APB2FaceV2

Official pytorch implementation of the paper that is under review for ICASSP'21: "APB2FACEV2: REAL-TIME AUDIO-GUIDED MULTI-FACE REENACTMENT".

Using the Code

Requirements

This code has been developed under Python3.7, PyTorch 1.5.1 and CUDA 10.1 on Ubuntu 16.04.

Datasets in the paper

Download AnnVI dataset from Google Drive or Baidu Cloud (Key:str3) to /media/datasets/zhangzjn/Audio2Face/AnnVI.

Test

python3 test.py 

You can view results in checkpoints/AnnVI-Big/results

Train

The complete training code will be released as soon as the paper is accepted.

Citation

TBA

Acknowledgements

TBA