UT-LUMPI

UCF UrbanTwin (UT) dataverse is a collection of high-fidelity synthetic roadside lidar datasets. Each dataset in this dataverse supports core perception tasks in intelligent transportation systems, including 3D object detection, multi-object tracking, semantic segmentation, and instance segmentation. Generated through precise simulation of real-world intersections and sensor settings, these datasets exhibit minimal sim-to-real domain gap and enable the training of deep learning models that perform competitively on real data benchmarks. The LUMPI dataset is a large-scale roadside lidar dataset captured from Königsworther Platz intersection, located in Hannover, Germany. It contains a 6 classes, that are, person, car, bicycle, motorcycle, bus, and truck. An unknown class is also added for non-background miscellaneous objects. This synthetic counterpart, UT-LUMPI, is constructed to mimic LUMPI’s spatial layout and sensor specifications. The dataset contains ~220K points and 8 road-user objects per frame in an 80x80x10 meter^3 region roughly centered at the intersection area, all matching closely to the original dataset.