Need help with 3D-Face-GCNs?
Click the “chat” button below for chat support from the developer who created it, or find similar developers for support.

About the developer

291 Stars 53 Forks MIT License 13 Commits 22 Opened issues


Towards High-Fidelity 3D Face Reconstruction from In-the-Wild Images Using Graph Convolutional Networks, CVPR 2020

Services available


Need anything else?

Contributors list

No Data

Towards High-Fidelity 3D Face Reconstruction from In-the-Wild Images Using Graph Convolutional Networks

CVPR 2020, [pdf] [arxiv]

by Jiangke Lin, Yi Yuan*, Tianjia Shao, Kun Zhou

In this paper, we seek to reconstruct the 3D facial shape with high fidelity texture from a single image, without the need to capture a large-scale face texture database.

An overview of our method.

Here is the official implementation.

1. Requirements


The shapes and coarse textures are from, the GCNs code is based on and mesh processing libraries from
Please refer to their project pages for the necessary files.

2. Prepare dataset

As we mentioned in the paper, we use a face segmentation network to segment out the non-face areas. Here, we treat the segmentation result as alpha channel and store it in a

file along with the face image. More specifically, we acquire the face segmentation result by a face segmentation network, then store the image and segmentation results as a
file, where the RGB channels store the image, and alpha channel stores the segmentation results. In the alpha channel, 0 means
non-skin region
and 255 represents
skin region
For efficiency, we then write all
images into a binary file in advance. Please change the data folder in
to yours.


3. Training

It is worth mentioning that, our network involves the mesh sampling algorithm. We save the sampling-related parameters into

files in advance and load them before training to avoid meaningless repeat calculation.
More details could be found in init_sampling().

After the dataset files are ready, the training can be started.

python --mode train

4. Future Works

As we continue to work on 3D face reconstruction and try to integrate such an algorithm into automatic game character creation, we can expect more results in the future.
We will also make some of our datasets for 3D face reconstruction public available in the future.
However, at present, you can try out our current automatic character creation method in the game Justice(逆水寒,


  title={Towards high-fidelity 3D face reconstruction from in-the-wild images using graph convolutional networks},  
  author={Lin, Jiangke and Yuan, Yi and Shao, Tianjia and Zhou, Kun},  
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},  

We use cookies. If you continue to browse the site, you agree to the use of cookies. For more information on our use of cookies please see our Privacy Policy.