Low Altitude Disaster Imagery (LADI) Dataset
MIT Lincoln Laboratory Humanitarian Assistance and Disaster Relief Systems Group
Aerial photos taken by Civil Air Patrol volunteers after U.S. disasters, labeled for damage and features.
Human and machine annotated airborne images collected by the Civil Air Patrol during disaster responses from 2015 to 2023. The low-altitude, oblique view and disaster-specific labels, such as flooding and damage, are rare in computer vision benchmarks. Built for training disaster assessment models.
Access
- Cost
- FreeNo charge to download; hosted in an AWS S3 bucket.
- License
- Open licenseCC BY 4.0
- Sign-up
- No sign-up
- How you get it
- API and Bulk downloadJPEG, CSV
What's in it
- One row is
- One aerial photograph with its labels
- Covers
- United States disaster areas
- Years
- 2015-2023
- Updated
- Irregular
- Size
- Not stated
Quality
Unusual domain-specific benchmark with documented annotation process. Label coverage is uneven across disaster types.
- Mix of human and machine labels
- Coverage limited to events with CAP flights
Connects to
- Maxar Open Data Program: Satellite-view pre and post event imagery of the same kinds of disasters
- OpenAerialMap on AWS: Community aerial imagery used in humanitarian mapping
Questions it could answer
- Can a model trained on LADI spot flood damage in new events?
- Which disaster labels are hardest to classify?