Home

Awesome

<h1> Deep Shells: Unsupervised Shape Correspondence with Optimal Transport </h1>

Implementation of the NeurIPS 2020 paper (arXiv). For a given pair of deformable 3D shapes, our algorithm produces high-quality correspondences.

Usage

Preprocessing

preprocess_data/preprocess_dataset.m

Datasets

Models

./models/

Train

Test

Citation

If you use our implementation, please cite:

@article{eisenberger2020deep,
  title={Deep Shells: Unsupervised Shape Correspondence with Optimal Transport},
  author={Eisenberger, Marvin and Toker, Aysim and Leal-Taix{\'e}, Laura and Cremers, Daniel},
  journal={Advances in Neural Information Processing Systems},
  volume={34},
  year={2020}
}

Axiomatic matching (smooth shells)

@inproceedings{eisenberger2020smooth,
  title={Smooth Shells: Multi-Scale Shape Registration with Functional Maps},
  author={Eisenberger, Marvin and Lahner, Zorah and Cremers, Daniel},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={12265--12274},
  year={2020}
}