An open library of aerial imagery labelled for anomaly detection.

Every scene here is a photograph taken from an aircraft or satellite by a special camera that records 100 – 400 colors per pixel — not just the red, green and blue a normal camera sees. Because different materials (asphalt, water, vegetation, metal) reflect each of those colors differently, an AI model can spot what does not belong in a scene — an aircraft on a runway, vehicles in a field, a building where the map shows only trees. That’s “anomaly detection.”

This dashboard is the ground truth researchers use to train and benchmark those models. Every one of the 655 scenes here has been sorted into an anomaly category, tagged with its source paper, and linked to the raw data on Google Drive. Click any scene to see its false-color preview, a labelled overlay showing where the unusual pixels sit, and the paper that published its ground truth.

Allotrope 21 Labs (2026). Hyperspectral Ground-Truth Corpus for Anomaly
Detection, v1.0. Curated by The Design Fusion.
Available at: https://groundtruthdatasets.vercel.app
Source: https://github.com/thedesignfusion/groundtruthdatasets