zhalok / apb2face Goto Github PK

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 [email protected]

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

• 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

1. Train Audio-to-Landmark model.

   python3 audio2landmark/main.py

2. Train Landmark-to-Face model.

   cd landmark2face
   sh experiments/train.sh

   you can watch the checkpoint in checkpoints/man1_Res9

3. 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

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

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}
}

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