FaceTuneGAN: Face Autoencoder for Convolutional Expression Transfer Using Neural Generative Adversarial Networks

In this paper, we present FaceTuneGAN, a new 3D face model representation\ndecomposing and encoding separately facial identity and facial expression. We\npropose a first adaptation of image-to-image translation networks, that have\nsuccessfully been used in the 2D domain, to 3D face geometry. Leveraging\nrecently released large face scan databases, a neural network has been trained\nto decouple factors of variations with a better knowledge of the face, enabling\nfacial expressions transfer and neutralization of expressive faces.\nSpecifically, we design an adversarial architecture adapting the base\narchitecture of FUNIT and using SpiralNet++ for our convolutional and sampling\noperations. Using two publicly available datasets (FaceScape and CoMA),\nFaceTuneGAN has a better identity decomposition and face neutralization than\nstate-of-the-art techniques. It also outperforms classical deformation transfer\napproach by predicting blendshapes closer to ground-truth data and with less of\nundesired artifacts due to too different facial morphologies between source and\ntarget.\n