Mariupol Damage Identification

Public dataset identifying damage sustained in Mariupol as of March 21st, 2022 on a by-structure level. This dataset aims to support the mission-driven AI community across government, humanitarian, and academic sectors.


Mariupol Damage Identification Dataset

Developed with commercial satellite imagery, this dataset supports structure and damage identification use cases. This data is composed of:


Sq. KM Scene of Mariupol


Levels of damage assessment


structures in total


Satellite Constellation Imagery

The image used is an electro-optical multispectral satellite image, with a GSD of approximately 50cm and spherical error of 10 meters.

Map loader img


Identifying And Assessing Damage

Scale annotates large areas through best in class human-machine teaming. After imagery ingestion, machine learning data curation separates out areas without structures. The remaining areas are annotated by human labelers using smart annotation tools, programmatic checks and quality sampling to meet statistical confidence goals.

Polygon Labeled

In this Mariupol dataset, each structure is labeled with a polygon. Because of the difficulty of placing precise polygons in dense urban areas, Scale created an auto-segment polygon tool. This tool allows labelers to place a bounding box over an area with a building and leverages machine learning to transform that bounding box into a precise polygon covering the building’s rooftop.


All buildings receive a damage assessment on a three-point scale:

  • Undamaged building

    No Damage

    No visible holes or cracks in the roof

  • Damaged building


    Roofs with holes or partially collapsed

  • Destroyed building


    Roof is completely collapsed


Get Started with Mariupol Dataset
Ukraine dataset cover