Getting the best of on-device ML with ML Kit

on YouTube ↗21m

" summary

Covers practical ML Kit details developers need to ship: which APIs are on-device vs cloud (on-device image labeling ~400 labels, cloud ~10,000), 18× faster face model with 13–24% accuracy gains, TensorFlow Lite custom models, hosting via Firebase or bundling, thin vs thick SDK tradeoffs, App Bundle and dynamic feature tips to cut APK size, runtime rules (face >100×100 px, contour >200×200, char sizes), and a real case of model compression 80MB → <1MB showing retained accuracy.

AI-written summary · about

" chapters (5)

1. 0:00Intro
2. 8:30Enhancements
3. 9:39Tips and Best Practices
4. 16:28Base API
5. 18:16Upcoming features

" related talks

• 24mML Kit: Machine learning for mobile developersSachin Kotwani, Isabella Chen · Firebase Summit 2018
• 04mTop 5 reasons to add ML Kit to your Android or iOS appSachin Kotwani · Firebase Summit 2018
• 29mPushing Machine Learning into your app When TensorFlow and MLKit rule the worldSandra Dupré · DevFest Nantes 2018
• 30mMachine Learning in Android Apps with ML KitFelipe Pedroso · AndroidFest Campinas 2018
• 39mFrom TensorFlow to ML Kit: power your mobile application with machine learningQian Jin · Mobilization 8
• 12mML Kit: Turnkey APIs to use on-device ML in mobile appsChenxi Song, Kenny Sulaimon · Google I/O 2021
• 39mCustomize Your App with MLKitVictoria Gonda · Øredev 2019
• 10mMachine Learning for mobileKemal El Moujahid · Google's Machine Learning Virtual Community Day 2021