It’s not only really cool, but also free and publicly available, thanks to the researchers who developed it.
Demonstration obtained with an example face.
3D reconstructions have long been a painstaking task. You generally need several pictures from different angles, and for something as complex as the human face, results have typically been lackluster. But a recent paper from UK scientists demonstrated impressive capabilities using a single, forward-facing image.
The researchers from the University of Nottingham and Kingston University developed a so-called convolutional neural network (CNN). This type of network allows algorithms to learn a bit like humans do, using interconnected neuron-like nodes.
An example of how the algorithm "sees" the faces. Credits: Jackson et al.
Convolutional networks were inspired by biological processes in that the connectivity pattern between neurons resembles the organization of the animal visual cortex. As a result, they are especially used in analyzing visual imagery.
Given enough time and resources, these networks can find patterns in all sorts of data, though the relevancy and usefulness of those patterns remains a human challenge. Still, some networks have proven surprisingly capable of mimicking creativity.
An example of how the algorithm “sees” the faces. Credits: Jackson et al.
Using this network, the team was able to produce remarkably accurate 3D facial reconstructions from a single photo. You can try it here, either using 8 example faces provided on the website, or by uploading your own images.
If you want to dig a little deeper, researchers also uploaded the paper and the code they are using — both of which can be accessed for free.
So far, almost a million faces have been uploaded and over 2 million models have been generated (presumably most models are done with the example faces, which include Barack Obama, Alan Turing, and Marie Curie).
For now, the technology is limited to human faces — as was dramatically highlighted by my failed first attempt, which was to upload a picture of my cat.
The most important use for this technology is in video monitoring and facial recognition. It could also be implemented in computer games or augmented reality where you could generate a character that looks just like you.
However, there are also some concerns: as smartphone producers are replacing fingerprint sensors with facial recognition, there are concerns that this type of algorithm could trick the security software. Aaron Jackson, the study’s first author thinks that’s unlikely. He does concede, however, that the technology could be used by authoritarian governments for population monitoring.
For now, we can enjoy the simple pleasure of viewing our faces in 3D. Ah, the joys of the internet!
Enjoyed this article? Join 40,000+ subscribers to the ZME Science newsletter. Subscribe now!
