PIANO: A Parametric Hand Bone Model from Magnetic Resonance Imaging

PIANOLayer is a differentiable PyTorch layer that deterministically maps from pose and shape parameters to hand bone joints and vertices. It can be integrated into any architecture as a differentiable layer to predict bone meshes for data-driven fine-grained hand bone anatomic and semantic understanding from MRI or even RGB images.

To learn about PIANO, please visit our website: https://liyuwei.cc/proj/piano

You can find the PIANO paper at: https://www.ijcai.org/proceedings/2021/0113.pdf

---

For comments or questions, please email us at: Yuwei Li (liyw@shanghaitech.edu.cn)

System Requirements:

Python Dependencies:

• Numpy
• pickle
• Pytorch
• Trimesh (for mesh saving)

Getting Started:

Model file: PIANO_RIGHT_dict.pkl

Demo pose: demo_pose.pkl

python demo.py

Acknowledgements:

This model and code was developped and used for the paper PIANO: A Parametric Hand Bone Model from Magnetic Resonance Imaging for IJCAI21. See project page

It reuses part of the great code from manopth by Yana Hasson and pytorch_HMR by Zhang Xiong!

If you find this code useful for your research, consider citing:

@inproceedings{li2021piano,
  title     = {PIANO: A Parametric Hand Bone Model from Magnetic Resonance Imaging},
  author    = {Li, Yuwei and Wu, Minye and Zhang, Yuyao and Xu, Lan and Yu, Jingyi},
  booktitle = {Proceedings of the Thirtieth International Joint Conference on
               Artificial Intelligence, {IJCAI-21}},
  publisher = {International Joint Conferences on Artificial Intelligence Organization},
  editor    = {Zhi-Hua Zhou},
  pages     = {816--822},
  year      = {2021},
  month     = {8},
  note      = {Main Track},
  doi       = {10.24963/ijcai.2021/113},
  url       = {https://doi.org/10.24963/ijcai.2021/113}
}