Benchmark for semantic façade segmentation on MLS point clouds

• MLS point clouds with facade-level labels (i.e., windows, doors, balconies, moldings, etc.)
• 33 annotated facades and 8 non-annotated for further benchmark extension or testing
• ~333 mln annotated points
• in local and global (i.e., UTM) coordinate reference system (if struggling with the height transformation - apply 45.66 m vertical offset)
• file names point to the official CityGML LoD2 building models of Bavaria (e.g., DEBY_LOD2_4959462 -> gml_id=DEBY_LOD2_4959462) [visualisation]
• settings file for adding your own data

• Annotate the outstanding building point clouds in the dataset, contact us (e.g., [email protected]), and extend the repository of the open façades!
• OR use our open-source-based approach and create a new open point cloud dataset!

In the linked paper you will find the comprehensive description of this dataset, how we achieved it, and the landscape of the point cloud benchmark datasets. If you use this dataset please cite either:

1. The paper [ISPRS] [preprint]

@article{wysockiTUMFACADE,
	author = {Wysocki, O. and Hoegner, L. and Stilla, U.},
	title = {TUM-FAÇADE: REVIEWING AND ENRICHING POINT CLOUD BENCHMARKS FOR FAÇADE SEGMENTATION},
	journal = {The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences},
	volume = {XLVI-2/W1-2022},
	year = {2022},
	pages = {529--536},
	url = {https://www.int-arch-photogramm-remote-sens-spatial-inf-sci.net/XLVI-2-W1-2022/529/2022/},
	doi = {10.5194/isprs-archives-XLVI-2-W1-2022-529-2022},

2. The dataset

@misc{mediatum1636761, 
	author = {Wysocki, Olaf  and Zhang, Jiarui  and Stilla, Uwe},
	title = {TUM-FAÇADE},
	publisher = {Technical University of Munich},
	url = {https://mediatum.ub.tum.de/1636761},
	type = {Dataset},
	year = {2021},
	doi = {10.14459/2021mp1636761.001},
	keywords = {MLS point clouds; semantic segmentation, 3D reconstruction, facade reconstruction},
	language = {en},
    

This work wouldn't be possible without Zhu et al. and their excellent TUM-MLS-2016 dataset. Thank you!
We are indebted to Jiarui Zhang and Yue Tan for their diligent work in the annotation process. Thank you Hitachi group for releasing the Semantic Segmentation Editor as a user-friendly open-source tool that we could easily adapt to our needs.

Last but not least, thank you Matrone et al. for the inspiring ArCH dataset! Go ahead, check it out, and test your algorithms there too!