40 high-resolution satellite video sequences containing 167,488 annotated instances across five categories including airplane, ship, car, train, and van. The dataset focuses on the technical challenges of identifying and monitoring small, dense moving targets within complex orbital imagery.
Use Cases
- Train multi-object tracking (MOT) models using the instance IDs and bounding box coordinates provided for each video frame
- Develop specialized small-object detection algorithms using the category labels for targets with minimal pixel footprints
- Benchmark motion-based object detection techniques against the temporal sequences of satellite imagery
Strengths
- 40 satellite video sequences captured from the Jilin-1 satellite constellation
- 167,488 annotated object instances across five distinct classes: airplane, ship, car, train, and van
- High-resolution frames cropped into sub-images to facilitate the detection of dense object clusters
- Temporal consistency data for tracking moving objects across consecutive video frames