DroneVehicle-Night: A Nighttime UAV Aerial Dataset for Vehicle Detection

The DroneVehicle-Night dataset contains 10,357 pairs of training images, 868 pairs of validation images, and 6,013 pairs of test images, with each pair including both RGB and IR modalities. In total, the dataset provides 276,819 annotated object instances across five vehicle categories: car, truck, bus, van, and freight vehicle. The samples are collected from representative nighttime aerial scenes—such as urban roads, residential areas, and parking lots—capturing the complexity of illumination interference and the diversity of spatial distributions.To achieve more accurate annotation, the dataset adopts the Oriented Bounding Box (OBB) format, which better describes vehicle shapes under oblique viewing angles. Compared with traditional Horizontal Bounding Boxes (HBB), OBB annotations are more suitable for UAV-based aerial imagery, where object orientations vary significantly and inter-object scale differences are prominent.