Although it might seem like a mere mobile application with which to play with different images of your face, it is really a program that is the result of a thorough work carried out by a group of researchers from the Universities of Kingston and Nottingham. In fact, it is an important advance in 3D facial reconstruction from a 2D image, since what had been done so far was a really difficult job to capture in a few seconds.
Project leaders say that “programmers have so far faced a series of methodological challenges such as establishing correspondences of different facial poses, as well as expressions and non-uniform lighting.” All that added to the fact that several images of the same 2D pose had to be loaded so that they could make a correct reconstruction of the face. Now, they have managed to eliminate all these potholes along the way, which only delayed the work, and created this program. Here’s an example of what it does:
They can do this thanks to what they call a “Convolutional Neural Network” (CNN), a learning system that helps a computer program to know how to perform, in this case, the 3D development on the 2D image. During this teaching period, a CNN has made the program know how to transform 2D pixels to 3D coordinates.
Although the tool works in many poses and with different facial expressions, it is recommended that close-up photos be used for better results.