The pytorch implementation for our ECCV2022 paper "Learning Dynamic Facial Radiance Fields for Few-Shot Talking Head Synthesis".

[Project] [Paper] [Video Demo]

• Python 3.8.11
• Pytorch 1.9.0
• Pytorch3d 0.5.0
• torchvision 0.10.0

For more details, please refer to the requirements.txt. We conduct the experiments with a 24G RTX3090.

• Download 79999_iter.pth from here to data_util/face_parsing

• Download exp_info.npy from here to data_util/face_tracking/3DMM

• Download 3DMM model from Basel Face Model 2009:

  cp 01_MorphableModel.mat data_util/face_tracking/3DMM/
  cd data_util/face_tracking
  python convert_BFM.py
  

Put the video ${id}.mp4 to dataset/vids/, then run the following command for data preprocess.

sh process_data.sh ${id}

The data for training the base model is [here].

sh run.sh ${id}

Some pre-trained models are [here].

Change the configurations in the rendering.sh, including the iters, names, datasets, near and far.

sh rendering.sh

This code is built upon the publicly available code AD-NeRF and GRF. Thanks the authors of AD-NeRF and GRF for making their excellent work and codes publicly available.

Please cite the following paper if you use this repository in your reseach.

@inproceedings{shen2022dfrf,
   author={Shen, Shuai and Li, Wanhua and Zhu, Zheng and Duan, Yueqi and Zhou, Jie and Lu, Jiwen},
   title={Learning Dynamic Facial Radiance Fields for Few-Shot Talking Head Synthesis},
   booktitle={European conference on computer vision},
   year={2022}
}