Songdo Vision: High-Altitude BEV Drone Imagery for Vehicle Detection in a Smart City

Songdo Vision is a high-resolution aerial vehicle-detection dataset of 5,419 annotated 4K (3840 × 2160) bird's-eye view RGB drone frames. It contains approximately 272,000 vehicle instances across four classes (car, bus, truck, motorcycle), captured during a multi-drone urban traffic monitoring experiment over Songdo. The dataset was created by author rfonod and was last updated on 2026-06-25.

Use Cases

• Train vehicle detection models based on high-resolution 4K bird's-eye view drone imagery.
• Benchmark object detection algorithms for autonomous driving based on the four annotated vehicle classes (car, bus, truck, motorcycle).
• Study urban traffic patterns and density in a smart city context based on the ~272k annotated vehicle instances.
• Develop models for multi-drone cooperative perception systems based on data from a large-scale multi-drone experiment.

Strengths

• Contains 5,419 high-resolution 4K (3840 × 2160) annotated images.
• Includes approximately 272,000 annotated vehicle instances.
• Provides annotations for four distinct vehicle classes: car, bus, truck, and motorcycle.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Last updated 2026-06-25 16:33:13; freshness should be verified.

Provenance

Source

rfonod on Hugging Face

Collection Method

Captured during a large-scale multi-drone urban traffic monitoring experiment.

Freshness

2026-06-25 16:33:13

Geography

Songdo (likely Songdo International Business District, South Korea)