Your image
?
Read in this browser and never uploaded —
there is no server in this application to send it to. Ground resolution is the
figure that matters: everything this app streams is 0.31–0.5 m per pixel, and at that
scale the wall between two touching houses is about one pixel wide. A drone frame at
5 cm is ten times finer, which is the whole reason for opening one here.
Position says whether the placement was read from the file or made by you.
Remove image
Placement
?
A GeoTIFF is placed from its own
georeferencing and nothing is guessed. A photograph has no coordinates, so you put
it roughly in place and snapping refines that against the imagery by correlation —
measured to 0.1 m from 33 m out.
Opacity
100%
Fit to view
Snap to imagery
Read the picture itself
?
A public building-segmentation model, 15 MB,
downloaded once and run in this browser. Everything else here re-serves what
someone else traced; this is the one thing that looks at your pixels —
which matters because every limit in this app comes from imagery at 0.31–0.5 m and
your file may be far finer. Its quality on dense informal fabric is unmeasured, so
the result is always scored against the published footprints.
Find buildings in it
Export that
Your polygons
?
Your survey, scored against the published data
building by building: which of yours sit on one of theirs, which you have and they do
not, which they have and you do not. Matched by centroid inside polygon , not by
overlapping bounding boxes — in dense fabric two neighbours share a wall, so their
boxes overlap heavily while neither centre is inside the other, and a box test scores
neighbours as matches. Only polygons can be matched; points are counted and ignored.
Remove them
Buildings
?
Everywhere
In selection
Off
Sources
?
Auto
Select
Confidence
off
?
Google scores every building it draws, from
0.65 to 1.0 — nothing exists below 0.65, so the slider starts there.
Raising it drops uncertain outlines and takes some real buildings with them:
measured on one view, 14,114 buildings at 0.65, 10,733 at 0.75, 5,480 at 0.90.
Overture buildings carry no score and are never filtered.
Your file
?
Open a file…
A GeoTIFF lands at its own coordinates. A JPEG
is placed from its EXIF GPS when a drone wrote one, otherwise by you and then
snapped to the imagery. GeoJSON, KML and CSV are compared against the published
data building by building. Everything is read in this browser — no file is
uploaded anywhere. Opening one switches this panel to that file's own controls
until you remove it.
Export
?
Download all
Mask only
One ZIP: the two-colour mask with touching
buildings separated, a colour-coded instance mask, the footprints as GeoJSON,
every statistic as CSV, and world files so the images land in the right place in
QGIS. Built in your browser — nothing is uploaded, and the README names every file
in the archive.
Click any building to read everything
its publisher records about it, or Select area to measure a
neighbourhood and export its mask.