Frequently asked questions¶

So what is fAIr, really?¶

fAIr runs AI models over aerial or drone imagery of your area and returns the features they find, such as buildings or trees, as a GeoJSON layer you can open in QGIS. You choose the model, and you stay in control at each step.

Is it AI on its own?¶

No. fAIr brings existing geo-AI models to mappers. You pick a model, run it, adapt it, and validate the results.

Do I need high-resolution imagery? Will Sentinel work?¶

fAIr needs openly licensed, very high-resolution RGB imagery, ideally hosted on OpenAerialMap or a public TMS server. The buildings model expects about 30 to 50 cm per pixel, so 10 m Sentinel-2 imagery is too coarse. Support for street-level imagery is being explored.

How do I validate the output? I do not trust AI.¶

You get the predictions as GeoJSON, and you choose how to check them: confirm them at scale in MapSwipe, verify them on the ground with FieldTM, or fix them by hand in QGIS.

Do I have to be advanced to start?¶

No. Begin as a mapper using a model someone already made. When you want it tuned to your area, you can grow into adapting and publishing your own.

Can I add my own model?¶

Yes. Developers write model code and contribute it to fAIr, where it becomes available to every kind of user. See Contributing a model.

Where does the finished data go?¶

You can download the outputs as points, lines, or polygons. HOT publishes validated outputs as open data layers on the Humanitarian Data Exchange and Hugging Face; see Datasets.

Does fAIr add its data to OpenStreetMap?¶

No. See fAIr and OpenStreetMap.