hotosm
In browser AI prediction in fAIr
fAIr is seeking community support to implement an in-browser AI perdition support using localized AI models (computer vision) on multiple imagery sources of users choice. fAIr will provide all the necessary requirements for this implementation and contributors can do the implementation in the front end application.
Feature request details fAIr would like to enable users to run prediction in browser by:
- select specific imagery provider (OAM, Bing or any other OSM available imagery)
- select a specific AI localized model in fAIr in .onnx format
- zoom to an area of interest world wide
- run prediction in-browser (no backend call so only client resources are used)
- allow users to give feedback on the predictions accuracy
- download or share the predictions to multiples sources (MapSwipe projects, open in JOSM, share to HDX ..)
Why We see this feature important and useful for the following reasons:
- Produces less load on the prediction API in fAIr
- Adopts recent advancements of in-browsers AI predictions
- Anticipating that prediction will be faster in-browser
Assumptions
- Contributors have Front End development skills
- Project setup documentation is available here and fAIr team is available to support
- Contributors will maintain the credit by raising a PR that will be reviewed, tested and merged
- In case of the need for physical meeting, HOT will cover the expenses considering environmental emission vs. the need
Quick technical explanation
https://github.com/user-attachments/assets/18b154a0-ac2f-4db8-b767-42a1e0a2cbef
@petya-kangalova @LeenDhondt @kshitijrajsharma We are seeking community support for this new feature in fAIr, would you please review so we can bring it to sharable state before spreading it with wider community? Thanks
Hey @omranlm thanks for sharing! Really excited to see such clear dedicated tasks to bring contributors on board!
Suggestion:
- Title - We are seeking open source contributors, specifically front end developers, for developing in browser AI predictions in fAIr
- to replace "assumptions" with "what we are looking for". Are there any additional skills we are looking for the frontend developer? If so, will be good to add.
- add when we want this work done - maybe something in the lines of - we would to find a contribute to work on this in April/May and we anticipate the work taking XX hours (or specifying an estimate).
What next?
- If you are interested, respond to this issue expressing your interest to contribute - @omranlm would you like people to add more detail on how they will approach the work?
Very happy to share more widely!
One more sub issue will be required for the postprocessing @omranlm , Which involves converting predictions to vectors and changing it based on the user configuration including centroid of the feature !
