This large-scale benchmark dataset was created using topographic maps and high-resolution 2D and 3D airborne data from the Netherlands, acquired in 2022. It consists of 2,413 spatially matching and non-overlapping tiles, including maps, 2D true orthophotos, digital surface models (DSMs), and 3D point clouds. The dataset covers approximately 217 square kilometers and represents diverse landscapes, including urban, suburban, industrial, rural, and forested areas.

The dataset provides (i) per-pixel and per-point semantic labels for 20 classes, supporting both 2D and 3D semantic segmentation tasks, and (ii) vector polygon annotations for object delineation tasks. The data is divided into two groups to support reproducible benchmarking of deep learning models.

Further details on the dataset, labeling methodology, and benchmark results are available in the associated publication: https://doi.org/10.1016/j.ophoto.2025.100112