Cascading Feature Extraction for Fast Point Cloud Registration

This repository contains the source code for the paper [Arxive link comming soon].

Method rotation error (˚) translation error (cm) chamfer error inferrence time (ms)
FGR 31.4 20 0.012 34
DCP-V2 12.6 17 0.011 12
RPMNet (Optim) 1.71 1.8 0.00085 58
RGM 1.56 1.5 0.00084 174
Proposed 1.23 1.3 0.00076 17

Getting Started

cascading_feature_extraction
|- datasets/
        |- modelnet40_ply_hdf5_2048.zip
        |- modelnet40_ply_hdf5_2048/
|- src/
    ...
|- pretrained/
        |- modelnet40.pth

Download dataset

Please add --insecure at the end of the installation command, in order of lack of certificate

curl -O https://shapenet.cs.stanford.edu/media/modelnet40_ply_hdf5_2048.zip

Requirements

pip install -r requirements

We confirmed that the code runs at Python>=3.7.

Train

cd src
python train.py

Test

cd src
python eval.py --resume ../pretrained/modelnet40.pth

Citation

@inproceedings{hisadome2021,
  title = {Cascading Feature Extraction for Fast Point Cloud Registration},
  author = {Hisadome, Yoichiro and Matsui, Yusuke},
  booktitle = {Proceedings of the British Machine Vision Conference},
  year={2021}
}

Acknowledgements

Our code is mainly based on RPMNet. We appriciate them for making the code available.

GitHub

View Github