Import/Catalogue/GlobalMLBuilding
About
This page is about importing building footprints into planet.osm. Dataset is available with Open Data Commons Open Database License. Since several actions shall be done on candidate data, it shall not be blind import.
This wiki is an attempt to define a method to achieve a quality close to manual tracing, hence areas involved should be up to mappers. Priority shall be for poor or no building OSM coverage.
Dataset
Global coverage areas and country geojson's are linked in GlobalMLBuildingFootprints Github repository. Authors declare "777M buildings from Bing Maps imagery between 2014 and 2021 including Maxar and Airbus imagery". As an example, italian coverage is patchy, but can potentially fill many uncovered areas.
Quality
Tested areas (Contessa-Italy, Tricase-Italy, Lefkada-Greece, Preveza-Greece, Mullaloo-Australia) feature different density and orography. Potential issues are similiar and can be listed as:
- general offset, due to source imagery
- overlapping, due to applied Microsoft AI
- false shapes, due to parallax (imagery capture angle)
- polygon orientation (due to applied Microsoft AI squaring on retained polygon vertices)
Legal
Source license is Open Data Commons Open Database License (ODbL), as defined in the above dataset Github repository.
Work pieces
Since data files involved can be from small to huge, data shall be splitted in manageable pieces. Depending on building density and anthropisation, most of cases may be easiliy managed on a municipality (admin_level=8) base.
If there is participation in the project, the activation of a task manager will be considered.
Filtering data
From input dataset, buildings already present in OSM ("building" and "demolished:building") shall be excluded from import. Accomplished by QGIS (Grass function v.select disjoint). Optionally, some pruning may be done on area, ie: geometries smaller than 40-50 square meters.
Tagging plan
Resulting filtered geometries will be tagged building=yes.
Conflating
Since input dataset does not feature specific building tags, no conflation will be performed. See the above Filtering paragraph.
Final check
Before upload, an overall check on import candidates shall be done with available tools (JOSM, umap, iD etc.) comparing geojson and available aerial imagery.
Workflow
Prior to JOSM visual check, reference imagery (Bing) shall be aligned to Strava heat map.
- import source geojson in JOSM
- general building=yes tagging
- validator for overlapping buildings fix
- parallax check for high buildings (typically in downtown areas)
- highway download in layer
- validator for building/highway crossings
Changeset tags
The following tags will be added to changeset:
- "source"="Microsoft GlobalMLBuildingFootprints"
- "source:url"="https://github.com/microsoft/GlobalMLBuildingFootprints"
- "url"="https://wiki.openstreetmap.org/wiki/Import/Catalogue/GlobalMLBuilding"
- "type"="import"
- "comment"="supervised import of building footprint"
Changeset author
Import specific author GlobalMLBuilding_import has been created.
Files
Please, find in Github ItalyBuildingImport repository cantidates osm files for import.
Changesets
| osm candidate | changeset | date |
|---|---|---|
| Contessa Entellina | 121366463 | 2022-05-23 |
| Cattolica Eraclea | 121414090 | 2022-05-24 |
External discussions
- Imports mailing list: received comments in https://lists.openstreetmap.org/pipermail/imports/2022-July/thread.html https://lists.openstreetmap.org/pipermail/imports/2022-August/thread.html https://lists.openstreetmap.org/pipermail/imports/2022-September/thread.html with opposition to import in the proposed form