Remember how Apple unveiled its $1,000 iPhone X with facial recognition software that allows you to unlock your phone with your face? Well good news, people who don’t have enough to worry about in the realm of cybersecurity already! Now there’s a neural net that can take just about any 2D face picture and model it as a 3D face. And while any security threats the tech poses are probably not imminent, it’s still pretty easy to imagine this technique being used for something nefarious.
Why, just look at this:
The Verge reported on the groundbreaking A.I. development, which was made by a team of researchers from the U.K. at The University of Nottingham and Kingston University. The researchers note in their paper that “3D face reconstruction is a fundamental Computer Vision problem of extraordinary difficulty” and that they were able to tackle it by training a Convolutional Neural Network (CNN) to guess at what 3D models of 2D images would look like.
Essentially, the researchers fed a large series of 2D facial pictures and their corresponding 3D facial models into the CNN, which learned how to spot patterns between the 2D pictures and 3D models. Now, when fed fresh 2D faces, the neural net is able to guess, based on prior examples, exactly what the 3D faces should look like.
And it works pretty dang well. Here’s former President Obama:
Simply feeding 2D images and their 3D model counterparts to a neural net likely undersells the amount of work that went into this simultaneously entertaining and frightening breakthrough. It took lead researcher, Aaron S. Jackson, about nine months to do this research, according to a comment he made on Reddit. It’ll probably only take a couple of weeks for Charlie Brooker to turn it into an episode of Black Mirror though.
If you want to test out a 2D picture of your own face, you can check out the 3D reconstruction tool here. But be prepared for a major face journey.
Image: Netflix
What do you think about this facial modeling technology? Construct some thoughts in the comments below!
Images: Wikimedia / Edward Liu
