DeformingThings4D-Matching Dataset

This is a new dataset adapted from DeformThings4D for non-isometric shape matching task as introduced in our paper "Smooth Non-Rigid Shape Matching via Effective Dirichlet Energy Optimization", by Robin Magnet, Jing Ren, Olga Sorkine-Hornung and Maks Ovsjanikov.

You can find more details at: paper | code | supplementary

Dataset Details

• We select 8 humanoid categories from the original DeformThings4D dataset, including Crypto, Zlorp, Mannequin, Drake, Ninja, Prisoner, Pumpkinhulk, Skeletonzombie, as shown in the teaser figure from left to right respectively.
• We also provide Ortiz and Mousey in this repository with inter-category correpondences, but the cross-category correspondences are hard to obtain for these two models
• Each model has 15-30 different shapes in different poses that are remeshed independently with inter-category correspondences.
• See demo1_visualize_cross_category_map for an example of loading the cross-category correspondences
• See demo2_visualize_inter_category_map for an example of loading the inter-category correspondences

Construction Details

• We first find the humanoid models in DeformingThings4D dataset whose largest connected component contains more than 75% vertices (so we do not lose too many details once we only keep the largest connected component, which is easier than stiching disconnected components to get a watertight shape)
• For each model, we select poses from provided animations that have large enough variations and save them as Xk.obj, i.e., the k-th frame of the X animation.
• For each pose we apply LRVD algorithm to remesh it independently (with around 8K vertices) and track the inter-category correspondences
• We then use Wrap3D, a commercial software, to wrap different models to the Crypto model, from which we can extract the cross-category correspondences.

Please see our paper for more details.

Citation

If this dataset is used in your work, please do not forget to cite the original DeformThings4D paper, and LRVD method (we used for remeshing), besides our paper:

@article{li20214dcomplete, 
    title={4dcomplete: Non-rigid motion estimation beyond the observable surface.}, 
    author={Yang Li, Hikari Takehara, Takafumi Taketomi, Bo Zheng, and Matthias Nießner},
    journal={IEEE International Conference on Computer Vision (ICCV)},
    year={2021}
}

@article{yan2014low,
  title={Low-resolution remeshing using the localized restricted Voronoi diagram},
  author={Yan, Dong-Ming and Bao, Guanbo and Zhang, Xiaopeng and Wonka, Peter},
  journal={IEEE Transactions on Visualization and Computer Graphics (TVCG)},
  year={2014}
}

@inproceedings{magnet2022smooth,
  title={Smooth Non-Rigid Shape Matching via Effective Dirichlet Energy Optimization},
  author={Magnet, Robin and Ren, Jing and Sorkine-Hornung, Olga and Ovsjanikov, Maks},
  booktitle={International Conference on 3D Vision (3DV)},
  year={2022}
}

Comments

• The provided correspondences are not perfect, but in reasonably good quality as shown in the teaser.
• Please let us know (🐼[email protected]🐼) if you have any question regarding the dataset.