RarePlanes
In-Q-Tel CosmiQ Works
Satellite images of aircraft, mixing real and synthetic scenes, to test synthetic training data.
A machine learning dataset from CosmiQ Works and AI.Reverie combining real and synthetically generated satellite imagery to test how synthetic data helps detect and characterize aircraft, including rare types. Includes annotations for aircraft attributes.
Access
- Cost
- FreeNo charge to download; hosted in an AWS S3 bucket.
- License
- Open licenseCC BY-SA 4.0
- Sign-up
- No sign-up
- How you get it
- API and Bulk downloadsee documentation
What's in it
- One row is
- One image tile with aircraft annotations
- Covers
- Airports worldwide (real images); synthetic scenes
- Years
- Not stated
- Updated
- One-time release
- Size
- Not stated
Quality
Well-defined study of synthetic data usefulness. Narrow task with no further updates planned.
- Narrow subject
- Synthetic scenes differ from real
Connects to
- SpaceNet: Related satellite benchmark for buildings and roads
- Maxar Open Data Program: Commercial satellite imagery of events
Questions it could answer
- How much real data can synthetic data replace in training?
- Which aircraft types are most often missed?