EFGH

This repository contains the code (in PyTorch) for "EFGHNet: A Versatile Image-to-Point Cloud Registration Network for Extreme Outdoor Environment" paper (IROS 2022).

Requirements

• Python 3.8
• PyTorch 1.10
• RELLIS-3D dataset

Environment

conda create -n efgh python=3.8
conda activate efgh
pip install -r requirements.txt

Set up

cd lib 
python build_khash_cffi.py 
cd ..

Data

Download RELLIS-3D dataset from https://unmannedlab.github.io/research/RELLIS-3D

data
└── RELLIS-3D
    ├── RELLIS-3D
    |   ├── 00000
    |   |   ├── os1_cloud_node_kitti_bin
    |   |   ├── pylon_camera_node
    |   |   ├── calib.txt
    |   |   ├── poses.txt
    |   |   └── camera_info.txt    
    |   ├── 00001
    |   └── ..
    ├── RELLIS_3D
    |   ├── 00000
    |   |   └── transforms.yaml
    |   ├── 00001
    |   └── ..
    ├── pt_test.lst
    ├── pt_train.lst
    └── pt_val.lst

Train

Set data_root and ckpt_dir in the train_rellis.yaml file.

python main.py configs/train_rellis.yaml

Test

Set ckpt_path in the test_rellis.yaml file.

python main.py configs/test_rellis.yaml

Pretrained model

April 2023 update

Please check the config.yaml file to set the parameters before using the pretrained model: Download link

Acknowledgements

Our BCL implementation is based on https://github.com/laoreja/HPLFlowNet.

Citation

If you use our code or method in your work, please cite the following:

@article{jeon2022efghnet,
  title={EFGHNet: A Versatile Image-to-Point Cloud Registration Network for Extreme Outdoor Environment},
  author={Jeon, Yurim and Seo, Seung-Woo},
  journal={IEEE Robotics and Automation Letters},
  volume={7},
  number={3},
  pages={7511--7517},
  year={2022},
  publisher={IEEE}
}

Please direct any questions to Yurim Jeon at [email protected]