STPLS3D stands for Semantic Terrain Points Labeling – Synthetic 3D, which aims to provide a large-scale aerial measurement dataset (including synthetic & real annotated 3D point clouds) for semantic segmentation and instance segmentation tasks.

The dataset contains:

•  Point cloud of 1.27 square kilometers of landscape.
• 62 synthetic point clouds with different building styles, vegetation types and terrain shapes

The synthetic dataset covers about 16 square kilometers of urban landscape, with up to 18 fine-grained semantic classes and 14 instance classes.

Cornell University released this dataset.