prbonn/mapmos

Building Volumetric Beliefs for Dynamic Environments Exploiting Map-Based Moving Object Segmentation (RAL 2023)

Building Volumetric Beliefs for Dynamic Environments Exploiting Map-Based Moving Object Segmentation

Our approach identifies moving objects in the current scan (blue points) and the local map (black points) of the environment and maintains a volumetric belief map representing the dynamic environment.

Click here for qualitative results!

mapmos.mp4

Our predictions for the KITTI Tracking sequence 19 with true positives (green), false positives (red), and false negatives (blue).

Installation

First, make sure the MinkowskiEngine is installed on your system, see here for more details.

Next, clone our repository

git clone [email protected]:PRBonn/MapMOS && cd MapMOS

and install with

make install

or

make install-all

if you want to install the project with all optional dependencies (needed for the visualizer). In case you want to edit the Python code, install in editable mode:

make editable

How to Use It

Just type

mapmos_pipeline --help

to see how to run MapMOS.

This is what you should see

Check the Download section for a pre-trained model. Like KISS-ICP, our pipeline runs on a variety of point cloud data formats like bin, pcd, ply, xyz, rosbags, and more. To visualize these, just type

mapmos_pipeline --visualize /path/to/weights.ckpt /path/to/data

Want to evaluate with ground truth labels?

Because these lables come in all shapes, you need to specify a dataloader. This is currently available for SemanticKITTI and NuScenes as well as our post-processed KITTI Tracking sequence 19 and Apollo sequences (see Downloads).

Want to reproduce the results from the paper?

For reproducing the results of the paper, you need to pass the corresponding config file. They will make sure that the de-skewing option and the maximum range are set properly. To compare different map fusion strategies from our paper, just pass the `--paper` flag to the `mapmos_pipeline`.

Training

To train our approach, you need to first cache your data. To see how to do that, just cd into the MapMOS repository and type

python3 scripts/precache.py --help

After this, you can run the training script. Again, --help shows you how:

python3 scripts/train.py --help

Want to verify the cached data?

You can inspect the cached training samples by using the script python3 scripts/cache_to_ply.py --help.

Want to change the logging directory?

The training log and checkpoints will be saved by default to the current working directory. To change that, export the export LOGS=/your/path/to/logs environment variable before running the training script.

Downloads

You can download the post-processed and labeled Apollo dataset and KITTI Tracking sequence 19 from our website.

The weights of our pre-trained model can be downloaded as well.

Publication

If you use our code in your academic work, please cite the corresponding paper:

@article{mersch2023ral,
  author = {B. Mersch and T. Guadagnino and X. Chen and I. Vizzo and J. Behley and C. Stachniss},
  title = {{Building Volumetric Beliefs for Dynamic Environments Exploiting Map-Based Moving Object Segmentation}},
  journal = {IEEE Robotics and Automation Letters (RA-L)},
  volume = {8},
  number = {8},
  pages = {5180--5187},
  year = {2023},
  issn = {2377-3766},
  doi = {10.1109/LRA.2023.3292583},
  codeurl = {https://github.com/PRBonn/MapMOS},
}

Acknowledgments

This implementation is heavily inspired by KISS-ICP.

License

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