Upgrade to Pro — share decks privately, control downloads, hide ads and more …

Superpower Your Android apps with ML: Android 11 | Devfest 2020

Superpower Your Android apps with ML: Android 11 | Devfest 2020

In this session, my major aim would be to provide an overview of the different tools one could use to power their Android apps with Machine Learning and also discuss the new additions for Machine Learning in Android 11 specifically the Model Binding Plugin and ML Kit. I would first go on to explain the standard procedure of using pre-trained models with MLKit. I would show how we could take the idea of MLKit forward and use pre-trained models from TensorFlow Hub to run right in the app, which would provide support to build high-quality machine learning apps based on models contributed from the community. I would then show how we could use custom TFLite models in Android apps, I would also talk about TensorFlow Model Maker and ML Model binding plugin in Android Studio through which I plan to show how easy it is to now use custom TF Lite models in Android apps. With Android 11 the NN API now supports Asymmetric integer weights making model sizes and inferences even smaller opening up an even larger opportunities for edge ML.

0d7c1e828ec0afbf29c0d37702c4637d?s=128

Rishit Dagli

October 17, 2020
Tweet

Transcript

  1. Superpower your Android Apps with ML: Android 11 Rishit Dagli

    High School TEDx, TED-Ed Speaker rishit_dagli Rishit-dagli
  2. • 11 Grade Student • TEDx and Ted-Ed Speaker •

    ♡ Hackathons and competitions • ♡ Research • My coordinates - www.rishit.tech $whoami rishit_dagli Rishit-dagli
  3. • Sayak Paul (ML GDE) • Khanh LeViet (Google) •

    Hoi Lam (Google) Acknowledgements
  4. • Mobile Devs looking for ways to build smarter apps

    • Mobile Devs looking for ways to integrate ML in their existing apps easily Ideal Audience
  5. Why care about ML in Android?

  6. Why care about on-device ML in Android?

  7. None
  8. None
  9. None
  10. None
  11. • Power Consumption Why should you care?

  12. • Power Consumption • Inference Time Why should you care?

  13. • Power Consumption • Inference Time • Network availability Why

    should you care?
  14. • Power Consumption • Inference Time • Network availability •

    Privacy Why should you care?
  15. ML Model Binding Plugin

  16. Easier to use Enable Hardware acceleration Faster Development ML Model

    Binding Plugin What’s new for on-device ML in Android?
  17. Importing a TF Lite Model

  18. Importing a TF Lite Model

  19. Importing a TF Lite Model

  20. Using the TF Lite Model

  21. Creating an Instance of model

  22. Processing images

  23. Passing in data

  24. Passing in data

  25. Passing in data

  26. Passing in data

  27. Adding labels

  28. We are done

  29. GPU Acceleration

  30. A new ML Kit

  31. Face detection Barcode scanning Image labeling Smart Reply Language Identification

    Vision Natural Language Object detection and tracking On-device Translation Text recognition Digital Ink Recognition Pose Detection N EW N EW
  32. On-Device ML Better customizability Generic use cases A new ML

    Kit What does the latest ML Kit focus on?
  33. Setup the model

  34. Customize the model

  35. None
  36. • High performance • Super easy to use • High

    customization too! TF Lite Model Maker
  37. TF Hub tfhub.dev

  38. TF Hub

  39. bit.ly/a11-ml Demos!

  40. Q & A rishit_dagli Rishit-dagli

  41. Thank You rishit_dagli Rishit-dagli