readme.md

April 27, 2019 ยท View on GitHub

How to build

  1. Prepaing reconstructed 3D dense alignment result

You can download the reconstructed results from BaiduYun / Google Drive or run main.py in the root directory.

  1. Run rendering.m to rendering 3D face meshes. The rendered images will be in obama_res@dense directory.

  2. Run python3 convert_imgs_to_video.py obama_res@dense to convert images to one video.

A frame of the video:

Vertex 3D

The output video named obama_res@dense.mp4 lies in this directory.

A simple Python render

Thanks for liguohao96 for contributions of the original version of this simple Python render.

Speed: ~25ms for one frame (720p), CPU.

After preparing the vertices, just run

python3 rendering.py && python3 convert_imgs_to_video.py obama_res@dense_py

A frame of the outputed video:

Vertex 3D

The output video named obama_res@dense_py.mp4 lies in this directory.

Future

Welcome for contributions of a FASTer Python/PyTorch render.