News for hackers. Stuff that matters.
Re:Machine Learning with TensorFlow Enabled Mobile Proof-Of-... by amelius on Tuesday October 17, 2017 @08:19AM (#15490298)
> This is undoubtedly incredibly impressive as a feat of engineering.

For years, post offices around the world have been automatically recognizing handwritten text, which is of much poorer quality and consistency than those codes.

Not sure what is new here. That it runs on a smartphone?

Re:Machine Learning with TensorFlow Enabled Mobile Proof-Of-... by fredley on Tuesday October 17, 2017 @10:33AM (#15491203)
Yes, and it runs on an image with a shadow occluding half the cap, held in someone's shaky hand, with low-contrast colouring. AFAIK, postal scanning is done in good lighting with known optical parameters of blue/black on white. The main part used for sorting the postal/zip code, which is relatively clear and in caps/numerals. Almost all mail has printed labels. Accuracy rates are not published, many people are hired to sort through the ones that cannot be automatically classified.

It may not seem like a huge jump conceptually but in engineering terms it's a big deal.