DAT Datenverarbeitung, Informatik; GEO Geowissenschaften
Resource type:
Simulationen / simulations; Abbildungen von Objekten / image of objects
Data type:
Bilder / images
Description:
SegmentationUQ is a synthetic "error-free" benchmark dataset for semantic segmentation of building footprints and evaluation of uncertainty quantification methods. This proposed synthetic dataset not only contains labels of the building masks but also the reference aleatoric uncertainty given different noise level and noise type of the input images.
Method of data assessment:
We render the data from 3D building models and very high resolution aerial images of Berlin, Germany in the software Blender with the Cycles engine with rendering settings emulating real-world environmental conditions closely. Images with different noise levels, noise distributions, and viewing angles of the sensors were rendered.
Links:
Additional contact if questions arise: Wang, Yuanyuan y.wang@tum.de