Research data

Datasets

Flood imagery for disaster-management models and ready-to-map tourism shapefiles. Add items to your cart and pay by card, bank transfer or IRIS.

Machine learning datasets

Deucalion v1.0-anonymized

All 10,240 photos of the two classes, as per the official publication (not the preprint), anonymized.

Delivered within 1 day

1000€

or order by e-mail

Pyrrha v1.0-anonymized

All 2,004 photos and 11,393 masks of the 20 classes, as per the official publication (not the preprint), anonymized. YOLO-ready (ultralytics) for “flooded” and “people”.

Delivered within 1 day

2500€

or order by e-mail

Deucalion v1.0

10,240 photos in two classes, as per the official publication. Not anonymized.

On requestAsk

Pyrrha v1.0

2,004 photos and 11,393 segments in 20 classes, as per the official publication. Not anonymized.

On requestAsk

The original (non-anonymized) versions are normally not for sale — feel free to ask if you need them.

Deucalion v1.0-anonymized

Updated 22 July 2026. Delivered within 1 day, or within a day of being at the storage location.
All photos anonymized with GIMP's smudge tool and a pen pad — the only difference from the official Deucalion v1.0.
Cross-checks found minimal errors (<0.5%). For consistency they are not revised but annotated in a CSV; filter them to obtain Deucalion v1.0-anonymized-sp1.
Find another, unlisted mistake and receive a $50 refund per image, plus a free future v1.0-anonymized-sp2. For published research, please don't correct the dataset alone.
For operational use, modifying the dataset is probably low-risk, but not recommended.

Pyrrha v1.0-anonymized

Updated 3 August 2026. Delivered within 1 day.
2,004 photos split into train (1,804) and test (200) as per the official paper, plus all 11,393 segments.
YOLO format, ready to train with ultralytics models for “flooded” and “people”; all other class names and segments included.
Need a different template? Mention it in your order email — usually delivered free within a few days.

Mapping datasets