DeKH — German Hospital Dataset

DeKH (German Hospital Dataset) is a multi-scene dataset of real hospital environments comprising high-resolution 3D point clouds with semantic annotations and ground-truth IFC BIM models. It is introduced alongside BIMStruct3D, a fully automated hybrid Scan-to-BIM pipeline accepted at EC3 2026.

---

Dataset Contents

The dataset covers four distinct hospital scenes across three buildings:

Buildings/
├── A/
│   ├── 1st_floor/
│   │   ├── DeKH_A_1st_floor.laz
│   │   └── DeKH_A_1st_floor.npy
│   ├── 2nd_floor/
│   │   ├── DeKH_A_2nd_floor.laz
│   │   └── DeKH_A_2nd_floor.npy
│   └── DeKH_A.ifc
├── B/
│   ├── DeKH_B_ICU.laz
│   ├── DeKH_B_ICU.npy
│   └── DeKH_B_ICU.ifc
└── C/
    ├── DeKH_C_surgery.laz
    ├── DeKH_C_surgery.npy
    └── DeKH_C_surgery.ifc

File Formats

• .laz — Compressed point cloud (LAS/LAZ format). Each file contains the raw 3D point positions of the scanned environment.
• .npy — NumPy array of per-point semantic labels aligned to the corresponding .laz point cloud (float32, shape (N,)).
• .ifc — Ground-truth Building Information Model in the IFC standard, usable as reference for Scan-to-BIM evaluation.

---

Requirements

To load the point clouds and labels you need:

• laspy with LAZ support (via lazrs or laszip backend):

  pip install laspy[lazrs]
  

  See the laspy installation guide for details on LAZ backends.

• NumPy:

  pip install numpy
  

Quick Start

import laspy
import numpy as np

# Load point cloud
las = laspy.read("Buildings/B/DeKH_B_ICU.laz")
points = np.vstack([las.x, las.y, las.z]).T  # (N, 3)

# Load semantic labels
labels = np.load("Buildings/B/DeKH_B_ICU.npy")  # (N,) float32

---

Annotation

Semantic labels were produced following the ontology and labeling methodology introduced in:

F. Kaufmann, M. Chamseddine, S. Guttikonda, C. Glock, D. Stricker, J. Rambach, "Ontology-Based Semantic Labeling for RGB-D and Point Cloud Datasets", EC3 2023.

---

License

This dataset is released under CC BY-NC-SA 4.0 (Attribution-NonCommercial-ShareAlike). It may not be used for commercial purposes.

---

Citing this Work

If you use DeKH in your research, please cite:

@article{chamseddine2026bimstruct3d,
    title   = {BIMStruct3D: A Fully Automated Hybrid Learning Scan-to-BIM Pipeline with Integrated Topology Refinement},
    author  = {Chamseddine, Mahdi and Kaufmann, Fabian and Schellen, Marius and Glock, Christian and Stricker, Didier and Rambach, Jason},
    journal = {arXiv preprint arXiv:2604.24311},
    year    = {2026}
}

---

Acknowledgement

This research was funded by the European Union as part of the projects: HumanTech (Grant Agreement 101058236) and ShieldBOT (Grant Agreement 101235093).