This codebase proposes modular light python and pytorch implementations of several LiDAR Odometry methods

Overview

pyLiDAR-SLAM

This codebase proposes modular light python and pytorch implementations of several LiDAR Odometry methods, which can easily be evaluated and compared on a set of public Datasets.

It heavily relies on omegaconf and hydra, which allows us to easily test the different modules and parameters with few but structured configuration files.

This is a research project provided "as-is" without garanties, use at your own risk. It is actively used for Kitware Vision team internal research thus is likely to be heavily extended, rewritten (and hopefully improved) in a near future.

Overview

KITTI Sequence 00 with pyLiDAR-SLAM

pyLIDAR-SLAM is designed to be modular, multiple components are implemented at each stage of the pipeline. Its modularity can make it a bit complicated to use. We provide this wiki to help you navigate it. If you have any questions, do not hesitate raising issues.

The documentation is organised as follows:

  • INSTALLATION: Describes how to install pyLiDAR-SLAM and its different components
  • DATASETS: Describes the different datasets integrated in pyLiDAR-SLAM, and how to install them
  • TOOLBOX: Describes the contents of the toolbox and the different modules proposed
  • BENCHMARK: Describes the benchmarks supported in the Dataset /!\ Note: This section is still in construction

The goal for the future is to gradually add functionalities to pyLIDAR-SLAM (Loop Closure, Motion Segmentation, Multi-Sensors, etc...).

News

[08/10/2021]: We also introduce support for individual rosbags (Introducing naturally an overhead compared to using ROS directly, but provides the flexibility of pyLiDAR-SLAM)

[08/10/2021]: We release code for Loop Closure with pyLiDAR-SLAM accompanied with a simple PoseGraph Optimization.

[08/10/2021]: We release our new work on arXiv. It proposes a new state-of-the-art pure LiDAR odometry implemented in C++ (check the project main page). python wrappings are available, and it can be used with pyLiDAR-SLAM.

Installation

See the wiki page INSTALLATION for instruction to install the code base and the modules you are interested in.

DATASETS

pyLIDAR-SLAM incorporates different datasets, see DATASETS for installation and setup instructions for each of these datasets. Only the datasets implemented in pyLIDAR-SLAM are compatible with hydra's mode and the scripts run.py and train.py.

But you can define your own datasets by extending the class DatasetLoader.

New: We support individual rosbags (without requiring a complete ROS installation). See the minimal example for more details.

A Minimal Example

Download a rosbag (e.g. From Rosbag Cartographer): example_rosbag

Note: You need the rosbag python module installed to run this example (see INSTALLATION for instructions)

Launch the SLAM:

python3 run.py num_workers=1 /          # The number of process workers to load the dataset (should be at most 1 for a rosbag)
    slam/initialization=NI /            # The initialization considered (NI=No Initialization / CV=Constant Velocity, etc...)
    slam/preprocessing=grid_sample /    # Preprocessing on the point clouds
    slam/odometry=icp_odometry /        # The Odometry algorithm
    slam.odometry.viz_debug=True /      # Whether to launch the visualization of the odometry
    slam/loop_closure=none /            # The loop closure algorithm selected (none by default)
    slam/backend=none /                 # The backend algorithm (none by default)
    dataset=rosbag /                    # The dataset selected (a simple rosbag here)
    dataset.main_topic=horizontal_laser_3d /    # The pointcloud topic of the rosbag 
    dataset.accumulate_scans=True /             # Whether to accumulate multiple messages (a sensor can return multiple scans lines or an accumulation of scans) 
    dataset.file_path=
   
    /b3-2016-04-05-15-51-36.bag / #  The path to the rosbag file 
    hydra.run.dir=.outputs/TEST_DOC   #  The log directory where the trajectory will be saved

   

This will output the trajectory, log files (including the full config) on disk at location .outputs/TEST_DOC.

Our minimal LiDAR Odometry, is actually a naïve baseline implementation, which is mostly designed and tested on driving datasets (see the KITTI benchmark). Thus in many cases it will fail, be imprecise or too slow.

We recommend you install the module pyct_icp from our recent work, which provides a much more versatile and precise LiDAR-Odometry.

See the wiki page INSTALLATION for more details on how to install the different modules. If you want to visualize in real time the quality of the SLAM, consider also installing the module pyviz3d.

Once pyct_icp is installed, you can modify the command line above:

python3 run.py num_workers=1 /          
    slam/initialization=NI /            
    slam/preprocessing=none /    
    slam/odometry=ct_icp_robust_shaky / # The CT-ICP algorithm for shaky robot sensor (here it is for a backpack) 
    slam.odometry.viz_debug=True /      
    slam/loop_closure=none /            
    slam/backend=none /                 
    dataset=rosbag /                    
    dataset.main_topic=horizontal_laser_3d /    
    dataset.accumulate_scans=True /             
    dataset.file_path=
   
    /b3-2016-04-05-15-51-36.bag / 
    hydra.run.dir=.outputs/TEST_DOC   

   

It will launch pyct_icp on the same rosbag (running much faster than our python based odometry)

With pyviz3d you should see the following reconstruction (obtained by a backpack mounting the stairs of a museum):

Minimal Example

More advanced examples / Motivation

pyLiDAR-SLAM will progressively include more and more modules, to build more powerful and more accessible LiDAR odometries.

For a more detailed / advanced usage of the toolbox please refer to our documentation in the wiki HOME.

The motivation behind the toolbox, is really to compare different modules, hydra is very useful for this purpose.

For example the script below launches consecutively the pyct_icp and icp_odometry odometries on the same datasets.

python3 run.py -m /             # We specify the -m option to tell hydra to perform a sweep (or grid search on the given arguments)
    num_workers=1 /          
    slam/initialization=NI /            
    slam/preprocessing=none /    
    slam/odometry=ct_icp_robust_shaky, icp_odometry /   # The two parameters of the grid search: two different odometries
    slam.odometry.viz_debug=True /      
    slam/loop_closure=none /            
    slam/backend=none /                 
    dataset=rosbag /                    
    dataset.main_topic=horizontal_laser_3d /    
    dataset.accumulate_scans=True /             
    dataset.file_path=
   
    /b3-2016-04-05-15-51-36.bag / 
    hydra.run.dir=.outputs/TEST_DOC   

   

Benchmarks

We use this functionality of pyLIDAR-SLAM to compare the performances of its different modules on different datasets. In Benchmark we present the results of pyLIDAR-SLAM on the most popular open-source datasets.

Note this work in still in construction, and we aim to improve it and make it more extensive in the future.

Research results

Small improvements will be regularly made to pyLiDAR-SLAM, However major changes / new modules will more likely be introduced along research articles (which we aim to integrate with this project in the future)

Please check RESEARCH to see the research papers associated to this work.

System Tested

OS CUDA pytorch python hydra
Ubuntu 18.04 10.2 1.7.1 3.8.8 1.0

Author

This is a work realised in the context of Pierre Dellenbach PhD thesis under supervision of Bastien Jacquet (Kitware), Jean-Emmanuel Deschaud & François Goulette (Mines ParisTech).

Cite

If you use this work for your research, consider citing:

@misc{dellenbach2021s,
      title={What's in My LiDAR Odometry Toolbox?},
      author={Pierre Dellenbach, 
      Jean-Emmanuel Deschaud, 
      Bastien Jacquet,
      François Goulette},
      year={2021},
}
Owner
Kitware, Inc.
Kitware develops software for web visualization, data storage, build system generation, infovis, media analysis, biomedical inquiry, cloud computing and more.
Kitware, Inc.
Pytorch Implementation of Interaction Networks for Learning about Objects, Relations and Physics

Interaction-Network-Pytorch Pytorch Implementraion of Interaction Networks for Learning about Objects, Relations and Physics. Interaction Network is a

117 Nov 05, 2022
Official Pytorch Implementation of: "Semantic Diversity Learning for Zero-Shot Multi-label Classification"(2021) paper

Semantic Diversity Learning for Zero-Shot Multi-label Classification Paper Official PyTorch Implementation Avi Ben-Cohen, Nadav Zamir, Emanuel Ben Bar

28 Aug 29, 2022
Adaptive Attention Span for Reinforcement Learning

Adaptive Transformers in RL Official implementation of Adaptive Transformers in RL In this work we replicate several results from Stabilizing Transfor

100 Nov 15, 2022
Based on Yolo's low-power, ultra-lightweight universal target detection algorithm, the parameter is only 250k, and the speed of the smart phone mobile terminal can reach ~300fps+

Based on Yolo's low-power, ultra-lightweight universal target detection algorithm, the parameter is only 250k, and the speed of the smart phone mobile terminal can reach ~300fps+

567 Dec 26, 2022
This is the winning solution of the Endocv-2021 grand challange.

Endocv2021-winner [Paper] This is the winning solution of the Endocv-2021 grand challange. Dependencies pytorch # tested with 1.7 and 1.8 torchvision

Vajira Thambawita 14 Dec 03, 2022
QMagFace: Simple and Accurate Quality-Aware Face Recognition

Quality-Aware Face Recognition 26.11.2021 start readme QMagFace: Simple and Accurate Quality-Aware Face Recognition Research Paper Implementation - To

Philipp Terhörst 59 Jan 04, 2023
Unit-Convertor - Unit Convertor Built With Python

Python Unit Converter This project can convert Weigth,length and ... units for y

Mahdis Esmaeelian 1 May 31, 2022
A script depending on VASP output for calculating Fermi-Softness.

Fermi softness calculation for Vienna Ab initio Simulation Package (VASP) Update 1.1.0: Big update: Rewrote the code. Use Bader atomic division instea

qslin 11 Nov 08, 2022
MVS2D: Efficient Multi-view Stereo via Attention-Driven 2D Convolutions

MVS2D: Efficient Multi-view Stereo via Attention-Driven 2D Convolutions Project Page | Paper If you find our work useful for your research, please con

96 Jan 04, 2023
Segmentation Training Pipeline

Segmentation Training Pipeline This package is a part of Musket ML framework. Reasons to use Segmentation Pipeline Segmentation Pipeline was developed

Musket ML 52 Dec 12, 2022
Info and sample codes for "NTU RGB+D Action Recognition Dataset"

"NTU RGB+D" Action Recognition Dataset "NTU RGB+D 120" Action Recognition Dataset "NTU RGB+D" is a large-scale dataset for human action recognition. I

Amir Shahroudy 578 Dec 30, 2022
Animation of solving the traveling salesman problem to optimality using mixed-integer programming and iteratively eliminating sub tours

tsp-streamlit Animation of solving the traveling salesman problem to optimality using mixed-integer programming and iteratively eliminating sub tours.

4 Nov 05, 2022
PySOT - SenseTime Research platform for single object tracking, implementing algorithms like SiamRPN and SiamMask.

PySOT is a software system designed by SenseTime Video Intelligence Research team. It implements state-of-the-art single object tracking algorit

STVIR 4.1k Dec 29, 2022
Extracts data from the database for a graph-node and stores it in parquet files

subgraph-extractor Extracts data from the database for a graph-node and stores it in parquet files Installation For developing, it's recommended to us

Cardstack 0 Jan 10, 2022
A U-Net combined with a variational auto-encoder that is able to learn conditional distributions over semantic segmentations.

Probabilistic U-Net + **Update** + An improved Model (the Hierarchical Probabilistic U-Net) + LIDC crops is now available. See below. Re-implementatio

Simon Kohl 498 Dec 26, 2022
Official implementations of PSENet, PAN and PAN++.

News (2021/11/03) Paddle implementation of PAN, see Paddle-PANet. Thanks @simplify23. (2021/04/08) PSENet and PAN are included in MMOCR. Introduction

395 Dec 14, 2022
PyTorch code for Vision Transformers training with the Self-Supervised learning method DINO

Self-Supervised Vision Transformers with DINO PyTorch implementation and pretrained models for DINO. For details, see Emerging Properties in Self-Supe

Facebook Research 4.2k Jan 03, 2023
Unifying Global-Local Representations in Salient Object Detection with Transformer

GLSTR (Global-Local Saliency Transformer) This is the official implementation of paper "Unifying Global-Local Representations in Salient Object Detect

11 Aug 24, 2022
An implementation of MobileFormer

MobileFormer An implementation of MobileFormer proposed by Yinpeng Chen, Xiyang Dai et al. Including [1] Mobile-Former proposed in:

slwang9353 62 Dec 28, 2022
Use MATLAB to simulate the signal and extract features. Use PyTorch to build and train deep network to do spectrum sensing.

Deep-Learning-based-Spectrum-Sensing Use MATLAB to simulate the signal and extract features. Use PyTorch to build and train deep network to do spectru

10 Dec 14, 2022