This paper was submitted to IROS 2022. [youtube]
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
bash setup_directories.sh
Enter dataset name
KITTI # KITTI dataset
Rellis-3D # Rellis-3D dataset
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/
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/
conda create -n LoRCoN-LO python=3.8
conda activate LoRCoN-LO
pip install -r requirements.txt
Change the config file (config/config.yaml
) depending on your directory configuration.
- transform ground truth poses from cam to vel
python preprocess/transform_poses_cam_to_vel.py
python preprocess/relative_pose_calculator.py
python preprocess/gen_data.py
python train.py
python test.py
KITTI model wass trained with 00 to 08 sequences.
Rellis-3D model was trained with 00 to 03 sequences.
When downloading and running the model, please modify the checkpoint related code in confg/config.yaml
.
The input data generation module is adapted from Overlapnet.
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.