Skip to content

Latest commit

 

History

History
158 lines (115 loc) · 5.9 KB

README.md

File metadata and controls

158 lines (115 loc) · 5.9 KB
Raw

StyleFaceV - Official PyTorch Implementation

This repository provides the official PyTorch implementation for the following paper:

StyleFaceV: Face Video Generation via Decomposing and Recomposing Pretrained StyleGAN3
Haonan Qiu, Yuming Jiang, Hang Zhou, Wayne Wu, and Ziwei Liu
Arxiv, 2022.

From MMLab@NTU affiliated with S-Lab, Nanyang Technological University and SenseTime Research.

[Project Page] | [Paper] | [Demo Video]

Generated Samples

Updates

  • [07/2022] Paper and demo video are released.
  • [07/2022] Code is released.

Installation

Clone this repo:

git clone https://github.com/arthur-qiu/StyleFaceV.git
cd StyleFaceV

Dependencies:

All dependencies for defining the environment are provided in environment/stylefacev.yaml. We recommend using Anaconda to manage the python environment:

conda env create -f ./environment/stylefacev.yaml
conda activate stylefacev

Datasets

Image Data: Unaligned FFHQ

Video Data: RAVDESS

Download the processed video data via this Google Drive or process the data via this repo

Put all the data at the path ../data.

Transform the video data into .png form:

python scripts/vid2img.py

Sampling

Pretrained Models

Pretrained models can be downloaded from this Google Drive. Unzip the file and put them under the dataset folder with the following structure:

pretrained_models
├── network-snapshot-005000.pkl  # styleGAN3 checkpoint finetuned on both RAVDNESS and unaligned FFHQ.
├── wing.ckpt                    # Face Alignment model from https://github.com/protossw512/AdaptiveWingLoss.
├── motion_net.pth               # trained motion sampler.
├── pre_net.pth
└── pre_pose_net.pth
checkpoints/stylefacev
├── latest_net_FE.pth            # appearance extractor + recompostion 
├── latest_net_FE_lm.pth         # first half of pose extractor
└── latest_net_FE_pose.pth       # second half of pose extractor

Generating Videos

python test.py --dataroot ../data/actor_align_512_png --name stylefacev \
    --network_pkl=pretrained_models/network-snapshot-005000.pkl --model sample \
    --model_names FE,FE_pose,FE_lm --rnn_path pretrained_models/motion_net.pth \
    --n_frames_G 60 --num_test=64 --results_dir './sample_results/'

Training

Pre Stage

If you want to use new datasets, please finetune the StyleGAN3 model first.

This stage is purely trained on image data and will help the convergence.

python train.py --dataroot ../data/actor_align_512_png --name stylepose \
    --network_pkl=pretrained_models/network-snapshot-005000.pkl \
    --model stylevpose --n_epochs 5 --n_epochs_decay 5
python train.py --dataroot ../data/actor_align_512_png --name stylefacev_pre \
    --network_pkl=pretrained_models/network-snapshot-005000.pkl \
    --model stylepre --pose_path checkpoints/stylevpose/latest_net_FE.pth

You can also use pre_net.pth and pre_pose_net.pth from the folder of pretrained_models.

python train.py --dataroot ../data/actor_align_512_png --name stylefacev_pre \
    --network_pkl=pretrained_models/network-snapshot-005000.pkl --model stylepre \
    --pre_path pretrained_models/pre_net.pth --pose_path pretrained_models/pre_pose_net.pth

Decomposing and Recomposing Pipeline

python train.py --dataroot ../data/actor_align_512_png --name stylefacev \
    --network_pkl=pretrained_models/network-snapshot-005000.pkl --model stylefacevadv \
    --pose_path pretrained_models/pre_pose_net.pth \
    --pre_path checkpoints/stylefacev_pre/latest_net_FE.pth \
    --n_epochs 50 --n_epochs_decay 50 --lr 0.0002

Motion Sampler

python train.py --dataroot ../data/actor_align_512_png --name motion \
    --network_pkl=pretrained_models/network-snapshot-005000.pkl --model stylernn \
    --pre_path checkpoints/stylefacev/latest_net_FE.pth \
    --pose_path checkpoints/stylefacev/latest_net_FE_pose.pth \
    --lm_path checkpoints/stylefacev/latest_net_FE_lm.pth \
    --n_frames_G 30

If you do not have a 32G GPU, reduce the n_frames_G (12 for 16G). Or only add supervision on pose representations:

python train.py --dataroot ../data/actor_align_512_png --name motion \
    --network_pkl=pretrained_models/network-snapshot-005000.pkl --model stylernns \
    --pose_path checkpoints/stylefacev/latest_net_FE_pose.pth \
    --lm_path checkpoints/stylefacev/latest_net_FE_lm.pth \
    --n_frames_G 30

Citation

If you find this work useful for your research, please consider citing our paper:

@misc{https://doi.org/10.48550/arxiv.2208.07862,
  doi = {10.48550/ARXIV.2208.07862},
  url = {https://arxiv.org/abs/2208.07862},
  author = {Qiu, Haonan and Jiang, Yuming and Zhou, Hang and Wu, Wayne and Liu, Ziwei},
  keywords = {Computer Vision and Pattern Recognition (cs.CV), FOS: Computer and information sciences, FOS: Computer and information sciences},
  title = {StyleFaceV: Face Video Generation via Decomposing and Recomposing Pretrained StyleGAN3},
  publisher = {arXiv},
  year = {2022},
  copyright = {arXiv.org perpetual, non-exclusive license}
}