beijing.overture.sql

COPY (
    SELECT * EXCLUDE (geometry,
                      bbox,
                      names,
                      sources),
           ST_GEOMFROMWKB(geometry) geom,
           JSON(bbox)    AS bbox,
           JSON(names)   AS names,
           JSON(sources) AS sources
    FROM read_parquet('s3://overturemaps-us-west-2/release/2024-06-13-beta.0/theme=buildings/type=building/*00208*',
                      filename=true,
                      hive_partitioning=1)
    WHERE bbox.xmin BETWEEN 116.28749 AND 116.52790  AND
          bbox.ymin BETWEEN 39.83297  AND 40.01533
)
TO 'beijing.overture.gpkg'
    WITH (FORMAT GDAL,
          DRIVER 'GPKG',
          LAYER_CREATION_OPTIONS 'WRITE_BBOX=YES');

The footprints largely haven't changed since at least April. It looks like orthofication has taken place. The tall buildings were photographed at an angle so it's only with linestring outlines that it is clear the footprints are probably very accurate. This data looks fine and doesn't seem to suffer the offset issue the Beijing data is having.

Microsoft's ML buildings dataset data for Abu Dhabi leaves a lot to be desired.