Tomographic Map Matching

This repository hosts the code used to perform the studies in our paper titled Efficient and Distributed Large-Scale 3D Map Registration using Tomographic Features.

Installation

The algorithms have been implemented to run within Apptainer (formerly Singularity) containers for portability and execution on various platforms, including HPC clusters. Tested on Singularity CE 3.8.1, 4.1.2 and Apptainer 1.3.3.

1. Follow the installation steps from Apptainer website to install the latest version.
2. Clone the repository recursively & download additional files

git clone --recursive https://github.com/RISC-NYUAD/tomographic-map-matching
cd tomographic-map-matching
make init

Running an algorithm on data

Some example data configurations are provided for the data used in the experiment. Other configurations can be generated using scripts in utils.

To build the container for a particular algorithm, use make {name}, with container names:

• roitr for RoITr
• geotr for GeoTransformer
• dgr for DeepGlobalRegistration
• buffer for BUFFER
• consensus for the proposed method
• all for all of the above (takes a long time)

Once the container is built, the algorithm can be run by invoking launch_image.sh script with the folder name:

./launch_image.sh {Consensus,BUFFER,RoITr,DeepGlobalRegistration,GeoTransformer}

A simple demo pair can be run using

./launch_image.sh Consensus --data_config /data/config/demo/demo.json --parameter_config /workspace/config/consensus-interiornet.json

If you would like to see the visuals, allow xhost access by running xhost +local: and appending --visualize flag to the above sample:

./launch_image.sh Consensus --data_config /data/config/demo/demo.json --parameter_config /workspace/config/consensus-interiornet.json --visualize

Partial data used for the studies can be found in this NYU UltraViolet repository.