Coarse-to-Fine Registration with SE(3)-Equivariant Representations

This is the official code repository for ICRA 2023 paper, Coarse-to-Fine Registration with SE(3)-Equivariant Representations [arxiv].

Preprocessed Dataset

ModelNet40 with occupancy labels

Train

$ python train.py

Demo

$ python demo.py --weights [checkpoints]

Installation

$ conda create -n cfreg python=3.8
$ conda install pytorch=1.9.0 cudatoolkit=11.1 -c pytorch -c nvidia
$ pip install -r requirement.txt

Citation

If you want to use it in your work, please cite it as

@inproceedings{lin2023coarse,
  title={Coarse-to-fine point cloud registration with se (3)-equivariant representations},
  author={Lin, Cheng-Wei and Chen, Tung-I and Lee, Hsin-Ying and Chen, Wen-Chin and Hsu, Winston H},
  booktitle={2023 IEEE International Conference on Robotics and Automation (ICRA)},
  pages={2833--2840},
  year={2023},
  organization={IEEE}
}