Al Haouz UAV Building Damage Dataset: annotated low-altitude imagery of three rural villages affected by the 2023 Mw 6.8 Al Haouz earthquake, Morocco

This dataset contains 963 annotated tiles of low-altitude uncrewed aerialvehicle imagery covering three rural villages in the High Atlas affected bythe 8 September 2023 Mw 6.8 Al Haouz earthquake, Morocco: Douar Ighil (437tiles), Talat N'Yaakoub (283 tiles) and Tassaount (243 tiles). Each tile ispaired with a single-channel mask giving a pixel-level building footprint andone of four ordinal damage grades aligned with EMS-98 damage grades DG1 to DG5. Frames are processed independently, without orthomosaic generation, and tiled.A tile is retained only if its annotation contains at least 200 buildingpixels. Building outlines were verified against a textured three-dimensionalmesh reconstructed from the same survey imagery. The deposit includes the split definition files that reproduce theleave-one-village-out cross-validation protocol and the grouped validationsplit used in the associated publication, together with a per-tile indexgiving dimensions and per-class pixel counts. The grouping unit is the sourceframe for Douar Ighil and the image for the other two villages, so no sourceframe contributes tiles to more than one split. See README.md for the full file layout, the mask value mapping, the folddefinitions and the resulting split sizes.