We introduce the concept of unconstrained real-time 3D facial performance
capture through explicit semantic segmentation in the RGB input. To ensure
robustness, cutting edge supervised learning approaches rely on large training
datasets of face images captured in the wild. While impres