Awesome

Designing One Unified Framework for High-Fidelity Face Reenactment and Swapping

This repository contains the official PyTorch implementation of the paper Designing One Unified Framework for High-Fidelity Face Reenactment and Swapping (ECCV2022).

Using the Code

Requirements

conda create -y -n uniface python=3.6.12
conda activate uniface
./install.sh

Data

Please download CelebA-HQ in data.

Please download VoxCeleb2 in data and follow the instrcution in FOMM official repository to perform preprocessing.

Inference

Please put test images to examples and create pair.txt to indicate the source and target file names. For example, 001_002 means the source file name is 001 and the target is 002. Please put pre-trained models in session. We release the separately trained models reenact and swap, the unified one will be available soon after we open source the journal version.

git clone https://github.com/xc-csc101/UniFace
python generate_swap.py   # test for swapping
python generate_reenact.py   # test for reenactment

Train

python train_reenact.py  # train for reenactment
python train_swap.py    # train for swapping

Acknowledgements

Our project is built on the StyleMapGAN and some codes are borrowed from pSp. We thank the authors for their excellent work.