HiGS-Calib: A Hierarchical 3D Gaussian Splatting based Targetless Local-Consistent LiDAR-Camera Calibration Method

This repository contains the official authors implementation associated with the paper "HiGS-Calib: A Hierarchical 3D Gaussian Splatting based Targetless Local-Consistent LiDAR-Camera Calibration Method".

Testing Environment

We use Ubuntu 22.04 + CUDA 12.5 + PyTorch 2.4.0

Installing Dependencies

3DGS

Please first configure the environment according to the official configuration tutorial of gaussian-splatting .

ROMA

HiGS-Calib uses ROMA to extract inter-frame flows, please following the configuration tutorial of ROMA to configure the environment.

To use HiGS, you need to download the weights of ROMA from the official implementation or our link

Pytorch3D

Installing pytorch3D, using command

pip install "git+https://github.com/facebookresearch/pytorch3d.git"

Dataset

The reorganized version of KITTI360 prepared for HiGS-Calib is offered.

Running

To run the calibrator, simply use

python calibrate.py -s <path_of_the_dataset> --flow_path <path_of_the_flow_model_weights>  --optimizer_type sparse_adam  --cam_id X --data_seq X

For example

python calibrate.py -s HiGS-Calib-dataset/KITTI360 --flow_path weights/ --optimizer_type sparse_adam  --cam_id 00 (00 or 01) --data_seq 2 (0-4)