alenaliu/auto-mos

Automatic Labeling to Generate Training Data for Online LiDAR-based Moving Object Segmentation

Auto-MOS: Automatic Labeling to Generate Training Data for Online LiDAR-based Moving Object Segmentation

This repo contains the code for our Auto-MOS, which automatically generates training data for LiDAR-based moving objects segmentation PDF.

Table of Contents

1. Introduction
2. Publication
3. Logs
4. Dependencies
5. How to use
6. Application
7. License

Publication

If you use our implementation in your academic work, please cite the corresponding paper (PDF):

@article{chen2022ral,
         author      = {X. Chen and B. Mersch and L. Nunes and R. Marcuzzi and I. Vizzo and J. Behley and C. Stachniss},
         title       = {{Automatic Labeling to Generate Training Data for Online LiDAR-Based Moving Object Segmentation}},
         journal     = {IEEE Robotics and Automation Letters (RA-L)},
         year        = 2022,
         volume      = 7,
         number      = 3,
         pages       = {6107-6114},
         url         = {http://arxiv.org/pdf/2201.04501},
         issn        = {2377-3766},
         doi         = {10.1109/LRA.2022.3166544}
        }

Logs

Version 1.0

Note that, due to copyright and protection of our benchmark, this repo currently only provides the tracking and label generating parts of the proposed method. For Odometry/LiDAR-SLAM we refer to our SuMa (link), refer dynamic removal to ERASOR (link), refer instance clustering to HDBSCAN (link), and refer the LiDAR-MOS network to our LMNet (link).

Dependencies

Before using our code, you need to install some libraries.

• System dependencies:

  sudo apt-get update 
  sudo apt-get install -y python3-pip wget unzip
  sudo -H pip3 install --upgrade pip

• Python dependencies (may also work with different versions than mentioned in the requirements file)

  sudo -H pip3 install -r requirements.txt

How to run

Download data and intermediate results

To run the quick demo, please first download the data (link) extracting it to the data folder, and the intermediate instance results (link) extracting it to the results folder.

To visualize the final results, you could also directly download the mos results (link) and extract it into the results folder.

You could also download the data and intermediate results using command lines as follows:

• Download kitti demo dataset:

  wget -P data/ https://www.ipb.uni-bonn.de/html/projects/auto-mos/kitti.zip
  unzip data/kitti.zip -d data
  rm data/kitti.zip

• Download instance predictions:

  wget -P results/ https://www.ipb.uni-bonn.de/html/projects/auto-mos/instances.zip
  unzip results/instances.zip -d results
  rm results/instances.zip

• Download final mos predictions:

  wget -P results/ https://www.ipb.uni-bonn.de/html/projects/auto-mos/mos_predictions.zip
  unzip results/mos_predictions.zip -d results
  rm results/mos_predictions.zip

Quick run

• To automatic generate the mos labels, one could directly run:

  python3 auto-mos-tracking.py

• To visualize the mos results, one could directly run:

  python3 vis_mos_results.py

  To control the visualizer:

  • press n: play next scan,
  • press b: play previous scan,
  • press esc or q: exits.

• To visualize the intermediate instance predictions, one could directly run:

  python3 vis_instances.py

  To control the visualizer:

  • press esc or q: exits.

License

This project is free software made available under the MIT License. For details see the LICENSE file.