Drone Datasets Categorization

By Primary Purpose

Object Detection Focused

• Kaggle Drone Object Detection Dataset: Specialized for amateur drone detection
• DroneDetectionDataset (GitHub): Focused on quadcopter UAV detection
• Roboflow Drone Datasets: Various datasets specifically for drone detection

Tracking Focused

• Multi-view Drone Tracking Datasets: Specialized for tracking with multiple camera views
• UAV123: Designed specifically for low-altitude UAV target tracking

Comprehensive (Detection & Tracking)

• VisDrone: Complete benchmark for detection, tracking, and counting
• UAVDT: Supports detection, single-object tracking, and multi-object tracking

By Dataset Size

Large (100,000+ images/frames)

• VisDrone: 261,908 frames + 10,209 static images
• UAV123: 113,476 images

Medium (10,000-100,000 images/frames)

• UAVDT: ~80,000 frames
• DroneDetectionDataset: 56,821 images (51,446 training + 5,375 test)

Small (<10,000 images/frames)

• Kaggle Drone Object Detection: ~4,000 images
• Multi-view Drone Tracking Datasets: 5 datasets with varying sizes (minutes of video each)

By Environment Type

Urban Environments

• UAVDT: Focused on urban areas (squares, streets, highways, crossings)
• VisDrone: Includes urban environments across 14 different cities

Natural/Rural Environments

• VisDrone: Includes country environments
• UAV123: Includes various outdoor environments

Mixed Environments

• DroneDetectionDataset: Various environments from YouTube videos
• Multi-view Drone Tracking: Includes winter/snow conditions in dataset 5

By Annotation Type

Bounding Box Annotations

• VisDrone: 2.6+ million bounding boxes
• UAVDT: Bounding boxes with vehicle category and occlusion
• UAV123: Bounding box annotations for 10 different object classes
• DroneDetectionDataset: XML labels for bounding boxes

Trajectory Annotations

• Multi-view Drone Tracking: 3D trajectory ground truth from RTK system
• UAVDT: Trajectory information for tracking tasks
• VisDrone: Trajectory information for tracking tasks

Additional Metadata

• VisDrone: Scene visibility, object class, occlusion
• UAVDT: Vehicle category, occlusion information
• Multi-view Drone Tracking: Camera calibration, synchronization data

By Object Classes

Drone-Only Detection

• Kaggle Drone Object Detection: Focused only on drone detection
• DroneDetectionDataset: Specialized for drone detection
• Roboflow Drone Datasets: Primarily for drone detection

Vehicle-Focused

• UAVDT: Primarily focused on vehicles in urban environments

Multi-Class

• VisDrone: Pedestrians, cars, bicycles, tricycles, etc.
• UAV123: 10 classes including person, car, group, boat, uav, bike, etc.

By Technical Challenges

Occlusion Handling

• UAVDT: Includes occlusion information
• UAV123: Includes partial occlusion (73,677 images) and full occlusion (30,736 images)
• VisDrone: Includes occlusion attributes

Scale Variation

• UAV123: Scale variation (100,919 images)
• VisDrone: Various altitudes and perspectives

Lighting/Weather Conditions

• VisDrone: Various weather and lighting conditions
• UAV123: Illumination variation (32,803 images)
• Multi-view Drone Tracking: Winter/snow conditions in dataset 5

Camera Motion

• UAV123: Camera motion (75,025 images)
• Multi-view Drone Tracking: Multiple camera perspectives

By Accessibility and Licensing

Open License

• Kaggle Drone Object Detection: CC BY 4.0
• DroneDetectionDataset: MIT License for image tags

Research-Only

• UAVDT: Custom (research-only)

Not Clearly Specified

• VisDrone: License not clearly specified
• UAV123: License not clearly specified
• Multi-view Drone Tracking: License not clearly specified
• Roboflow Drone Datasets: Various licenses depending on dataset