Triplet-Graph: Global Metric Localization based on Semantic Triplet Graph for Autonomous Vehicles

This is the code for Triplet-Graph. [paper]

You are also very welcome to check my personal Github for more frequent update.

If you find Triplet-Graph helpful, please consider citing:

@ARTICLE{ma2024triplet,
  author={Ma, Weixin and Huang, Shoudong and Sun, Yuxiang},
  journal={IEEE Robotics and Automation Letters}, 
  title={Triplet-Graph: Global Metric Localization Based on Semantic Triplet Graph for Autonomous Vehicles}, 
  year={2024},
  volume={9},
  number={4},
  pages={3155-3162}}

Install

Operation System

Tested on Ubuntu 20.04

Dependencies

• ROS
• Ceres
• yaml-cpp

Build

mkdir -p ~/tripletgraph_ws/src
cd ~/tripletgraph_ws/src
git clone https://github.com/lab-sun/Triplet-Graph.git
cd ..
catkin_make
source devel/setup.bash

Data preparation

• Pair list

We evaluate our method on SeamnticKitti dataset following paper SSC: Semantic Scan Context for Large-Scale Place Recognition.

• LiDAR scans

Please download LiDAR scans from the offical website of SemanticKitti dataset. Sequence-00, 02, 05, 06, 07, and 08 are used.

• Semantic label

We use semantic label from SemanticKitti and RangeNet++.

For your convenience, you can download all required data here (onely seq-06 is available). Unzip test_data.zip to folder src/Triplet-Graph.

Run

We provide three different launch files.

*Note that you should replace file paths as yours appropriately in config.yaml.

• tripletgraph_single_pair.launch This launch file will only run Triplet-Graph on a single pair of LiDAR scan for the evaluated sequence with visualization using rviz. Please use the following command:

  roslaunch tripletgraph tripletgraph_single_pair.launch

• tripletgraph_part_seq.launch This launch file will run Triplet-Graph on part of pairs of LiDAR scans for the evaluated sequence with visualization using rviz. Please use the following command:

  roslaunch tripletgraph tripletgraph_part_seq.launch

• tripletgraph_entire_seq.launch This launch file will run Triplet-Graph on all pairs of LiDAR scans for the evaluated sequence without visualization(for acceleration). It will take some time to be finished. Once finished, you should be able to find results in /your path/tripletgraph_ws/src/Triplet-Graph/results/, including score.txt, rte.txt, and rre.txt.

  roslaunch tripletgraph tripletgraph_entire_seq.launch

Evaluation

Based on results files (or you can download raw data for results reported in the paper), you can use eval.py to get the final results. Copy eval.py into the your result folder, and then run command python3 eval.py. numpy, sklearn, and matplotlib are required dependences for eval.py.

Contact

If you have any questions, please contact:

• Weixin Ma [weixin.ma@connect.polyu.hk]

Support Material

Here are the support material docs/appendix.pdf, which investigates the effects of some important parameters in Triplet-Graph.