Skip to content

The code of paper 'Learning to Aggregate and Personalize 3D Face from In-the-Wild Photo Collection'

License

Notifications You must be signed in to change notification settings

TencentYoutuResearch/3DFaceReconstruction-LAP

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

10 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Learning to Aggregate and Personalize 3D Face from In-the-Wild Photo Collection (CVPR 2021 Oral) Paper | Arxiv

Official pytorch implemetation of 'Learning to Aggregate and Personalize 3D Face from In-the-Wild Photo Collection'.

Introduction

This repository contains LAP (Learning to Aggregate and Personalize) framework for reconstructing 3D face. Right now we provide a version of demo for testing on in-the-wild images. The output size is 128 and the model is finetuned on CelebAMask-HQ Dataset.

Requirments

The code is tested on pytorch 1.3.0 with torchvision 0.4.1

pip install torch==1.3.0
pip install torchvision==0.4.1

Neural renderer is needed to render the reconstructed images or videos

pip install neural_renderer_pytorch

It may fail if you have a GCC version below 5. If you do not want to upgrade your GCC, one alternative solution is to use conda's GCC and compile the package from source. For example:

conda install gxx_linux-64=7.3
git clone https://github.com/daniilidis-group/neural_renderer.git
cd neural_renderer
python setup.py install

Facenet is also needed to detect and crop human faces in images.

pip install facenet-pytorch

DEMO

Download the pretrained model, and then run:

python demo.py --input ./images --result ./results --checkpoint_lap ./demo/checkpoint300.pth

Options:

--gpu: enable gpu

--detect_human_face: enable automatic human face detection and cropping using MTCNN provided in facenet-pytorch

--render_video: render 3D animations using neural_renderer (GPU is required)

Note:

The output depth is transformed by several options and functions, including tanh(), depth_rescaler and depth_inv_rescaler for better visualization. You could search along these options to find the original output depth and rescale it to a required range. The defined direction of normal in normal maps may be different to your required setting. If you want to accelarate the inference procedure, you may delete the branches irrelavant to reconstruct depth, and set anti_aliasing=False in each renderer.

License

The code contained in this repository is under MIT License and is free for commercial and non-commercial purposes. The dependencies, in particular, neural-renderer-pytorch, facenet, may have its own license.

Citation

@InProceedings{Zhang_2021_CVPR,
    author    = {Zhang, Zhenyu and Ge, Yanhao and Chen, Renwang and Tai, Ying and Yan, Yan and Yang, Jian and Wang, Chengjie and Li, Jilin and Huang, Feiyue},
    title     = {Learning To Aggregate and Personalize 3D Face From In-the-Wild Photo Collection},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    year      = {2021},
    pages     = {14214-14224}
}

About

The code of paper 'Learning to Aggregate and Personalize 3D Face from In-the-Wild Photo Collection'

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages