UAV123 Dataset

Overview

UAV123 is a benchmark dataset specifically designed for visual object tracking from low-altitude UAVs. It contains 123 video sequences with more than 110,000 frames.

Key Statistics

• 123 video sequences (113,476 frames)
• 10 different object classes
• Average sequence length: 915 frames
• Resolution: 1280×720 pixels
• Frame rate: 30 FPS

Dataset Attributes

• Scale Variation: 100,919 images
• Partial Occlusion: 73,677 images
• Full Occlusion: 30,736 images
• Illumination Variation: 32,803 images
• Camera Motion: 75,025 images

Object Classes

• Person
• Car
• Group (of people)
• Boat
• UAV
• Bike
• Truck
• Building
• Animal
• Other

Strengths

• Long sequences for persistent tracking
• Diverse tracking scenarios
• Low-altitude perspective
• High-quality annotations

Limitations

• Single-object tracking only
• Limited object diversity
• No multi-object scenarios
• Less suitable for detection tasks

Source

Dataset Page: https://datasetninja.com/uav123