Paper | Project | Video | Supplement
[IEEE RA-L 2025]
Hanjun Kim
·
Minwoo Jung
·
Wooseong Yang
·
Ayoung Kim
- 2025/10/08 : SHeRLoc is accepted in IEEE RA-L 2025.
- 2025/10/10 : Our project page is available via [Project].
- 2025/10/23 : First release of SHeRLoc code.
- 2025/10/25 : Supplementary material is available via [Supplement].
Code was tested using Python 3.8 with PyTorch 1.9.1 and MinkowskiEngine 0.5.4 on Ubuntu 20.04 with CUDA 10.2. Note: CUDA 11.1 is not recommended as there are some issues with MinkowskiEngine 0.5.4 on CUDA 11.1. The following Python packages are required:
- PyTorch (version 1.9.1)
- MinkowskiEngine (version 0.5.4)
- pytorch_metric_learning (version 1.0 or above)
- wandb
- matplotlib
2-1. Clone the Repository
Clone the SHeRLoc repository and enter to the project directory:
git clone https://github.com/hanjun815/SHeRLoc.git
2-2. Build the Docker Image
Build the Docker image using the provided Dockerfile and start a container with GPU support. Build the image:
cd SHeRLoc/docker
docker build -t sherloc .
Run the container:
docker run -it --rm --gpus all --net host --shm-size=1g -v $(pwd)/../..:/code -v /data3:/data3 sherloc
Note: Adjust the docker run command to suit your environment.
Modify the PYTHONPATH environment variable to include absolute path to the project root folder:
export PYTHONPATH=$PYTHONPATH:/code/SHeRLoc
2-3. Download the Preprocessed HeRCULES Dataset
Move the downloaded tar.gz to each sequence directory inside SHeRLoc/datasets/HeRCULES, then extract it.
# example
cd SHeRLoc/datasets/HeRCULES/Bridge/01
tar -zxvf Continental.tar.gz
tar -zxvf Navtech.tar.gz
tar -zxvf Navtech_576.tar.gz
tar -zxvf Aeva.tar.gz
3-0 Preprocessing
Preprocess 4D radar point clouds, spinning radar scans, and LiDAR point clouds. Pass the folder containing the raw data to --input, and the folder where preprocessed results should be saved to --output. Preprocess all data you plan to use for training or evaluation in advance.
- (Skippable) Calibration with Mountain 01 Sequence
cd SHeRLoc/datasets
python calibration_continental.py
python calibration_navtech.py
python calibration_matching.py
The heterogeneous radar calibration constant Ccorrect obtained through calibration is already reflected in the preprocessing code (Ccorrect = 2.1025).
- 4D Radar Preprocessing
cd SHeRLoc/datasets
python preprocess_4D.py --input HeRCULES/Bridge/01/raw_continental --output HeRCULES/Bridge/01/Continental
- Spinning Radar Preprocessing
cd SHeRLoc/datasets
python preprocess_Spinning.py --input HeRCULES/Bridge/01/raw_navtech --output HeRCULES/Bridge/01/Navtech
- LiDAR Preprocessing
cd SHeRLoc/datasets
python preprocess_LiDAR.py --input HeRCULES/Bridge/01/raw_aeva --output HeRCULES/Bridge/01/Aeva
- Spinning Radar Preprocessing (384*576 size image with 360 degree FOV for homogeneous Spinning radar PR)
cd SHeRLoc/datasets
python preprocess_Spinning_576.py --input HeRCULES/Bridge/01/raw_navtech --output HeRCULES/Bridge/01/Navtech_576
3-1 Training
Train with the standard model (SHeRLoc):
- Heterogeneous Radar Place Recognition
cd SHeRLoc/training
python train.py --config ../config/config.txt --model_config ../models/Hetero_SHeRLoc.txt
- 4D Radar Place Recognition
cd SHeRLoc/training
python train.py --config ../config/config.txt --model_config ../models/4d_SHeRLoc.txt
- Spinning Radar Place Recognition
cd SHeRLoc/training
python train.py --config ../config/config.txt --model_config ../models/Spinning_SHeRLoc.txt
- LiDAR to Spinning Radar Place Recognition
cd SHeRLoc/training
python train.py --config ../config/config.txt --model_config ../models/LiDAR_SHeRLoc.txt
Train with the smaller model (SHeRLoc-S):
- Heterogeneous Radar Place Recognition
cd SHeRLoc/training
python train.py --config ../config/config.txt --model_config ../models/SHeRLoc_S.txt
- 4D Radar Place Recognition
cd SHeRLoc/training
python train.py --config ../config/config.txt --model_config ../models/4d_SHeRLoc_S.txt
- Spinning Radar Place Recognition
cd SHeRLoc/training
python train.py --config ../config/config.txt --model_config ../models/Spinning_SHeRLoc_S.txt
- LiDAR to Spinning Radar Place Recognition
cd SHeRLoc/training
python train.py --config ../config/config.txt --model_config ../models/LiDAR_SHeRLoc_S.txt
3-2 Pre-trained Models
Pre-trained models are available in the weights directory:
Hetero_SHeRLoc.pth: Heterogeneous Radar PR Model (320-dimension descriptor)Hetero_SHeRLoc_S.pth: Heterogeneous Radar PR Model (256-dimension descriptor)4d_SHeRLoc.pth: 4D Radar PR Model (320-dimension descriptor)4d_SHeRLoc_S.pth: 4D Radar PR Model (256-dimension descriptor)Spinning_SHeRLoc.pth: Spinning Radar PR Model (320-dimension descriptor)Spinning_SHeRLoc_S.pth: Spinning Radar PR Model (256-dimension descriptor)LiDAR_SHeRLoc.pth: LiDAR to Spinning Radar PR Model (320-dimension descriptor)LiDAR_SHeRLoc_S.pth: LiDAR to Spinning Radar PR Model (256-dimension descriptor)
3-3 Evaluation
To evaluate pretrained models run the following commands.
For the standard model (SHeRLoc):
- Heterogeneous Radar Place Recognition
cd SHeRLoc/eval
python evaluate.py --dataset_root /code/SHeRLoc/datasets/HeRCULES --map_seq Sports_Complex/01 --query_seq Sports_Complex/01 --model_config ../models/Hetero_SHeRLoc.txt --weights ../weights/Hetero_SHeRLoc.pth
- 4D Radar Place Recognition
cd SHeRLoc/eval
python evaluate.py --dataset_root /code/SHeRLoc/datasets/HeRCULES --map_seq Sports_Complex/01 --query_seq Sports_Complex/02 --model_config ../models/4d_SHeRLoc.txt --weights ../weights/4d_SHeRLoc.pth
- Spinning Radar Place Recognition
cd SHeRLoc/eval
python evaluate.py --dataset_root /code/SHeRLoc/datasets/HeRCULES --map_seq Sports_Complex/01 --query_seq Sports_Complex/02 --model_config ../models/Spinning_SHeRLoc.txt --weights ../weights/Spinning_SHeRLoc.pth
- LiDAR to Spinning Radar Place Recognition
cd SHeRLoc/eval
python evaluate.py --dataset_root /code/SHeRLoc/datasets/HeRCULES --map_seq Sports_Complex/01 --query_seq Sports_Complex/01 --model_config ../models/LiDAR_SHeRLoc.txt --weights ../weights/LiDAR_SHeRLoc.pth
For the smaller model (SHeRLoc-S):
- Heterogeneous Radar Place Recognition
cd SHeRLoc/eval
python evaluate.py --dataset_root /code/SHeRLoc/datasets/HeRCULES --map_seq Sports_Complex/01 --query_seq Sports_Complex/01 --model_config ../models/Hetero_SHeRLoc_S.txt --weights ../weights/Hetero_SHeRLoc_S.pth
- 4D Radar Place Recognition
cd SHeRLoc/eval
python evaluate.py --dataset_root /code/SHeRLoc/datasets/HeRCULES --map_seq Sports_Complex/01 --query_seq Sports_Complex/02 --model_config ../models/4d_SHeRLoc_S.txt --weights ../weights/4d_SHeRLoc_S.pth
- Spinning Radar Place Recognition
cd SHeRLoc/eval
python evaluate.py --dataset_root /code/SHeRLoc/datasets/HeRCULES --map_seq Sports_Complex/01 --query_seq Sports_Complex/02 --model_config ../models/Spinning_SHeRLoc_S.txt --weights ../weights/Spinning_SHeRLoc_S.pth
- LiDAR to Spinning Radar Place Recognition
cd SHeRLoc/eval
python evaluate.py --dataset_root /code/SHeRLoc/datasets/HeRCULES --map_seq Sports_Complex/01 --query_seq Sports_Complex/01 --model_config ../models/LiDAR_SHeRLoc_S.txt --weights ../weights/LiDAR_SHeRLoc_S.pth
- When using the code or dataset, please cite our paper:
@article{kim2025sherloc,
title = {SHeRLoc: Synchronized Heterogeneous Radar Place Recognition for Cross-Modal Localization},
author = {Kim, Hanjun and Jung, Minwoo and Yang, Wooseong and Kim, Ayoung},
journal = {IEEE Robotics and Automation Letters},
year = {2025},
publisher = {IEEE},
}
@INPROCEEDINGS { hjkim-2025-icra,
AUTHOR = { Hanjun Kim and Minwoo Jung and Chiyun Noh and Sangwoo Jung and Hyunho Song and Wooseong Yang and Hyesu Jang and Ayoung Kim },
TITLE = { HeRCULES: Heterogeneous Radar Dataset in Complex Urban Environment for Multi-session Radar SLAM },
BOOKTITLE = { Proceedings of the IEEE International Conference on Robotics and Automation (ICRA) },
YEAR = { 2025 },
MONTH = { May. },
ADDRESS = { Atlanta },
}
- Maintainer: Hanjun Kim ([email protected])