ofirkris / LatentAvatar

A PyTorch implementation of "LatentAvatar: Learning Latent Expression Code for Expressive Neural Head Avatar"

LatentAvatar: Learning Latent Expression Code for Expressive Neural Head Avatar

Paper | Project Page

Getting Started

• Create a conda environment conda env create -f environment.yml
• Install Pytorch3d pip install --no-index --no-cache-dir pytorch3d -f https://dl.fbaipublicfiles.com/pytorch3d/packaging/wheels/py38_cu113_pyt1120/download.html
• Download our dataset Google drive and place demo_dataset in this folder
• Download checkpoint of the landmark detector from mmpose

Training

First, train the avatar

python train_avatar.py

Second, train the Y-shaped architecture and the mapping MLP jointly!

python train_yvae.py

Evaluation

Self reenactment

python self_reenactment.py

Cross-identity reenactment

python cross_reenactment.py

Citation

@inproceedings{xu2023latentavatar,
  title={LatentAvatar: Learning Latent Expression Code for Expressive Neural Head Avatar},
  author={Xu, Yuelang and Zhang, Hongwen and Wang, Lizhen and Zhao, Xiaochen and Huang, Han and Qi, Guojun and Liu, Yebin},
  booktitle={ACM SIGGRAPH 2023 Conference Proceedings},
  year={2023}
}