional photograph? For computers, the problem is lighting and the strange crags of the human face. On a scale of Elijah Wood to Danny Trejo, humans have a huge variety of faces. And the more complicated the face, the harder it's been for computers to work out exactly how it all fits together.

But a new, AI-assisted method developed by researchers at The University of Nottingham and Kingston University has solved the problem once and for all. The system uses deep learning, where the machine develops its own set of parameters for determining the correct answer by being fed a series of data and that data's correct interpretation. In this instance, it was given a 3D model of a subject's face, then a series of images of that subject in all kinds of lighting.

The result? You can upload any images of face that you want, whether that's you and your loved ones, or your famous faves. Unfortunately, the links don't last long enough to share with you here, but two of our favorites were the Penguin and Voldemort, looking like the latter had donated half his nose to the former.

This ... Is CNN

The crucial ingredient in the development of this tool was a convolutional neural network, or a CNN for short. These systems have proven themselves extremely effective at identifying images since they work by recognizing patterns at both a large and a small scale.

CNNs have also been used to analyze video and writing, and have even successfully beat a human player in Go. Perhaps most astonishingly, they've been used in drug discovery, by predicting how drugs will interact on a chemical level with the human body. Even as the bots grow more competent, we should probably remember that their ability to learn is a bit beyond ours, and that maybe we shouldn't empower something that we can't understand.

Researcher Aaron Jackson on His Selfie to 3D Model Technology

Share the knowledge!
Written by Reuben Westmaas December 20, 2017

Curiosity uses cookies to improve site performance, for analytics and for advertising. By continuing to use our site, you accept our use of cookies, our Privacy Policy and Terms of Use.