madeye

如何用 TensorFlow 打造 Not Hotdog 的移动应用

  •  
  •   madeye · Jul 5, 2017 · 4062 views
    This topic created in 3265 days ago, the information mentioned may be changed or developed.

    原文: https://hackernoon.com/how-hbos-silicon-valley-built-not-hotdog-with-mobile-tensorflow-keras-react-native-ef03260747f3

    Not Hotdog 官网: https://www.seefoodtechnologies.com/nothotdog/

    为了提高移动设备上的执行效率,并减小 model 以及的安装包尺寸,作者做的主要优化包括:

    1. Rounding the weights of our network helped compressed the network to ~25% of its size.
    2. Optimize the TensorFlow lib by compiling it for production with -Os
    3. Removing unnecessary ops from the TensorFlow lib

    我自己也基于 TensorFlow 的 Android Example 做了一个 YOLOv2 的移植( https://github.com/madeye/yolo-android ),过程意想不到的顺利。经过 Quantization 后的安装包大小差不多 20MB,识别速度可以达到 2 FPS 左右 ( Google Pixel )。

    感觉可以尝试的事情还有很多,独立开发者们不妨在自己的应用里试试基于 Deep Learning 技术。

    No Comments Yet
    About   ·   Help   ·   Advertise   ·   Blog   ·   API   ·   FAQ   ·   Solana   ·   3202 Online   Highest 6679   ·     Select Language
    创意工作者们的社区
    World is powered by solitude
    VERSION: 3.9.8.5 · 30ms · UTC 03:51 · PVG 11:51 · LAX 20:51 · JFK 23:51
    ♥ Do have faith in what you're doing.