A regional rail authority asked for a site file of a small train station it wants to renovate: platform, freight shed, parking, and the two tracks. The catch was the audience. A structural engineer, a track designer, and a web developer all had to open the same file without calling me, and all three run different CAD workflows. Two tools, one boundary, and a very different set of outputs.
What CADMapper actually returns
CADMapper is a download service. You draw a box on a web map, pick layers, and get a CAD file built from open data. For this site I took buildings, roads, rail lines, and water, then exported DXF. The parcel boundary came through as a closed polyline, the platform edge as another one, and the shed as a set of footprints with no height.
That is the whole promise, and it is a good one. You get geometry in a coordinate system a CAD tool understands, in minutes, without a survey. What you do not get is elevation accuracy on the platform, or any indication of which line is the property line and which is a fence that has been there since 1974.
The units check that saves a week
Every DXF from a web service gets the same treatment before I trust it: measure a known object. Here the platform canopy is 42 meters long on the drawing set. The units check is the same every time. I pull the polyline length in the file's own units and compare.
A ratio near 1 means meters. Near 3.28 means feet dressed up as meters. Near 1000 means millimeters. I have seen all three from sources that swore the file was clean. The check takes ten seconds and has never once been wasted.
Where the geometry needs a human
Open data stops at the edge of what a satellite can see. The freight platform in the export was a clean rectangle. In reality it steps down 40 centimeters at the north end, which matters because the loading doors are there and a forklift does not like surprises. I fixed that in Rhino with two clicks and a note.
Mesh cleanup is the other recurring job. Rail alignments arrive as thousands of short segments. Merging them into polylines before export keeps the file usable for the track designer, who works in a different tool and does not want to select 4,000 line fragments by hand.
Then the same geometry gets a second life as GeoJSON. One script drops the layers I need into a feature collection, keeps the platform and parcel as polygons, and writes the track as lines with the same coordinates. The web developer opened that file in a WebGL viewer and had the site on screen before lunch, which never happens with a DXF.
GenAnywhere as a scene, not a site
GenAnywhere starts from the same box and produces a rendered scene: station, canopy, tracks, and a believable town behind it. For the public open house, that scene did what a DXF cannot. People understood the scale of the canopy from the platform, and the architect got useful feedback on the roof pitch in one evening.
As site data it is a dead end. The generated scene has no layers, no coordinates, and no dates. Its tracks curve for looks. Its platform is 9 meters wide where the survey says 6. It is a picture, and pictures are fine as long as nobody forwards one to a contractor.
The pipeline I keep
CADMapper output goes straight into a cleanup pass: merge segments, snap the parcel boundary, tag layers with the download date, and export DXF plus GeoJSON for the web viewer. The Three.js viewer reads the GeoJSON, so the developer gets the same geometry in a WebGL page without touching the DXF.
The GenAnywhere render goes into a concept folder, dated, and never enters the asset pipeline. When the structural engineer asks which polyline is surveyed, the layer name answers. When the developer asks why the platform looks different on the website, the GeoJSON and the render are in different folders with different jobs.
Two files, two truths, one rule: the coordinate system decides who gets to argue about it. Everything measured lives with the coordinates. Everything imagined lives with the colors.


Top comments (0)