RGB Face Tracking and Reconstrucion
2020, Feb 08
For the 3D-Scanning and Spatial Learning practical course we implemented an (almost) real-time face reconstruction algorithm based on Face2Face. The algorithm takes the RGB-stream from a webcam as input and fits a parametric model to the face using differential rendering and LM optimization. Facial keypoints are predicted using DLIB and serve as additional input for the optimization. The rendering pipeline was implemented in OpenGL and the optimization was done using CUDA.
Team Members: Patrick Radner, Wojciech Zielonka, Mustafa Işık
Instructor: Justus Thies