This project is used to reconstruct the 3d human face by fitting the landmarks of faces. By now, the fitting process which is based on flame has been implemented. The kenel is composed of two parts : 'landmarks detection' and '3d morphable model'.
The landmark detection is based on face-alignment
The fitting process of a sequence of images:
- camera calibration : only using the first 50 images
- identity fitting : only using the first 50 images
- expression and pose fitting : using of all images
- pytorch3d
- conda install -c pytorch pytorch=1.7.1 torchvision cudatoolkit=10.2
- conda install -c fvcore -c iopath -c conda-forge fvcore iopath
- conda install pytorch3d -c pytorch3d
- requirements.txt
Some scripts are in './scripts/'.
https://github.com/HavenFeng/photometric_optimization
- add eyes detection to align gaze
- photometric fitting
- implement the BFM fitting part