Skip to content

Latest commit

 

History

History
110 lines (63 loc) · 4.37 KB

train-custom-vision-model.md

File metadata and controls

110 lines (63 loc) · 4.37 KB
Raw

Train and export a TensorFlow model with Azure Custom Vision Service

To build an image classifier, you need to create an Azure Custom Vision Service project and provide training images. For more information about the steps that you take in this section, see How to build a classifier with Custom Vision.

Once your image classifier is trained, you can export it as a TensorFlow model that you can run in your Python Azure Functions.

Prerequisites

  • Microsoft Azure subscription - Although Custom Vision has a free tier, it must be accessed using an Azure subscription. If you need an Azure subscription, sign up for a free trial.

Create a Custom Vision account

  1. In your web browser, navigate and log in to the Azure portal.

  2. Select the Create a resource button (green +).

  3. Search for and select Custom Vision.

  4. In the Create screen, enter the following values.

    Field Value
    Name Provide a name for your Custom Vision account.
    Subscription Subscription in which to create the account.
    Location Select a region close to you.
    Prediction Pricing Tier Select F0 for the free tier. (may not be available if you already have another Custom Vision account)
    Training Pricing Tier Select F0 for the free tier. (may not be available if you already have another Custom Vision account)
    Resource Group Create or select a resource group.

  5. Select Create to create the account.

    Create Custom Vision account

Download and extract training images

You will use the Kaggle Cats and Dogs dataset to train the image classifier.

  1. Download a zip archive of the dataset from Microsoft at this web page.

  2. Extract the zip archive to a folder.

  3. In the PetImages folder, you will find a folder of cat photos, and another folder of dog photos.

Create a new Custom Vision project

  1. In your web browser, navigate to Azure Custom Vision Service.

  2. Select Sign in and sign in with the same account that you used to create the Custom Vision account.

  3. Select New project.

  4. Create your project with the following values:

    Field Value
    Name Provide a name for your project, like Dog or Cat.
    Description Optional project description.
    Resource Group Select the Custom Vision account you created.
    Project Types Classification
    Classification Types Multiclass (single tag per image)
    Domains General (compact)
    Export Capabilities Basic platforms (Tensorflow, CoreML, ONNX, ...)

  5. Select Create project.

Train the model

  1. In the new Custom Vision project, select Add images.

  2. Browse to where you extracted the training images and select approximately 100 image files in the Cat folder.

  3. Tag the images with the value cat.

  4. Select Upload images.

    Tag cat images

  5. Select Add images and select another 100 photos from the Dog folder. Tag them dog.

  6. Select Train and choose Fast Training as the training type.

  7. Select Train and wait for the training to complete.

To train an even better model, you can upload more images. You can also upload images that contain neither a dog nor a cat and apply the built-in Negative tag.

Test the model

  1. Select Quick Test.

  2. Provide a URL of an image containing a cat or a dog.

  3. Submit it to see the result.

    Quick test

Export the model

  1. Select the Performance tab.

  2. Select Export.

  3. Select TensorFlow.

    Exporting a TensorFlow model

  4. Select Export. Wait for the export process to complete.

  5. Select Download to download the model.

The download is a zip archive containing two files: model.pb and labels.txt. Copy these files into your Azure Functions project.