MoFA: Model-based Deep Convolutional Face Autoencoder for Unsupervised Monocular Reconstruction
https://arxiv.org/pdf/1703.10580v1.pdf