* This dataset supports online use.Click here to jump.

UAVDT stands for Unmanned Aerial Vehicle Benchmark Object Detection and Tracking. It is a large-scale video dataset for drone target detection and tracking. It contains 10 hours of raw video and about 8,000 representative video frames with manually annotated bounding boxes and some useful labels such as vehicle categories and occlusions. The dataset is captured by drones in various complex scenes and is mainly used to perform three basic tasks: target detection (DET), single target tracking (SOT) and multiple target tracking (MOT).

The benchmark of this dataset consists of 100 video sequences, which are selected from more than 10 hours of videos taken by drones at multiple locations in urban areas, representing various common scenes, including squares, main streets, toll booths, highways, intersections, and T-junctions. The more prominent target objects in this benchmark are vehicles. The videos are recorded at 30 frames per second (fps) and the JPEG image resolution is 1080 × 540 pixels.