An Explicit Framework for 3D Point Cloud Normalization

Justin Baker*, Shih-Hsin Wang*, Tommaso de Fernex, and Bao Wang

This repository contains the official implementation for "An Explicit Framework for 3D Point Cloud Normalization" (ICML 2024).

In Progress: more flexible data handling and distributed normalization.

Usage

PyOrbit is a library for normalizing 3D point clouds. It is designed to be used in conjunction with NumPy or PyTorch point cloud data.

Currently two types of normalization are supported: PointCloud and CategoricalPointCloud. The Frame and CatFrame classes can be used to normalized the point cloud by calling .get_frame(point_cloud) or .get_frame(point_cloud, categorical_data).

Several useful examples can be found in the examples directory.

Training requires additional installation and can be performed by running the following command:

python3 ./training/ae_qm9.py

Installation

Datasets

ModelNet40 can be downloaded by running

python3 ./datasets/modelnet40.py

and then processed by running the jupyter notebook ./datasets/modelnet40.ipynb.

QM9 will be downloaded automatically by the torch_geometric library.

Requirements

Install PyOrbit as a library:

git clone https://github.com/Bayer-Group/alignment
cd alignment
pip3 install -e .

Training the autoencoder requires the additional library:

git clone https://github.com/Bayer-Group/giae
cd giae
pip3 install -e .

Citation

If you find our work useful in your research, please consider citing:

@inproceedings{
baker2024explicit,
title={An Explicit Frame Construction for Normalizing 3{D} Point Clouds},
author={Baker, Justin and Wang, Shih-Hsin and De Fernex, Tommaso and Wang, Bao},
booktitle={Proceedings of the 41st International Conference on Machine Learning},
pages={2456--2473},
year={2024},
editor={Salakhutdinov, Ruslan and Kolter, Zico and Heller, Katherine and Weller, Adrian and Oliver, Nuria and Scarlett, Jonathan and Berkenkamp, Felix},
volume={235},
series={Proceedings of Machine Learning Research},
month={21--27 Jul},
publisher={PMLR},
pdf={https://raw.githubusercontent.com/mlresearch/v235/main/assets/baker24a/baker24a.pdf},
url={https://proceedings.mlr.press/v235/baker24a.html},
}

@inproceedings{
wang2024rethinking,
title={Rethinking the Benefits of Steerable Features in 3D Equivariant Graph Neural Networks},
author={Shih-Hsin Wang and Yung-Chang Hsu and Justin Baker and Andrea L. Bertozzi and Jack Xin and Bao Wang},
booktitle={The Twelfth International Conference on Learning Representations},
year={2024},
url={https://openreview.net/forum?id=mGHJAyR8w0}
}

Acknowledgements

Our implementation is based on NumPy, PyTorch, and PyTorch Geometric.