Python Computer Vision Model in Azure Functions

A python based computer vision model hosted in Azure Functions.

The python codebase is created from the Make machine learning predictions with TensorFlow and Azure Functions.

• 'app': Contains the python codebase
• 'client': Sample app to invoke the local running azure function.
• 'IaC': Contains ARM deployment template
• 'azure-pipelines.yml': Package and release pipeline
• 'Enable-LongPathSupport.ps1': Enables long path support on windows, as described in the reference article.

Deploy azure resources

Run local

On a windows pc with conda installed:

cd __app__
conda create -n venv python=3.7
activate venv
pip install --no-cache-dir -r requirements.txt
func start

Use the following test urls:

• http://localhost:7071/api/classify?img=https://raw.githubusercontent.com/Azure-Samples/functions-python-tensorflow-tutorial/master/resources/assets/samples/cat1.png
• http://localhost:7071/api/classify?img=https://raw.githubusercontent.com/Azure-Samples/functions-python-tensorflow-tutorial/master/resources/assets/samples/cat2.png
• http://localhost:7071/api/classify?img=https://raw.githubusercontent.com/Azure-Samples/functions-python-tensorflow-tutorial/master/resources/assets/samples/dog1.png
• http://localhost:7071/api/classify?img=https://raw.githubusercontent.com/Azure-Samples/functions-python-tensorflow-tutorial/master/resources/assets/samples/dog2.png

Optionally start the client and use the same github urls.

Create Azure DevOps pipeline

Register an Azure DevOps pipeline using the azure-pipelines.yml in the root of this repo:

Specify the following variables:

• azureSubscription: The Azure DevOps Resource Manager connection to use for deployment.
• functionAppName: The Azure Function App to deploy the model to.
• vmImageName: Use ubuntu-latest or windows-latest.

Test the model is running

In order to test if the model is working fine, install pytest

pip install pytest

and then navigate in the __app__ folder and run pytest:

cd __app__
pytest