A brief line-up of AI-mapping #Mapflow models. The models are the core stack of the platform that orchestrates them and provides services to manage the input data.
Mapflow turns satellite and aerial imagery into ready-to-use vector maps. Here's what's in the box:
🏠 Buildings — global, 0.3 m / z19. Extracts rooftops, plus optional building-type classification, contour regularization, simplification, merge with OSM, and height estimation (beta) that projects the roof down to ground level for true 3D footprints. 📊 v.2026-07-06: mean F1 0.893 (aerial), 0.881 (satellite)
🌲 Forest & trees — global, 0.6–0.3 m. A solid vegetation mask covering sparse forest, shrubland, small tree groups and narrow tree lines. Options: height classes (<4 m / 4–10 m / >10 m) and individual tree crowns as polygons or points. 📊 v.2026-07-03: mean F1 0.850 — up from 0.513 in the previous version
🚗 Roads — 0.3–0.5 m, tuned for rural and suburban areas. Multi-task learning keeps the mask connected where roads hide under trees; centerline extraction plus graph cleanup (gap merging, simplification, removal of stray segments).
🧩 Buildings + Roads + Forest — all three in a single run, returned as one topology-corrected GeoJSON.
🗺 Open Data — pull OpenStreetMap and Overture layers with no AI involved, as a complement to model output.
⚙️ Custom models, on request: construction sites, high-density housing, agriculture fields, Segment Anything (zoom-dependent, from land use down to single trees), swimming pools (F1 > 0.95), solar panels. You can also bring your own model — or ask us to train one for your use case.
Run it all in the web app, the #QGIS plugin, or via #API. 🔗 mapflow.ai/models · docs.mapflow.ai
