The creation and animation of digital characters in movies, TV productions and social interactions requires detailed 3D models of their facial morphologies. Typically, for digital doubles of real actors these models are obtained from 3D captures of the actor face using complex hardware setups or a high number of hours of manual work by artists.
The purpose of this internship is to investigate an alternative, where the facial appearance would be generated using a deep generative network (GAN) and some constraints to drive the result.
A first objective will be to evaluate performances and limitations of existing solutions (typically styleGAN2 and FTGAN). Then evaluate how to ingest “features” to drive the generation of faces (also called text to face generation). And finally develop a solution to map the generated “appearance” on a proprietary 3D adapted face model.
http://arxiv.org/abs/1912.04958
https://arxiv.org/abs/1904.05729
The internship is located at InterDigital Research & Innovation, Rennes, France
