The official PyTorch repository for Controllable Radiance Fields for Dynamic Face Synthesis, 3DV 2022.
In this repository, we propose a Controllable Radiance Field (CoRF): 1) Motion control is achieved by embedding motion features within the layered latent motion space of a style-based generator; 2) To ensure consistency of background, motion features and subject-specific attributes such as lighting, texture, shapes, albedo, and identity, a face parsing net, a head regressor and an identity encoder are incorporated. On head image/video data we show that CoRFs are 3D-aware while enabling editing of identity, viewing directions, and motion.
We are cleaning code and will release it soon.
We also verify the effacacy of CoRF on monocular human body RGB videos. In the figure below we show 9 synthetic human bodies from multiple views. In each row, we use different motion representations to control the standing gesture.
If you find our work or repo useful in your research, please consider citing our paper:
@inproceedings{zhuang2022controllable,
author = {P. Zhuang and L. Ma and O. Koyejo and A.~G. Schwing},
title = {Controllable Radiance Fields for Dynamic Face Synthesis},
booktitle = {Proc. 3DV},
year = {2022},
}