{"id":46748,"date":"2020-03-27T10:36:01","date_gmt":"2020-03-27T17:36:01","guid":{"rendered":"https:\/\/devblogs.microsoft.com\/xamarin\/?p=46748"},"modified":"2020-09-15T11:13:55","modified_gmt":"2020-09-15T18:13:55","slug":"image-classification-xamarin-android","status":"publish","type":"post","link":"https:\/\/devblogs.microsoft.com\/xamarin\/image-classification-xamarin-android\/","title":{"rendered":"Android Image Classification with TensorFlow Lite & Azure Custom Vision Service"},"content":{"rendered":"

This is a guest blog by Daniel Hindrikes<\/a>. Daniel is a Microsoft MVP, Speaker, and Xamarin + Cloud Solution Architect\u00a0at\u00a0tretton37<\/a>.<\/span><\/em><\/p>\n

Image Classification allows our Xamarin apps to recognize objects in a photo.<\/p>\n

Being able to take a photo and recognize its contents is becoming more and more common. We experience it in our banking apps when making a mobile deposit, in our photo apps when adding filters, and in our HotDog apps to determine whether or not our meal is a hotdog.<\/p>\n

Thanks to Azure Custom Vision Service, we don’t have to learn\u00a0complex machine learning algorithms to implement image classification.<\/p>\n

This article will walkthrough how to implement it using Azure’s Custom Vision Service<\/a>, TensorFlow Lite<\/a>\u00a0(an open source machine learning platform) and Xamarin.Android<\/a>.<\/p>\n

Note:<\/strong> For Xamarin.iOS<\/a>, we can also use Azure’s Custom Vision Service<\/a> with CoreML<\/a>, but we’ll save that for another blog post.<\/p>\n

\"Image<\/center><\/p>\n

<\/h2>\n

Image Classification Libraries<\/h2>\n

We’ll be leveraging Azure Custom Vision Service and TensorFlow Lite to implement our image classification.<\/p>\n

1. Azure Custom Vision Service<\/h3>\n

Azure’s Custom Vision Service<\/a> makes it easy to create and train machine learning models – no previous Artificial Intelligence (AI) or Machine Learning (ML) experience is necessary.<\/p>\n

Using the Custom Vision Service Web Portal<\/a>, we can do the following without writing any AI\/ML code:<\/p>\n

    \n
  1. Upload a training image<\/li>\n
  2. Tag \/ Label the object(s) in the image<\/li>\n
  3. Repeat (the model gets better with more training data)<\/li>\n
  4. That’s it – Custom Vision Service takes care of the rest!<\/li>\n<\/ol>\n

    <\/h3>\n

    2. TensorFlow Lite<\/h3>\n

    TensorFlow Lite<\/a> is an open source machine learning platform\u00a0that allows us to use TensorFlow<\/a> on IoT and Mobile devices.<\/p>\n

    Both TensoryFlow Lite and TensorFlow are completely open-source on GitHub<\/a>.<\/p>\n

     <\/p>\n

    Implementing Image Classification with Azure + Xamarin.Android<\/h2>\n

     <\/p>\n

    The completed\u00a0Image Classification sample app is available here on GitHub: https:\/\/github.com\/dhindrik\/MushroomDetector\/<\/a><\/p>\n

     <\/p>\n

    1. Training the Model<\/h3>\n

    Using the Custom Vision Service Web Portal<\/a>, we will first train models for image classification.<\/p>\n

      \n
    1. In the Custom Vision Service Web Portal<\/a>, click New Project<\/strong><\/li>\n<\/ol>\n

      \"Image<\/center>2. In the\u00a0Create new project<\/strong> window, make the following selections:<\/p>\n