This repository contains our work in ICCV19 This paper proposes a deep Dense-Fine-Finer Network (DF2Net) to address the challenging problem of high-fidelity 3D face reconstruction from a single image. DF2Net is composed of three modules, namely D-Net, F-Net, and Fr-Net. It progressively refines the subtle facial details such as small crow’s feet and wrinkles. We introduce three types of data to train DF2Net with different training strategies. More details can be seen in our paper.
Xiaoxing Zeng, Xiaojiang Peng, Yu Qiao. DF2Net: A Dense-Fine-Finer Network for Detailed 3D Face Reconstruction. ICCV, 2019

• Pytorch 0.4.0
• Python 2.7
• Dlib
• PIL

We crop the raw face image with bounding box of face detection, you can change to other more advanced detector.

We also masking the cropped face with 68 face landmarks detector.

python demo.py

show_output.m

python pointcloud2rawmesh.py

There may be some visual different from our results of paper to our demo, this is due to the different of cropping and masking way.

@inproceedings{zeng2019df2net,
title={DF2Net: A Dense-Fine-Finer Network for Detailed 3D Face Reconstruction},
author={Zeng, Xiaoxing and Peng, Xiaojiang and Qiao, Yu},
booktitle={Proceedings of the IEEE International Conference on Computer Vision},
pages={2315--2324},
year={2019}
}

Thanks the authors of extreme 3d faces,PRNet, SfSNet, pix2vertex for their inspiring works.