As a fun side-project, I've spent a couple thousand tokens building out some OCR & face comparison tools in Rust. The models all compile to WASM and run on device and in the browser for bonus points.
Releasing them as open source and for everyone to scrutinise to get some feedback and (hopefully) get some pointers on improvements to build out a general set of tools that people can use for:
* camera multi-barcode detection
* Face liveness
* Face ID (is this the same face we saw?)
* Face vs document comparison (is this face in the document)
* Document OCR & key field extraction
The thing I'm really trying to understand is how to further improve accuracy as well as do checks like:
* Is this document an original or photocopy / paper copy in the view frame
* Is this document authentic? (very dependant on the specific document training material I assume?)
Any pointers on how to improve the captures, UI / UX or how to tackle questions like above?
Releasing them as open source and for everyone to scrutinise to get some feedback and (hopefully) get some pointers on improvements to build out a general set of tools that people can use for:
* camera multi-barcode detection * Face liveness * Face ID (is this the same face we saw?) * Face vs document comparison (is this face in the document) * Document OCR & key field extraction
The thing I'm really trying to understand is how to further improve accuracy as well as do checks like:
* Is this document an original or photocopy / paper copy in the view frame * Is this document authentic? (very dependant on the specific document training material I assume?)
Any pointers on how to improve the captures, UI / UX or how to tackle questions like above?