3D face: rapid, merely and accumulate reconstruction

By Jianzhu Guo, Xiangyu Zhu, Yang Yang, Fan Yang, Zhen Lei and Stan Z. Li.

Introduction

This work extends 3DDFA, named 3DDFA_V2, titled In opposition to Swiftly, Correct and Stable 3D Dense Face Alignment, authorized by ECCV 2020. The supplementary area topic is right here. The gif above exhibits a demo of the monitoring result.
This repo is the legit implementation of 3DDFA_V2.

Compared to 3DDFA, 3DDFA_V2 achieves higher efficiency and steadiness. Moreover, 3DDFA_V2 contains the rapid face detector FaceBoxes as a substitute of Dlib. An easy 3D render written by c++ and cython is also incorporated. Whilst you are attracted to this repo, lawful strive it on this google colab! Welcome for precious considerations and PRs ?

Getting started

Requirements

Stare necessities.txt, tested on macOS and Linux platforms. Show that this repo makes spend of Python3. Basically the most valuable dependencies are PyTorch, numpy and opencv-python, and heaps others.

Utilization

Clone this repo

git clone https://github.com/cleardusk/3DDFA_V2.git
cd 3DDFA_V2

Discover the cython model of NMS, and Sim3DR

Flee demos

# working on tranquil image, four alternatives: 2d_sparse, 2d_dense, 3d, depth
python3 demo.py -f examples/inputs/emma.jpg

# working on movies
python3 demo_video.py -f examples/inputs/movies/214.avi

# working on movies smoothly by taking a look ahead by `n_next` frames
python3 demo_video_smooth.py -f examples/inputs/movies/214.avi

The implementation of monitoring is merely by alignment. If the top pose > 90° or the motion is just too rapid, the alignment can also fail. A threshold is ancient to trickly test the monitoring state, but it’s unstable.

It is seemingly you’ll well focus on over with demo.ipynb or google colab for the step-by-step tutorial of working on the tranquil image.

As an example, working python3 demo.py -f examples/inputs/emma.jpg -o 3d will give the final result below:

Acknowledgement

The FaceBoxes module is modified from FaceBoxes.PyTorch

Quotation

If your work or research advantages from this repo, please cite two bibs below : )

@inproceedings{guo2020in direction of,
    title =        {In opposition to Swiftly, Correct and Stable 3D Dense Face Alignment},
    writer =       {Guo, Jianzhu and Zhu, Xiangyu and Yang, Yang and Yang, Fan and Lei, Zhen and Li, Stan Z},
    booktitle =    {Complaints of the European Conference on Computer Imaginative and prescient (ECCV)},
    year =         {2020}
}

@misc{3ddfa_cleardusk,
    writer =       {Guo, Jianzhu and Zhu, Xiangyu and Lei, Zhen},
    title =        {3DDFA},
    howpublished = {url{https://github.com/cleardusk/3DDFA}},
    year =         {2018}
}