Synthetic MBES Dataset for Underwater 3D Object Detection
[research data]
Shaukat, Muhammad Salman
VerfasserIn
aut
software, multimedia
Computerdaten
Forschungsdaten
xx
2026
monographic
Rostock
University
publisher
2026
Rostock
University Library
distributor
eng
1 Online-Ressource
This dataset contains 100 synthetically generated multi-beam echosounder (MBES) 3D point clouds for the development and evaluation of underwater 3D object detection methods. The simulated environments were inspired by the artificial reef at the Digital Ocean Lab (DOL) near Nienhagen, Germany, and consist of procedurally generated seabeds populated with four types of artificial reef structures: small tetrapods, large tetrapods, reef cones, and concrete rings. The dataset contains a total of 5,491 annotated object instances. Each point cloud is accompanied by ground-truth 3D bounding-box annotations specifying the object class, dimensions, position, and orientation. The dataset was generated using a physics-based simulation pipeline and is intended to support research on 3D object detection from sonar point clouds, particularly in scenarios where annotated real-world sonar data are scarce.
Open Access
Namensnennung 4.0 International
M. Salman Shaukat
This research was supported by the BMBF, OTC-Digital Twin & Analytics (OTC-DaTA) project (03ZU1107FB)
004
http://purl.uni-rostock.de/rosdok/id00005050
10.1109/JOE.2026.3675841
M. S. Shaukat, Y. Käckenmeister, S. Bader and T. Kirste, "Toward Training-Free Underwater 3-D Object Detection From Sonar Point Clouds: A Comparison of Traditional and Deep Learning Approaches," in IEEE Journal of Oceanic Engineering, vol. 51, no. 3, pp. 2284-2297, July 2026.
10.18453/rosdok_id00005050
1613390107
rda
DE-627
260827
1982579714
20260827T120106.0
Converted from MARCXML to MODS version 3.8 using MARC21slim2MODS3-8_XSLT1-0.xsl
(Revision 1.174 20250328)
ger