Earth Parser Dataset: A new dataset to train and evaluate parsing methods on large, uncurated aerial LiDAR scans

We introduce a new dataset to train and evaluate parsing methods on large, uncurated aerial LiDAR scans. We use data from the French Mapping Agency associated to the LiDAR-HD project. We selected 7 scenes, covering over 7.7km2 and a total of 98 million 3D points, with diverse content and complexity, such as dense habitations, forests, or complex industrial facilities.

You can download sequences individually or use zenodo-get to download all sequences at once:

pip install zenodo-get

zenodo-get 7820686

See companion github repository and the dedicated wepage for more information.

Cite as:

@misc{loiseau2023learnable,
      title={Learnable Earth Parser: Discovering 3D Prototypes in Aerial Scans}, 
      author={Romain Loiseau and Elliot Vincent and Mathieu Aubry and Loic Landrieu},
      year={2023},
      eprint={2304.09704},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}

Acknowledgements :

• This work was supported by ANR project READY3D ANR-19-CE23-0007.
• The work of MA was partly supported by the European Research Council (ERC project DISCOVER, number 101076028).
• The scenes of the Earth Parser Dataset were acquired and annotated by the LiDAR-HD project.
• We thank Zenodo for hosting the dataset.