LoRCoN-LO: Long-term Recurrent Convolutional Network-based LiDAR Odometry

Video [link] Paper [link]

Download datasets

KITTI dataset (http://www.cvlibs.net/datasets/kitti/eval_odometry.php).

• velodyne laser data, calibration files, ground truth poses

Rellis-3D dataset (https://github.com/unmannedlab/RELLIS-3D).

• SemanticKITTI Format, Scan Poses files

Setup directories

bash setup_directories.sh

Enter dataset name

KITTI # KITTI dataset
Rellis-3D # Rellis-3D dataset

Move datasets

KITTI

Move calib files (from 00 to 10)

mv data_odometry_calib/dataset/sequences/00/calib.txt data/KITTI/calib/00.txt

Move pose files (from 00 to 10)

mv data_odometry_poses/dataset/poses/00.txt data/KITTI/pose/00.txt

Move scan files (from 00 to 10)

mv data_odometry_velodyne/dataset/sequences/00/velodyne data/KITTI/scan/00/

Rellis-3D

Move pose files (from 00 to 04)

mv Rellis_3D_lidar_poses_20210614/Rellis_3D/000000/poses.txt data/Rellis-3D/pose/00.txt

Move scan files (from 00 to 04)

mv Rellis_3D_os1_cloud_node_kitti_bin/Rellis_3D/000000/os1_cloud_node_kitti_bin data/Rellis-3D/scan/00/

Setup the environment

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

Change the config file

Change the config file (config/config.yaml) depending on your directory configuration.

Pre-process

• transform ground truth poses from cam to vel

python preprocess/transform_poses_cam_to_vel.py

Compute relative poses

python preprocess/relative_pose_calculator.py

Generate input data

python preprocess/gen_data.py

Train and Test

python train.py
python test.py

Pre-trained models

KITTI model wass trained with 00 to 08 sequences.

• Download

Rellis-3D model was trained with 00 to 03 sequences.

• Download

When downloading and running the model, please modify the checkpoint related code in confg/config.yaml.

Paper

@inproceedings{jung2023lorcon,
  title={LoRCoN-LO: Long-term Recurrent Convolutional Network-based LiDAR Odometry},
  author={Jung, Donghwi and Cho, Jae-Kyung and Jung, Younghwa and Shin, Soohyun and Kim, Seong-Woo},
  booktitle={2023 International Conference on Electronics, Information, and Communication (ICEIC)},
  pages={1--4},
  year={2023},
  organization={IEEE}
}

Acknowledgement

The input data generation module is adapted from Overlapnet.

License

Copyright 2022, Donghwi Jung, Jae-Kyung Cho, Younghwa Jung, Soohyun Shin, Seong-Woo Kim, Autonomous Robot Intelligence Lab, Seoul National University.

This project is free software made available under the MIT License. For details see the LICENSE file.