Flyover Ghent

Open-source 3D art built on real city data. All three projects use the "Gent in 3D" dataset from data.stad.gent to render the city of Ghent. Two run in the browser using Three.js, one outputs a slicer-ready STL ready for 3D printing.

As presented at BYOB 2025

Projects

Fly Over Gent

Live DemoGitHub

Free-flight through all of Ghent using official municipal 3D data. Explore 150+ tiles of the city with WASD controls.

WASD free-flight controls

Dynamic tile loading 150+ city tiles

Python pipeline DWG→DXF→STL preprocessing

3-tier tile caching Service Worker → memory → scene

Direction-aware loading only fetches tiles the camera faces

Live minimap color-coded tile readiness

Built with Three.js. Python preprocessing with ODA File Converter, ezdxf, and trimesh. Lambert-72 coordinates.

Drones Over Ghent

Live DemoGitHub

Autonomous drone flocking over a 3D model of Ghent. Drones use boid algorithms to exhibit natural swarming behavior above the city skyline.

Boid flocking with autonomous agents

4 camera modes follow, chase, orbit, free-flight

Real-time GUI controls for tweaking parameters

Built with Three.js and vanilla JavaScript.

Print Ghent

Live DemoGitHub

Pick an area on the map, get a printable slab of the city. Free, no signup, runs entirely in your browser. Uses the same tiles as Fly Over Gent so the whole Ghent dataset is available to crop from.

Map selection Shift-drag for square, Alt for freeform

Watertight slab plane-clipped terrain + 4 walls + flat bottom

Drop edge buildings optional clean cut along the bbox

Configurable base set slab thickness in meters

Pure browser pipeline no backend, STLExporter output

Built with Three.js, Leaflet, and proj4. Tiles served from the same GCS bucket as Fly Over Gent.

Data & Pipeline

The Data

Gent in 3D Dataset

The city of Ghent publishes its full 3D model as open data through the "Gent in 3D" dataset. It covers the entire city in 1km² tiles using the Lambert-72 coordinate system — buildings as LOD2 geometry (rooftops modeled from LiDAR, walls extruded from cadastral contours) and terrain from ground-point elevation data.

Processing Pipeline

The entire pipeline is automated with python run_all.py --download — it downloads all 150+ tiles from the Gent open data API, extracts the DWG files, and batch converts them to STL in parallel.

.ZIP downloaded from data.stad.gent API

↓

download_gent_data.py

.DWG LiDAR scan data

↓

ODA File Converter (cached)

.DXF vector geometry

↓

ezdxf — parse 3DFACE entities

.STL triangle meshes (parallel)

↓

trimesh — mesh export

Three.js rendered in browser

Map Caching

Direction-Aware Tile Caching

Caching to speed up loading times, takes into account camera direction and position. Tiles persist between sessions so revisited areas load instantly.

Tile States

Available — exists in dataset, never fetched

SW Cached — stored in browser cache

In Memory — geometry parsed, ready to re-add

Loading — fetch in-flight

Visible — rendered in scene