Facial Performance Capture with Deep Neural Networks

We present a deep learning technique for facial performance capture, i.e., the transfer of video footage into a motion sequence of a 3D mesh representing an actor’s face. Specifically, we build on a conventional capture pipeline based on computer vision and multiview video, and use its results to train a deep neural network to produce similar output from a monocular video sequence. Once trained, our network produces high-quality results for unseen inputs with greatly reduced effort compared to the conventional system. In practice, we have found that approximately 10 minutes worth of high-quality data is sufficient for training a network that can then automatically process as much footage from video to 3D as needed. This yields major savings in the development of modern narrativedriven video games involving digital doubles of actors and potentially hours of animated dialogue per character.