100 video sequences containing 140,000 frames and 1.21 million manually annotated bounding boxes of vehicles. The dataset covers four vehicle categories—car, bus, van, and others—captured across varying weather, scale, and occlusion levels.
Use Cases
- Train vehicle detection models using the bounding box coordinates and category labels
- Develop multi-object tracking (MOT) algorithms using the frame-to-frame identity associations
- Analyze model performance across environmental factors using the weather and occlusion attributes
Strengths
- 1.21 million manually annotated bounding boxes across 140,000 video frames
- Environmental metadata including weather (sunny, cloudy, rainy, night) and illumination
- Four vehicle categories: car, bus, van, and others
- 100 video sequences recorded at 24 frames per second