Dense 3D Face Landmarks

Repository for the paper - A lightweight 3D dense facial landmark estimation model from position map data

Prepare Data -

Follow the instruction in face3d page to generate the UV map data from the 300W-LP dataset.

Make sure to install mesh_core_cython in the local python environment by running -

# directory - face3d/mesh/cython
pip install cython 
python3 setup.py build_ext --inplace 
python3 setup.py install

• Point the BFM.mat and BFM_UV.mat in generate_posmap_300WLP.py
• Update the input_path and output_path in generate_posmap_300WLP.py

Training -

run - train_mobilnet.py

Inference -

run - inference.py

Pretrained Checkpoint -

https://drive.google.com/file/d/1pfUZRMzLh8m53RI3mOqbjDfJGeOZHE_i/view?usp=sharing

Check the Configs/config.py for the configuration details.

Sample Results -

Citation

If you use this code, please consider citing:

@article{basak2023lightweight,
  title={A lightweight 3D dense facial landmark estimation model from position map data},
  author={Basak, Shubhajit and Mangapuram, Sathish and Costache, Gabriel and McDonnell, Rachel and Schukat, Michael},
  journal={arXiv preprint arXiv:2308.15170},
  year={2023}
}