TensorFlow Lite for Android Neural Networks API

  • Posted on: 14 November 2017
  • By: oon
TensorFlow Lite Architecture

TensorFlow untuk aplikasi mobile sudah mulai berkembang jauh sejak dari versi 1. (sekitar setahun yang lalu saya tuliskan [1])

Kali ini disebut dengan TensorFlow Lite [2][8], dengan arsitektur yang bisa (rencananya) cross-platform, setidaknya saat ini untuk Android dan iOS. Arsitekturnya dapat dilihat pada gambar.

TensorFlow Lite secara bawaan sudah mendukung sejumlah model yang sudah di-training dan dioptimalkan untuk perangkat mobile (ponsel atau embedded), antara lain:

  • MobileNet: A class of vision models able to identify across 1000 different object classes, specifically designed for efficient execution on mobile and embedded devices [3]
  • Inception v3: An image recognition model, similar in functionality to MobileNet, that offers higher accuracy but also has a larger size[4]
  • Smart Reply: An on-device conversational model that provides one-touch replies to incoming conversational chat messages. First-party and third-party messaging apps use this feature on Android Wear [5]

Oh ya, Inception V3 dan MobileNets di-training dengan dataset ImageNet [6], tapi dengan transfer learning [7] bisa dengan mudah di-training ulang dengan dataset image sendiri.

Satu hal lagi tentang Android Neural Networks API (NNAPI) [9], sejak Android versi 8.1.

[1]http://oo.or.id/content/tensorflow-better-support-mobile-and-macos
[2]https://developers.googleblog.com/2017/11/announcing-tensorflow-lite.html
[3]https://research.googleblog.com/2017/06/mobilenets-open-source-models-fo...
[4]https://arxiv.org/abs/1512.00567
[5]https://research.googleblog.com/2017/11/on-device-conversational-modelin...
[6]http://www.image-net.org/
[7]https://www.tensorflow.org/tutorials/image_retraining
[8]https://github.com/tensorflow/tensorflow/tree/master/tensorflow/contrib/...
[9]https://developer.android.com/ndk/guides/neuralnetworks/index.html

 

Add new comment

Filtered HTML

  • Web page addresses and e-mail addresses turn into links automatically.
  • Allowed HTML tags: <a> <em> <strong> <cite> <blockquote> <code> <ul> <ol> <li> <dl> <dt> <dd>
  • Lines and paragraphs break automatically.

Plain text

  • No HTML tags allowed.
  • Web page addresses and e-mail addresses turn into links automatically.
  • Lines and paragraphs break automatically.
By submitting this form, you accept the Mollom privacy policy.