osm-to-3dprint

Export OpenStreetMap (OSM) features to STL for 3D printing.
Now supports streets (buffered & extruded roadbeds) in addition to buildings.

• License: MIT (free for commercial use)
• Why this over CadMapper? No 1 km cap; export much larger areas.
• Printer-friendly: Models are watertight and tuned for slicers.
• Fast: Typical city tiles generate in under a minute.

Support the project by grabbing a 3D print from Etsy.com!

---

What’s new (Roads!)

• Roads are pulled from OSM edges (e.g., highway=residential, primary, …).
• Each edge is buffered per-edge (no global union), triangulated, and extruded slightly above the base to avoid being fused by slicers.
• You can choose a single uniform width (simplest) or turn on per-type widths for realism (motorway > residential).
• Tunable parameters:
  • UNIFORM_WIDTH_MM (or per-type map) – model-space road width
  • road_height_mm – relief above the base (e.g., 1.4–1.8 mm)
  • EPS – tiny Z-offset (e.g., 0.02 mm) to prevent coplanar merging

Tip: At ~2 km mapped to ~180–200 mm, 1 mm ≈ 9–11 m. Start with 0.6–1.0 mm for residential roads.

---

Downtown Philly ^^

Installation

We recommend OSMnx v2 and recent geospatial libs.

# (optional) new virtual environment
# python -m venv .venv && . .venv/Scripts/activate   # Windows PowerShell
# source .venv/bin/activate                          # macOS/Linux

pip install "osmnx>=2.0.2" "shapely>=2.0.4" "geopandas>=0.14" "pyproj>=3.6" "networkx>=3.2" "numpy-stl"

On Windows, using Conda can be even smoother:
conda create -n osm2stl -c conda-forge python=3.11 osmnx shapely geopandas pyproj networkx numpy-stl -y

---

Usage

1) Pick a bounding box

This repo uses the internal convention:

# our code: (north, east, south, west)
bbox = (N, E, S, W)

OSMnx v2 calls that as bbox=(west, south, east, north) when fetching features/graphs, and the code already converts for you.

Center City Philadelphia (Broad & Market as center, ~2 km square):

bbox = (39.9615, -75.1520, 39.9435, -75.1750)  # (north, east, south, west)

2) Run

python main.py

This produces e.g. city_with_roads_and_buildings.stl.

---

Key script knobs (in main.py)

Buildings

• target_size (mm): width of the modeled area on your print bed (the base is slightly larger to frame).
• max_height_mm: tallest building height in mm after scaling.
• default_height & building:levels:
  The code checks ['height', 'building:height', 'building:levels'] (levels × ~3 m) if explicit heights are missing.

Roads (per-edge buffering & extrusion)

Inside scale_coordinates(...) the roads block:

• Uses per-edge buffering (no global dissolve/union), then triangulates/extrudes each piece.
• Keeps a tiny EPS Z-offset so slicers don’t merge roads into the base.
• Logs a diagnostic line so you can tune widths:

  [roads-per-edge] pieces = 1234, summed_coverage = 0.064
  

  Aim for ~0.04–0.10 coverage at city scale; if you see ~0.20, your roads are too wide (they’ll look like a second base slab).

Tuning:

• Start with UNIFORM_WIDTH_MM = 0.8 and road_height_mm = 1.6.
• If roads look too faint in slice, bump width to 1.0–1.2; if they “blanket” blocks, drop to 0.6–0.7.
• Once you like it, turn on per-type widths (motorways bigger; residentials smaller).

---

Example: Philly quick start

In main():

# bbox = (north, east, south, west)
bbox = (39.9615, -75.1520, 39.9435, -75.1750)

buildings = fetch_building_data(bbox)          # OSMnx v2: features_from_bbox under the hood
roads     = fetch_road_data(bbox, 'drive')     # OSMnx v2: graph_from_bbox(bbox=(W,S,E,N))

vertices, faces = scale_coordinates(
    gdf_buildings=buildings,
    bbox=bbox,
    target_size=180,
    max_height_mm=40,
    default_height=40,
    base_thickness=1.8,
    roads_gdf=roads,
    road_height_mm=1.6
)

save_to_stl(vertices, faces, 'city_with_roads_and_buildings.stl')

The roads section (already in the file) is configured for per-edge extrusion and prints a coverage diagnostic.

---

Troubleshooting

“I don’t see roads in the slicer.”

• Ensure per-edge block is enabled (no global unary_union).
• Keep EPS = 0.02 (or 0.03) so roads start slightly above the base.
• Set road_height_mm ≥ 1.4 so the terrace is visible across a few layers.
• Check the coverage log:
  • summed_coverage > 0.15 → widths too large; reduce to 0.6–0.8 mm.
  • summed_coverage < 0.02 → widths too small; raise to 1.0–1.2 mm.

“Diagonal ‘rays’ from a corner.”

That’s fan-triangulation on concave polygons. The script now uses Shapely triangulation for roads to prevent this.

“OSMnx function signature errors / bbox order.”

You’re likely mixing OSMnx v1 and v2 docs. This repo assumes OSMnx ≥ 2.0.

• Use features.features_from_bbox(bbox=(W, S, E, N), tags=...).
• Use graph_from_bbox(bbox=(W, S, E, N), network_type='drive', simplify=True).

---

Road width realism (optional)

Once the geometry looks good, switch to per-type widths (still per-edge, no union):

USE_PER_TYPE = True
width_map = {
    "motorway": 3.2, "trunk": 2.8, "primary": 2.6,
    "secondary": 2.2, "tertiary": 1.8, "residential": 1.2,
    "service": 1.0, "living_street": 1.0, "unclassified": 1.2,
    "cycleway": 0.9, "footway": 0.8, "path": 0.8
}
# clamp to [0.6, 3.2] in model mm; 1 mm ~ 9–11 m at this scale

---

Example outputs

---

Acknowledgments

• Building height attributes may be missing in some regions; the code checks multiple tags and falls back to defaults.
• Special thanks to ChatGPT by OpenAI for pairing on design and debugging.

---

Roadmap

• Optional separate STLs: export roads and buildings as distinct files for dual-color prints or per-part slicing.
• Optional sidewalks, water polygons, and parks for additional detail.
• CLI & presets per city scale (auto width/height suggestions).

---

Happy printing! 🗺️🧱🛣️