Datasetpublic

hotosm/fAIr-guru-td-buildings

HOT OAM Segmentation Dataset Semantic segmentation training data. Aerial imagery tiles from OpenAerialMap paired with OpenStreetMap labels, drawn from HOT Tasking Manager projects. The labelled feature is recorded in the feature_type column and in the generated statistics below. Structure One row per image tile. Column Description image 256x256 RGB aerial chip, georeferenced mask Binary raster, 0 background, 255 feature. Empty when the feature is not… See the full description on the dataset page: https://huggingface.co/datasets/hotosm/fAIr-guru-td-buildings.

sourceHugging Facecc-by-4.0updated 2mo agoView on Hugging Face

0likes10downloads

Dataset Card

HOT OAM Segmentation Dataset

Semantic segmentation training data. Aerial imagery tiles from OpenAerialMap paired with OpenStreetMap labels, drawn from HOT Tasking Manager projects. The labelled feature is recorded in the feature_type column and in the generated statistics below.

Structure

One row per image tile.

Column	Description
`image`	256x256 RGB aerial chip, georeferenced
`mask`	Binary raster, 0 background, 255 feature. Empty when the feature is not rasterized
`label_geojson`	Label geometries for the tile, EPSG:4326
`tile_id`, `tile_x`, `tile_y`, `tile_z`	XYZ tile identity
`bbox_west`, `bbox_south`, `bbox_east`, `bbox_north`	Tile bounds, EPSG:4326
`project_id`, `project_name`, `country`, `organisation`	Source Tasking Manager project
`imagery_url`	OpenAerialMap TMS endpoint the chip came from
`feature_type`	OSM tag key the labels were filtered on
`num_features`	Label count in the tile
`class_counts`	JSON of tag values in the tile, e.g. `{"residential": 3, "path": 1}`
`project_url`	Tasking Manager project page, including the mapping instructions given to mappers

Ground resolution depends on zoom level and latitude. At zoom 19 a pixel covers roughly 0.25 m and a tile spans about 63 m.

Labels

Labels come from OpenStreetMap through the HOT Raw Data API, filtered on the feature_type tag key. Geometry follows the feature: polygons for areas, lines for roads and waterways. Full OSM tags are preserved per feature under the tags property. Tiles containing no labels are kept as negative samples.

Label completeness follows OpenStreetMap coverage at extraction time. Imagery and labels can be captured at different dates, so a feature visible in a chip is not guaranteed to appear in the labels.

Splits

Splits are assigned per project, keeping tiles from one project together to avoid spatial leakage. Projects sharing an imagery URL cover the same ground and are assigned as one group.

Sources and licensing

Component	Source	License
Imagery	OpenAerialMap	CC-BY 4.0, attribute contributors of Open Imagery Network
Labels	OpenStreetMap	ODbL 1.0, attribute OpenStreetMap contributors
Project metadata	HOT Tasking Manager

Provenance

Built with tm-oam-ds-builder. Per-project metadata, areas of interest, and aggregate statistics ship alongside the Parquet.

Generated Stats

—Dataset version: 0.2.1
—Tile pairs (num_examples): 323
—Total projects: 1
—Feature type: building
—Tiles with labels: 321
—Tiles without labels: 2
—Total label features: 18,716
—Avg buildings per tile: 57.9
—Total area: ~2.9 sq km
—Countries: 1
—Generated at: 2026-08-05T23:41:43.504650+00:00

Label classes (OSM `building` tag)

Class	Features	Share
`yes`	18,654	99.7%
`school`	33	0.2%
`house`	21	0.1%
`tent`	3	0.0%
`construction`	2	0.0%
`carport`	1	0.0%
`parking`	1	0.0%
`hangar`	1	0.0%

Coverage by Country

Country	Projects	Tiles	Buildings	Area (sq km)
South Africa	1	323	18,716	2.9