Building predictions for Zushi, Japan missing

It looks like there might have been a decision to not run inference around Zushi, Japan based on tiles as there is a clear break ~5KM before the coastline.

OpenStreetMap doesn't have any building footprints for this area.

There is a PLATEAU project by the Japanese government to map out buildings in 3D but they haven't covered Zushi yet. Other Japanese public dataset have points for each of the buildings but no polygon footprints.

Any inference of the buildings in this area would produce a pretty unique dataset.

These were the steps I took to extract the buildings:

$ wget https://minedbuildings.blob.core.windows.net/global-buildings/dataset-links.csv

$ cat dataset-links.csv \
    | ~/duckdb \
            -json \
            -c "SELECT Url
                FROM READ_CSV('/dev/stdin')
                WHERE Location = 'Japan'" \
    | jq '.[].Url' \
    > japan.txt

$ cat japan.txt | xargs -P4 -I% wget -c %

$ ~/duckdb buildings.duckdb

CREATE OR REPLACE TABLE buildings (
    height     DOUBLE,
    confidence DOUBLE,
    geom       GEOMETRY);

$ for FILENAME in *.csv.gz; do
      gunzip -c $FILENAME | gzip -1 > working.jsonl.gz # Fix trailing garbage complaint
  
      echo "INSERT INTO buildings
            SELECT properties.height,
                   properties.confidence,
                   ST_GEOMFROMGeoJSON(geometry) geom
            FROM READ_NDJSON('working.jsonl.gz')" \
          | ~/duckdb buildings.duckdb
  done


$ ~/duckdb buildings.duckdb

COPY(
    SELECT height,
           confidence,
           ST_AsWKB(geom) AS geom
    FROM   buildings
    WHERE  ST_Y(ST_CENTROID(geom)) IS NOT NULL
    AND    ST_X(ST_CENTROID(geom)) IS NOT NULL
    ORDER BY HILBERT_ENCODE([ST_Y(ST_CENTROID(geom)),
                             ST_X(ST_CENTROID(geom))]::double[2])
) TO 'japan.pq' (
    FORMAT 'PARQUET',
    CODEC  'ZSTD',
    ROW_GROUP_SIZE 15000);

Between the trailing garbage complaint when decompressing the GZIP files and some geometry producing NULL centroid coords there might be some other issues at play.

Start by reproducing the extraction described with dataset-links.csv, japan.txt, the downloaded *.csv.gz files, and buildings.duckdb. Inspect whether the missing Zushi coverage is caused by source data, decompression, geometry filtering, or the inference pipeline. Done means identifying the cause and documenting or correcting the coverage issue, with a reproducible check for Zushi.