The largest three-dimensional facial database ever created.

Or the biggest theft of biometric data

As humanoid robots become more advanced, a curious challenge begins to follow: how to make a machine recognize and even reproduce human expressions in a truly natural way, such as laughter, micro-expressions, subtle movements of the eyes and mouth.

All of this is part of human communication, but it is extremely difficult to replicate in artificial systems. Now, Chinese researchers may have taken an important step in that direction. A team led by scientists from the Chinese Academy of Sciences developed one of the largest three-dimensional facial databases ever created.

The project gathered nearly 200,000 high-fidelity 3D facial scans captured directly from real human faces, this data forms a huge set of geometric information that allows the structure of the human face to be precisely studied.

Unlike traditional facial recognition systems that depend on two-dimensional images, this new model works directly with V of 3D points, in this type of representation, the face is not a textured image, but a collection of thousands of points in space that describe the exact geometry of the face.

To interpret this data, the researchers developed an artificial intelligence model called CF-GAT, designed to analyze these three-dimensional structures more efficiently. The system uses a graphical attention network that manages to identify subtle variations in curvature in the face, essential details for detecting facial expressions and anatomical reference points.

The above allows regions such as eyes, nose, mouth and expression lines to be located with greater precision, even when the face is moving and this can have important implications. Humanoid robots, digital assistants and virtual avatars will be able to interpret human expressions in a much more natural way, in addition, biometric systems will be able to become more precise with the identification of people.

The database also includes information on dynamic 4D facial expressions, as well as full three-dimensional models of the human body. Ultimately, this research reveals something interesting. To make robberies seem human, it is not enough to teach artificial intelligence to think. It is also necessary to teach machines to understand the geometry of the human face itself.