Creating Models

Step 1. Log onto the Custom Vision portal with your Azure credentials.

Note: If you don’t have an Azure subscription, you can opt for a free Azure for Students version.

Step 2. Hit New Project on the portal and fill in the specifications for the Custom Vision Project you wish to create.

Name: Name of your choice
Description: Valid description for your project
Resource: Resource Group created beforehand
Project type: Classification
Classification types: Multiclass (single tag per image)
Domains: General [A2]

Step 3. Hit Create Project

You would now be led to the Training Images pane on the custom vision portal.

Step 4. Click on Add Images to select images you wish to train your model on, and then select appropriate tags or ‘labels’ for your images. In our example, only one label/ class can be associated with one given image.

If you are working on binary classification (only 2 classes), the Negative for a tag would appear as a default option. 

Step 5. Click on Upload all ___ Images once you are done. 

Step 6. Once the images have been tagged /labeled and uploaded, it is time to train your model based on the images.

Step 7. Hit the Train button on the top-right corner of the screen and select Quick Training. Click Train.

Step 8. Once the training is done (after 15-20 minutes tentatively), you can take a look at the performance metrics for the model by navigating to the Performance tab. You can analyze the Precision, Recall, and Average Precision (AP) metrics. 

Step 9. If you are unhappy with your model’s performance, you can train it again. Once you are satisfied with the metrics, you can test its performance on sample images by navigating to the Quick test button. 

Step 10. Upload the image file or paste the image URL in the Quick test box and get the prediction results.

This is how you can train a custom vision model based on your own images and test its performance on a number of images. The larger the image dataset used to train the model, the better its performance.

Azure Custom Vision is a computer vision service offered by Microsoft Azure that allows users to create customized computer vision models for image classification and object detection. It is a no-code interface that simplifies the process of making computer vision models without code-based training. Using a very user-friendly UI, an end-to-end image classification/ object detection solution could be developed from scratch without writing a single line of code.  

In this article, we are going to discover how we can create resilient Custom Vision models once we have created a Custom Vision resource.