This is the Real-time generic face tracking in the wild with CUDA (MMSys 2014) written by Shiyang Cheng and Akshay Asthana. It is a real-time system that performs 66-landmarks face tracking along with 3D head pose estimation. We use a hybrid CPU-GPU implementation of Discriminative Response Map Fitting (DRMF) algorithm in this system, which are capable of achieving real-time performance at 30 to 45 FPS on ordinary consumer-grade computers.
The code requires a NVIDIA graphics card with least 2.0 compute capability, and CUDA Toolkit 5.0 or above should be installed in the system.
Intelligent Behaviour Understanding Group (iBUG)
 S. Cheng, A. Asthana, S. Zafeiriou, J. Shen and M. Pantic. Real-Time Generic Face Tracking in the Wild with CUDA [pdf]. MMSys 2014.
 A. Asthana, S. Zafeiriou, S. Cheng and M. Pantic. Robust Discriminative Response Map Fitting with Constrained Local Models [pdf]. CVPR 2013.