TensorFlow Lite in Android
summary
TensorFlow Lite runs from Google Play services so apps can reduce binary size up to 5 MB by requesting the TFLite runtime and GPU delegate at runtime instead of bundling libraries. Migration means swapping dependencies, calling TFLite.initialize (async), choosing runtime options (FROM_SYSTEM_ONLY/FROM_APPLICATION_FALLBACK), and enabling delegates via a DelegateFactory to move inference from CPU to GPU or DSP; more than 10 apps already use it across over 1 billion monthly active users, and an Acceleration Service for per-device benchmarking is in early access.
AI-written summary · about
chapters (17)
- 0:00Introduction
- 0:35What is TensorFlow Lite
- 9:24Benefits
- 11:20Early Access Program
- 12:06Sample App
- 13:55Dependencies
- 14:40Classes
- 17:10Initialization
- 19:40Interpreter API
- 22:45Dependency
- 27:50move model to GPU
- 29:15performance
- 31:05languages support
- 31:25Java support
- 34:05API delegate
- 36:45Optimization
- 39:15Language support
related talks
34mTensorFlow Lite
30mOn-device machine learning in mobile and web apps
13mWhat's new in Android machine learning
32mAI for Mobile and IoT Devices: TensorFlow Lite
17mDeploy a custom Machine Learning model to mobile
47mMachine Learning on Android Demystified
14mOptimize your TensorFlow Lite models
21mHow to use hardware acceleration for Machine Learning inference on Android