RangePlace

The code for our paper:

RangePlace: A Hierarchical Range Image Transformer for LiDAR-based Place Recognition

Developed by Ji Li.

Beijing Institute of Technology

Dependencies

coming soon ...

Datasets

KITTI Odometry Dataset

Ford Campus Dataset

Before the network training or evaluation, run the below code to generate pickles with positive and negative point clouds for each anchor point cloud.

# Generate training tuples for the KITTI Dataset
cd generating_queries/ 
python generate_training_tuples_kitti.py --dataset_root <dataset_root_path>

# Generate evaluation tuples
python generate_test_kitti.py --dataset_root <dataset_root_path>
python generate_test_ford.py --dataset_root <dataset_root_path>

<dataset_root_path> is a path to dataset root folder, e.g. /data/kitti_datasets/. Before running the code, ensure you have read/write rights to <dataset_root_path>, as training and evaluation pickles are saved there.

Dataset Structure

  data_root_folder (KITTI for example) follows:
  ├── 00
  │   ├── depth_map
  │     ├── 000000.png
  │     ├── 000001.png
  │     ├── 000002.png
  │     ├── ...
  │   └── pointcloud_locations.csv
  ├── 01
  ├── 02
  ├── ...
  └── 10

Training

To train the network, run:

cd training
python train.py --config ../config/config_kitti.txt --model_config ../models/rangplace.txt

Testing

coming soon ...

License

Our code is released under the MIT License (see LICENSE file for details).

Acknowledgement

We have intensively borrowed code from the following repositories. Many thanks to the authors for sharing their code.

OverlapTransformer

MinkLoc3Dv2

SwinTransformer

JiLiBIT/RangePlace