APB2Face: A Python repository from zhangzjn

APB2Face — Official PyTorch Implementation

Official pytorch implementation of the paper "APB2FACE: AUDIO-GUIDED FACE REENACTMENT WITH AUXILIARY POSE AND BLINK SIGNALS, ICASSP'20".

For any inquiries, please contact Jiangning Zhang at 186368@zju.edu.cn

Using the Code

Requirements

This code has been developed under Python3.7, PyTorch 1.3.1 and CUDA 10.1 on Ubuntu 16.04.

# Install python3 packages
pip3 install -r requirements.txt

Inference

Download pretraind Audio-to-Landmark model for the person man1 to the path landmark2face/APB/man1_best.pth.
Download pretraind Landmark-to-Face model for the person man1 to the path landmark2face/checkpoints/man1_Res9/latest_net_G.pth

python3 test.py

You can view the result in result/man1.avi

Training

Train Audio-to-Landmark model.
```
python3 audio2landmark/main.py
```
Train Landmark-to-Face model.
```
cd landmark2face
sh experiments/train.sh
```
you can watch the checkpoint in checkpoints/man1_Res9

Do following operations before the test.

copy audio2landmark/APBNet.py landmark2face/APB/APBNet.py  # if you modify APBNet.py
copy audio2landmark/APBDataset.py landmark2face/APB/APBDataset.py  # if you modify APBDataset.py
copy audio2landmark/checkpoints/man1-xxx/man1_best.pth landmark2face/APB/man1_best.pth

Datasets in the paper

We propose a new AnnVI dataset, you can download it from Google Drive or Baidu Cloud (Key:str3).

Citation

If you think this work is useful for your research, please consider citing:

@inproceedings{zhang2020apb2face,
  title={APB2FACE: Audio-Guided Face Reenactment with Auxiliary Pose and Blink Signals},
  author={Zhang, Jiangning and Liu, Liang and Xue, Zhucun and Liu, Yong},
  booktitle={ICASSP 2020-2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
  pages={4402--4406},
  year={2020},
  organization={IEEE}
}

Acknowledgements

We thank for the source code from the great work pytorch-CycleGAN-and-pix2pix.

zhangzjn/APB2Face