VIDA Open Buildings: Romania

source.coop/vida/google-microsoft-osm-open-buildings

The footprint of every building in Romania, from a global layer that merges three sources: Google's Open Buildings V3 (detected from satellite imagery), Microsoft's GlobalMLBuildingFootprints (also imagery-detected) and OpenStreetMap (drawn by contributors). VIDA deduplicates and merges the three into one 2.7-billion-footprint global layer published on Source Cooperative; this dataset is that layer's Romanian slice, verified directly: 12,655,928 footprints, of which 12,491,968 (98.71%) are Microsoft-sourced, 163,960 (1.30%) are OpenStreetMap-sourced, and 0 are Google-sourced — Romania was not covered by Google's Open Buildings detection run. Every row carries its exact source (bf_source), because the three upstream sources carry different licences: OpenStreetMap is ODbL with a share-alike obligation, Microsoft is CDLA Permissive 2.0, and Google is CC BY 4.0. VIDA republishes the merged layer under a single licence, ODbL v1.0 — the strictest of the three — and this dataset does the same. The source has no stable per-building id; geometry is published as WKT polygons in WGS 84.

referenced liveRows

1Tables

—Size

last monthLast ingest

ODbL-1.0Open Data Commons Open Database License v1.0

Attribution, terms and provenanceWhat you must carry if you reuse this

Attribution: Building footprints © VIDA (google-microsoft-osm-open-buildings dataset), merging Google Open Buildings V3 (CC BY 4.0), Microsoft GlobalMLBuildingFootprints (CDLA Permissive 2.0) and © OpenStreetMap contributors (ODbL) — Open Data Commons Open Database License (ODbL) v1.0.
Notes: VIDA's own "Dataset Licenses" statement, read directly on the dataset's Source Cooperative page (https://source.coop/vida/google-microsoft-osm-open-buildings) and in the bucket's own README.md, says plainly: "The data is shared under the Open Data Commons Open Database License (ODbL) v1.0 license." That is the licence this connector republishes under, in full. IMPORTANT CONTEXT for a reuser. (1) This is a MERGE of three differently-licensed sources, each still identifiable per row via bf_source: OpenStreetMap (https://www.openstreetmap.org/copyright) is ODbL-1.0, which requires attribution and "share-alike" — a database produced from it must itself be offered under ODbL or a compatible licence; Microsoft's GlobalMLBuildingFootprints (https://github.com/microsoft/GlobalMLBuildingFootprints) is CDLA Permissive 2.0, which permits redistribution and modification with attribution and no share-alike obligation; Google's Open Buildings V3 (https://sites.research.google/open-buildings/) is CC BY 4.0, attribution required, no share-alike. VIDA's choice to publish the WHOLE merged product under ODbL v1.0 — the strictest of the three — is the safe, conservative position for a combined database that includes ODbL-licensed OSM data, and this dataset follows VIDA's own published position rather than trying to relicense a subset. (2) Romania's current extract carries no Google-sourced rows (bf_source is never "google" here), so in practice every row in this table today is Microsoft- (CDLA Permissive 2.0) or OSM-sourced (ODbL). A reuser who specifically needs a licence more permissive than ODbL can filter to bf_source <> 'osm' and rely on CDLA Permissive 2.0 for what remains — CDLA Permissive itself does not carry a share-alike clause — but the WHOLE published table, mixed, is offered under ODbL v1.0 as VIDA licenses it. (3) A "building" here means a detected or mapped footprint polygon, not an inspected structure: confidence is only ever populated for Google-sourced rows (none in Romania today), and Microsoft- and OSM-sourced footprints carry no model confidence at all. (4) What is republished is the layer's own GeoParquet columns for Romania, reshaped: the footprint geometry as WKT, the source, the area, the confidence where one exists, and a bounding box; VIDA's boundary_id and s2_id partition keys are dropped because they are constant across the whole Romanian extract and carry no information here.
Version: snapshot 2026-09-04 · 01M1NQT7YTFY5YJ067PWBKZP89
Machine-readable:: llms.txt manifest.json

Tables

Open a table for its columns

Loading the schema…

Queries

Copy a snippet, or run it in the playground

Query it yourselfConnection snippets for DuckDB — Parquet or the live lake

Generated from this exact manifest version, so the URLs are the ones actually published.

Attach the catalog, then query it

INSTALL spatial;

LOAD spatial;

ATTACH IF NOT EXISTS ':memory:' AS "vida_buildings_ro";

CREATE SCHEMA IF NOT EXISTS "vida_buildings_ro"."main";

CREATE OR REPLACE VIEW "vida_buildings_ro"."main"."buildings" AS SELECT
  row_number() OVER (
    ORDER BY geohash, bbox.xmin, bbox.ymin, bbox.ymax, bbox.xmax
  )                                AS building_id,
  bf_source                        AS bf_source,
  confidence                       AS confidence,
  area_in_meters                   AS area_in_meters,
  geohash                          AS geohash,
  bbox.xmin                        AS bbox_min_longitude,
  bbox.ymin                        AS bbox_min_latitude,
  bbox.xmax                        AS bbox_max_longitude,
  bbox.ymax                        AS bbox_max_latitude,
  ST_AsText(CAST(geometry AS GEOMETRY)) AS geometry_wkt,
  TIMESTAMPTZ '2026-09-04T08:15:57.402Z'                       AS ingested_at
FROM read_parquet('https://data.source.coop/vida/google-microsoft-osm-open-buildings/geoparquet/by_country/country_iso=ROU/ROU.parquet');

SELECT *
FROM vida_buildings_ro.main.buildings
LIMIT 10;

Execute