Draw a box on the map, get a color 2D map and a 3D white-model scene of that area, export to 17 formats, and take them into Blender / AutoCAD / QGIS / Cesium.
English | 简体中文
A web tool that runs on your own computer. Hold the left mouse button and drag a box on the map; the backend goes off to fetch OpenStreetMap data and terrain elevation for that patch of ground, and about a minute later you can:
- View a 2D color map — buildings, roads, railways, water and green areas drawn in their own colors, like a proper paper map
- View a 3D white model — buildings stand up at their real heights, the ground undulates, six layers you can toggle on and off
- Export to 17 formats and carry on in whatever software you like
All data comes from free public sources (OSM + AWS terrain tiles). No account, no API keys.
draw a box ──► fetch OSM ──► fetch Overture buildings ──► fetch terrain ──► parse ──► build ──► results
│ (inside China, an extra open building dataset │
│ is fetched and merged with OSM) │
│ │
│ ┌─────────────┼─────────────┐
│ ▼ ▼ ▼
└── progress streamed live hub.glb preview.geojson meta.json
│ │
3D formats 2D/GIS formats
Six steps in total; the web page shows where each one is at, and you can cancel at any time.
- Drag a box straight on the map, area shown live; boxes that are too big (over 25 km²) get rejected
- No missing buildings in mainland China: an extra set of ML-extracted building footprints from Overture Maps (free open data, no key needed) is fetched and merged with OSM — OSM lags badly in China and new districts are full of gaps; Overture fills them. The GCJ-02 "Mars coordinates" offset in that data is corrected automatically, and a satellite basemap toggle lets you verify the alignment by eye
- Six feature classes with one consistent palette: SVG / PDF are vector and stay sharp at any zoom; PNG / DXF / GeoJSON each serve their own purpose
- Buildings extruded by their
heighttag, estimated from floor count when missing, then from building-class priors; roads and railways get widths by class; terrain follows AWS elevation data - Six 3D formats (GLB / OBJ / STL / USDZ / FBX / DAE), four 2D formats (SVG / PDF / PNG / DXF), seven GIS formats (GeoJSON / GPKG / SHP / KML / KMZ / CityJSON / 3D Tiles)
- "Up" is already sorted for you: GLB / USDZ use Y-up (Blender, three.js convention), OBJ / STL / CityJSON use Z-up (CAD convention), and 3D Tiles carry their own georeference — drop them into Cesium and they land in place
- Flaky networks don't leave you hanging: data fetching rotates through mirrors; if terrain can't be fetched at all it falls back to flat ground, tells you why, and the task keeps going
- Progress is pushed over WebSocket, reconnects automatically, and falls back to another channel if WebSocket won't connect at all
- Layer names survive the whole way: nodes inside the GLB are literally named TERRAIN, BUILDING and so on, so they line up with the layer panel in Blender
| Software | Version | Why |
|---|---|---|
| Python | ≥ 3.11 (3.12 / 3.13 recommended) | runs the backend |
| Node.js | ≥ 20 | runs the frontend |
| Blender (optional) | any recent version | only needed for FBX / DAE |
Two commands on Windows:
# 1. Start the backend (first run creates the venv and installs dependencies)
powershell -ExecutionPolicy Bypass -File scripts\dev-backend.ps1
# 2. In another window, start the frontend
powershell -ExecutionPolicy Bypass -File scripts\dev-frontend.ps1Open http://127.0.0.1:5173, drag a box, hit "Generate".
Hate two windows? Run
scripts\run_all.ps1. Not sure your environment is complete? Runscripts\check_env.ps1first.
docker compose up -d --build
# open http://localhost:8080Everything generated is stored under ./data, so it survives container rebuilds.
Don't want to set up Python and Node at all? Grab the desktop build from Releases: unzip, double-click map2model.exe, done. It opens in its own window, all 17 export formats included (FBX / DAE light up automatically if Blender is installed on that machine — something the Docker image can't do).
A few things worth knowing:
- The exe is not code-signed, so the first launch shows Windows SmartScreen: click More info → Run anyway.
- Blender installed somewhere the auto-detect doesn't look (it checks
PATHandC:\Program Files\Blender Foundation\)? Create amap2model.envfile next to the exe withM2M_BLENDER_PATH=D:\Blender\blender.exe— seedesktop/map2model.env.examplefor the knobs. - Your data (database, task outputs, logs) lives in
%LOCALAPPDATA%\map2model. Delete that folder to reset everything. - The app needs internet access — it fetches map data and terrain tiles from the network, like the web version does.
- Clicking Download opens a save-file dialog, starting in your system Downloads folder.
- If the system lacks the WebView2 runtime (rare on Windows 10/11), the app falls back to opening in your default browser instead.
To build the exe yourself from source: powershell -ExecutionPolicy Bypass -File desktop\build-desktop.ps1 -Smoke -Zip.
| Group | Format | Status | Good for |
|---|---|---|---|
| 3D | GLB | ✅ | the default choice. Opens directly in Blender and three.js, layer names included |
| 3D | OBJ | ✅ | universal in modeling tools, Z-up |
| 3D | STL | ✅ | 3D printing, millimeters |
| 3D | USDZ | ✅ | send to an iPhone, preview in AR with the Files app |
| 3D | FBX | Unity / Unreal / Maya | |
| 3D | DAE | classic 3D interchange format | |
| 2D | SVG | ✅ | vector map, scale it as much as you like |
| 2D | ✅ | printing | |
| 2D | PNG | ✅ | image snapshot |
| 2D | DXF | ✅ | AutoCAD, layers pre-sorted as M2M_* |
| GIS | GeoJSON | ✅ | the most universal vector format, everything reads it |
| GIS | GPKG | ✅ | multiple layers in one file, opens straight in QGIS / ArcGIS |
| GIS | SHP | ✅ | traditional GIS format, zipped per layer |
| GIS | KML / KMZ | ✅ | Google Earth, buildings extruded |
| GIS | CityJSON | ✅ | city modeling standard (version 2.0) |
| GIS | 3D Tiles | ✅ | big scenes in Cesium, georeferenced |
✅ = works out of the box; C:\Program Files\Blender Foundation\ by itself; on other platforms set M2M_BLENDER_PATH).
Buttons in the UI tell the truth about availability: usable ones are lit, the rest are grey, and hovering tells you what's missing.
The six layers use the same names and colors in 2D and 3D:
| Layer | Color | Notes |
|---|---|---|
| Terrain | #c8c3ba |
the ground (undulates when elevation data is available) |
| Green | #b7d6a8 |
parks, grass, woods |
| Water | #a8cfe8 |
rivers, lakes |
| Building | #ded7cb |
buildings, extruded to real or estimated heights |
| Road | #a8a8a8 |
roads, main and side roads get different widths |
| Railway | #6b6f73 |
railways |
It runs with zero configuration. To tweak, set these environment variables (all prefixed M2M_):
| Variable | Default | Description |
|---|---|---|
M2M_DATA_DIR |
data |
where the database and generated files live |
M2M_OVERPASS_ENDPOINTS |
a built-in list | comma-separated, tried front to back |
M2M_OVERTURE_RELEASE |
2026-09-23.1 |
Overture buildings release; changing it rebuilds the row-group index |
M2M_OVERTURE_TIMEOUT |
60 |
per-file timeout for Overture fetches (seconds) |
M2M_MAX_BBOX_AREA_KM2 |
25 |
max area per task |
M2M_MAX_BUILDINGS |
20000 |
max buildings per task |
M2M_TASK_CONCURRENCY |
2 |
how many tasks run at once |
M2M_BLENDER_PATH |
auto |
where Blender is; auto = go find it |
Optional extras: pip install -e ".[gis,usd]" makes GPKG go through geopandas and USDZ through the official library; pip install -e ".[dev]" installs development and test tools. Everything works without them — the UI just reports what's available.
Q: npm keeps failing with corrupt files?
On some Windows machines antivirus interferes with npm's unpacking (TAR_ENTRY_ERROR, zero-byte files). This project uses pnpm throughout and has no such trouble: corepack pnpm install.
Q: How long does a run take? Most time goes to pulling data from Overpass: about a minute for 0.3 km² of urban area, slower the bigger the box. The second run on the same area hits the cache and comes back almost instantly.
Q: First run in China sat ten-plus minutes on "fetching Overture data"? That's expected. Overture's building data is shuffled globally, so the first use scans it once to build a local index (a one-time cost, cached afterwards) — every later run takes seconds. If you can't wait, switch the building source to "OpenStreetMap only" in the task panel.
Q: Why is my terrain flat? Some networks can't reach AWS terrain tiles; the program then falls back to flat ground and says why in the task's warnings. Use a proxy or another network and it comes back.
Q: FBX / DAE buttons are grey? Those two formats borrow a hand from Blender. Install Blender (free) and restart the backend — they light up. The other 15 formats are unaffected.
Q: The model lies on its side in Blender? That's not a bug. "Up" inside a GLB is Y, and Blender straightens it automatically on open. If you imported OBJ (Z-up) and it looks lying down, rotate it manually.
# backend tests (82 cases)
cd backend; .venv\Scripts\python.exe -m pytest -q
# frontend type check + unit tests
cd frontend; corepack pnpm typecheck; corepack pnpm test
# frontend build
cd frontend; corepack pnpm buildOn Python 3.14: mapbox-earcut has no 3.14 package yet, so the program automatically switches to its own triangulation code — same results, just slower. The Docker image is pinned to 3.12 and doesn't have this issue.
Leo Xie — GIS / 3D / full-stack. Open to custom development and consulting: 742875110@qq.com · GitHub
The code is licensed under the GNU AGPL-3.0 (or later). You are free to use, study, modify, deploy and run it — including commercially — with one condition: if you modify it and make the modified version available to others, even only as a network service, you must release your modifications under the same license.
Commercial license: if you want to embed map2model in a closed-source product, or run it without open-sourcing your modifications, contact 742875110@qq.com.
One reminder: the maps and models this tool produces come from two open datasets — OpenStreetMap (ODbL license) and building footprints from the Overture Maps Foundation (CDLA-Permissive 2.0, used mainly for the China building infill). Personal use is fine; if you publish or commercially distribute the results, include the attributions: one "© OpenStreetMap contributors" line and one "Buildings © Overture Maps Foundation" line.