This dataset supports detecting small aerial objects in surveillance-style CCTV imagery. It includes labeled images for five classes: propeller drones, fixed-wing drones, helicopters, birds, and planes, to reduce confusion between drones and other objects.
The dataset is about 12 GB with roughly 15200 images. It’s organized into Train, Dev, and Test splits, then REAL and AI subfolders, and finally per-class folders. Annotations are Pascal VOC-style… See the full description on the dataset page:
https://huggingface.co/datasets/ratio1/Drone_Detection_Survaillance.
