The new technology has been devised by computer scientists based at the University of Nottingham, U.K. Attempst at creating 3D facial images have been variable and to do so typically requires multiple pictures of the same face from different angles. This is necessary to map every contour. With the new study, The Verge reports, the researchers simply fed a set of photographs and corresponding 3D models into a neural network. From this, they were able to teach and artificial intelligence system how to extrapolate the shape of a face from a single photograph. This has led to a rapid, 3D facial image generation program.
Central to the development was a Convolutional Neural Network (CNN). This is an area of artificial intelligence that uses machine learning to give computers the ability to learn without being explicitly programmed. According to the lead researcher, Dr Yorgos Tzimiropoulos: “The main novelty is in the simplicity of our approach which bypasses the complex pipelines typically used by other techniques. We instead came up with the idea of training a big neural network on 80,000 faces to directly learn to output the 3D facial geometry from a single 2D image.”
Potential applications
In terms of what the technology could be used for, apart from entertainment, there are several business applications, such as face and emotion recognition. In addition the image rendering process could be used to personalize computer games, enhance augmented reality, and to allow people to shop online for things that have until now, proved difficult unless a physical object is such as reading glasses. There are also potential medical applications, such a testing out medical devices that need to fit different body shapes.
With this latter point, the university are researching how deep learning can be applied to the human face. This includes 3D reconstruction and segmentation applied to the human face as well as to other parts of the human body.
Try it for yourself
For those wishing to test out the software, it can be tired out for free here.
To test out the imaging, Digital Journal’s Tim Sandle put the software to the test. Here’s before:
And here is one of the possible 3D images:
The research has been published in a white paper titled “Large Pose 3D Face Reconstruction from a Single Image via Direct Volumetric CNN Regression.”
