MCG-NJU / SADRNet

[TIP 2021] SADRNet: Self-Aligned Dual Face Regression Networks for Robust 3D Dense Face Alignment and Reconstruction
Apache License 2.0
127 stars 27 forks source link
face-alignment face-reconstruction

SADRNet

Paper link: SADRNet: Self-Aligned Dual Face Regression Networks for Robust 3D Dense Face Alignment and Reconstruction

image

Requirements

python                 3.6.2
matplotlib             3.1.1  
Cython                 0.29.13
numba                  0.45.1
numpy                  1.16.0   
opencv-python          4.1.1
Pillow                 6.1.0                 
pyrender               0.1.33                
scikit-image           0.15.0                
scipy                  1.3.1
torch                  1.2.0                 
torchvision            0.4.0

Pretrained model

Link: https://drive.google.com/file/d/1mqdBdVzC9myTWImkevQIn-AuBrVEix18/view?usp=sharing .

Please put it under data/saved_model/SADRNv2/.

Please set ./SADRN as the working directory when running codes in this repo.

Predicting

The network takes cropped-out 256×256×3 images as the input.

Training

Acknowledgements

We especially thank the contributors of the face3d codebase for providing helpful code.