What users like about fAIr

[adiatmad]

Following the presentation on the fAIr update plan to the 3.0 version, here are my thoughts. Thank you!

1. What did you like about fAIr?

• I liked that fAIr allows models to be trained on locally labeled data, which respects and leverages local knowledge. It’s a big step toward inclusive AI.
• The design encourages human validation instead of replacing mappers, which helps maintain the quality of OSM data and community trust.
• I like the wizard-style setup, which makes it more accessible for the general public.

2. If you were to mention one key improvement, what would it be?

• Improve post-processing to cluster or simplify predictions so it’s easier to validate areas with dense building footprints.
• If feasible, fAIr could collect anonymized validator feedback (accept/reject/edit actions) and provide a mechanism to feed this back into model refinement or retraining.

[omranlm]

Great, thanks @adiatmad for that ... Totally agree about the post processing .. it is part of our 2025 roadmap and I am pretty sure it will come to focus in Q3/4 this year.