Our research group applies deep learning technique to restore 3D face models from 2D portrait photos. Lots of sample data in Revopoint’s 3D face database (ZX-3DFaceData-2018.0) are used to conduct neural network training to output 3D models. With a single 2D portrait photo, high-resolution depth image can be output, and 3D point cloud can be restored to make the face more lifelike and detailed.
The restored 3D face model has vivid color and texture and a precise size, breaking the limits of 2D photos and offering better visual experience. It can be widely used in cosmetic surgery, games and video communication.