This repository contains the code for the "Pose-Aware 3D Talking Face Synthesis using Geometry-guided Audio-Vertices Attention" paper.

The environment we used was python 3.8, a 3090 GPU, and Windows OS.

Note : Requirements

Before you start using our model, you need to install the requirements. To do so, we advise you to create a virtual environment. Then run pip install -r requirements.txt. It is worth noting that you will need to install pytorch3d manually.

Next, you will need to download some pre-trained models and pre-processing files at here. The file directories in the web drive and the local file directories are corresponding. Please download the files in checkpoints, ds_graph and template folders from the web drive and put them into the corresponding local folders respectively.

To try our model trained on VOCASET, you can run the following:

python test_voca.py

To try our model trained on MULTIFACE, you can run the following:

python test_multiface.py

If all goes well, you can see the video in the root directory.

If you need to change the audio or the 3D head model, you need to change the wav_path and template_path in test_voca.py or test_multiface.py (lines 16 and 17) and just run them again.

If you used this code or the paper, please consider citing our work:

@article{li2024pose,
  title={Pose-Aware 3D Talking Face Synthesis using Geometry-guided Audio-Vertices Attention},
  author={Li, Bo and Wei, Xiaolin and Liu, Bin and He, Zhifen and Cao, Junjie and Lai, Yu-Kun},
  journal={IEEE Transactions on Visualization and Computer Graphics},
  year={2024},
  publisher={IEEE}
}