Hack the North 2019 - Machine Learning Workshop

Microsoft

Overview

In this workshop, you will build an image classifier with Azure Custom Vision Service and export it as a TensorFlow model. You will learn how to import the model into a Python application and use it to classify images.

Part 1 - Train and export the TensorFlow model

Follow the following instructions to create a Custom Vision project. You can use the training images linked to in the tutorial to build a cats vs dogs image classifier.

Tutorial: Train and export a TensorFlow model with Azure Custom Vision Service

If you are feeling adventurous or if you have another idea in mind, you can download and use your own images to train a model that recognizes other objects.

Part 2 - Import and use the model in Python

1. If you haven't cloned this repo already, clone it to your local machine.

2. In a terminal, browse to the app folder.

3. Create and activate Python 3 virtual environment by entering:

   python3.7 -m venv .venv
   source .venv/bin/activate
   # windows:
   # .venv\scripts\activate
   

4. Copy the model.pb and labels.txt that you downloaded from Azure Custom Vision Service into model folder or use the ones that are already there.

5. Run the script and a prediction should appear:

   python run.py
   

Resources

Azure Custom Vision Service can export models that run on many platforms.

• iOS and Android
• Xamarin (cross-platform)
• Serverless HTTP API (Python)
• You can also export it as a Docker container (Dockerfile)

Also check out these other resources:

• Object detection with Custom Vision
• All Azure Cognitive Services