My Imagery updated! π°οΈ Organise your data collections and run AI workflows in one shot. π
AI models for satellite imagery
We develop the models, train them, and run them in our cloud β you send an area of interest and get map-ready vectors back. Need something we don't have yet, or a model that has to stay on your own infrastructure? Both are available.
Run a ready model
Pick a model, draw an area, get GeoJSON. No GPUs to provision, no weights to manage, no MLOps. New model versions ship to every account automatically.
Open Mapflow βHave one built
Fine-tune an existing model on your data, or have a new one developed for classes and imagery we don't cover yet. Delivered as a hosted service on the same platform.
Custom models βRun it on your infrastructure
For data that cannot leave your perimeter, Mapflow deploys on-premise or in your private cloud, and orchestrates your own models alongside ours on your hardware.
Talk to us βAvailable in every account
Production models you can run today from the web app, the QGIS plugin or the API. Each page below documents the current version, options and measured accuracy.
Buildings
Rooftops and footprints from high-resolution imagery, with type classification, regularization and height estimation.
- Resolution:
- 0.3 m / z19
Forest and trees
Tree and shrub cover including sparse forest, shrubland, small tree groups and narrow tree lines.
- Resolution:
- 0.6β0.3 m / z18β19
Roads
Road networks extracted as centrelines and inflated back to polygons, tuned for connectivity under tree cover.
- Resolution:
- 0.3β0.5 m
Buildings + Roads + Forest
All three models in one pipeline, returned as a single topology-corrected GeoJSON.
Model reference βAvailable on request
Trained and validated, but enabled per account β usually because the geography, imagery type or use case needs a short conversation first.
Construction sites
Construction sites and buildings under construction, for monitoring development over time.
Model reference βHigh-density housing
Roofs in terraced and densely built areas typical of the Middle East and parts of Africa.
Model reference βBuildings (aerial)
Small buildings and detailed outlines from very high-resolution aerial imagery, for rural and suburban areas.
- Resolution:
- 15β10 cm
Agriculture fields
Field segmentation with boundary delineation, trained primarily on European and Russian farmland.
- Resolution:
- 1β1.2 m
Solar panels
Photovoltaic elements on rooftops and on the ground, from aerial imagery.
- Resolution:
- z19β21
- Benchmark:
- F1 > 0.87
Swimming pools
Swimming pool detection and classification from high-resolution aerial imagery.
- Resolution:
- z19β20
- Benchmark:
- F1 > 0.95
Segment Anything
Meta's universal segmentation model, adapted to run over large areas in Mapflow workflows.
Model reference βThis is not the whole catalog
The list above is what is documented publicly today. We regularly train models for classes that never appear here β they were built for a specific customer, on a specific imagery type, for a specific region. If what you need is missing, it usually means nobody has asked for it yet.
How custom models work βModels you don't have to operate
Most geospatial AI tooling hands you a model and leaves the hard part β inference at scale, versioning, GPU capacity, retraining β to you. Mapflow is the opposite: we build the model, we run it, and you consume the output.
Hosted by Mapflow
Models run in our cloud on our hardware. You get an API, a web app and a QGIS plugin, plus every model update as it ships. Nothing to deploy, nothing to keep running.
Deployed on your infrastructure
When data residency or security rules out a hosted service, the same platform runs on-premise or in your private cloud β and Mapflow orchestrates your own models there too, so in-house and Mapflow models share one workflow engine.
Try a model on your own area
Free credits on signup β run the production models on your area of interest before talking to anyone.