Using GitHub Actions to Automate Image Classification with Microsoft Azure Custom Vision Services

Setting up GitHub Actions to automate the process of image-classification using Azure.

Posted by : Mritunjay Sharma at Sep 18, 2020

Category : GitHub-Actions, Automating, Python

My Workflow

I created this new GitHub Action available here with which we can automate the entire process of image classification with Azure Custom Vision. With this - all a user needs to do is:

Add the 4 SECRETS which are basically Azure Custom Vision Services Credentials. The process is explained in the README
Upload Image Dataset in the GitHub Repo and modify the tags, tagsVar and trainSize accordingly in the below yml file.

Here’s a sample workflow YML file using this Action:

name: Auto Custom Vision Classifier
on:
  push:
    paths:
    - '**.yml'
jobs:
  build_model:
    runs-on: ubuntu-latest
    steps:
    - name: Automating Image Classification with Microsoft Azure Custom Vision Training and Prediction
      uses: mritunjaysharma394/autoCustomVision@v1.0
      with:
        tags: "[Hemlock,Japanese Cherry]" # Rename it according to the folder name under images/ which will also be our name to the tags
        tagsVar: "[hemlock_,japanese_cherry_]" # Rename it according to the symmetry of file names under images/tag/ 
        trainSize: "10" # Train Size of Each Tag
        endpoint: $
        trainingKey: $
        predictionKey: $
        predictionResourceid: $

Using GitHub Actions to Automate Image Classification with Microsoft Azure Custom Vision Services

Setting up GitHub Actions to automate the process of image-classification using Azure.

My Workflow

Working Video Example: