PCA-based 3D Facial Reenactment From Single Image

Facial re-enactment from videos and images is one of the essential tasks in character animation applied in the creation of video games and movies. In this paper, we propose an efficient method for 3D facial re-enactment for online facial expression transferring. By approximating the pose, the approximation of expression parameters is reduced to a linear regression problem of which the closed-form solution is readily available, resulting in the improvement in inference speed. We also test the proposed method with multiple variants of initial blendshapes estimation and compare the results with the prior in term of both quality and speed.