RFC: Proposal for Face detection API


Certainly would not hurt to add “accelerated” - specially if it clears up confusion about what the API is about.

1 Like





This API seems to be in quite wrong level. Way too higher level. Wouldn’t it be better to expose lower level capabilities to the platform, possibly explicit facial detection being among those. I’m hoping there will be soon some proposal (not from me) to handle this kind of detection in a more uniform way.



I’m totally exited about this feature. I’m very interested in TextDetection and am already testing it in my application to learn languages. That something I was looking forward for a long time. My use case is when one read a book (real, physical one) written on a foreign language and encounter an unknown word. Therefore with mobile camera and TextDetection API he is able to save this word with its context in order to train it later. It’s a huge time saving feature I always wanted.



Shape API for Face, Barcode and Text Detection would be a huge value add to the financial sector for client on boarding processes. Every customer needs to be FICA’d using their identity document similarly to the “voting” use case.

  • The face points needs to be determined and qualified
  • The barcode details needs to be extracted for customer verification
  • The Identification values extracted to be compared against another system. (if not on barcode)

The current OCR solutions (cloud or onpremise) on the market require the capture of the image to have a certain quality for these steps to happen 100% of the time (lighting, resolution-dpi, image format, boarder control\cropping). This becomes a problem when left to the general public to do.

I am only aware of mobile capture SDK’s that provide this capability however in our use-cases the customers interact from the web page before downloading the mobile app therefore bypassing the need for an SDK. To be in a position to dictact the quality of the image capture upfront through the web would put us in a position effectively onboard the customer 100% of the time, or much closer than what it is now.