State of the Art Mobile Machine Learning Vision App Release!

Standing on the shoulders of giants, my version of TFL Classify and FB MLVision are published. Much like Bluetooth Chat, my goal is to just have minimum modification to cross the finish line to publish. Now they are running on light weight but powerful mobile devices, I am amazed and feel magical to see face recognition, image classification and text recognition happen in real time. I guess it is fair to say those are state of the art machine learning capacities well packaged for developers to play with.

Comparison between two SDKs

  • Tensorflow Lite is more lower level and more open, it seems on device only (or may I say Local Native :-)

  • ML Kit for Firebase needs register the project and embed some google service configuration file downloaded from Firebase console, otherwise:

File google-services.json is missing. The Google Services Plugin cannot function without it.

  • ML Kit for Firebase also has cloud API for more advanced features, which requires developer to pay for usage.

So Tensorflow comes earlier and presumably is going to last longer given it is the foundation underlying Firebase, which as a rapid application framework, (sort of like AWS Amplify) , might subject to rapid change. Certain features are overlapping so they are also kind of competing with each other.

Other notes:

  • Rename java package com.google for android
  • The usual graphics work
  • Time from "Ready for Sale" to actually available on App Store can vary
  • Android icon on newer device seems to be round and if only square image is used it appears as small square within the circle with ugly gaps (目前得到也有这个问题,但少年得到没有:-)