Object Detection with Azure Custom Vision
Pluralsight Hands-On Lab — AI & Containers
At a Glance
| Platform | Pluralsight |
|---|
| Category | Azure AI Services |
|---|
| Lab Type | Guided + Challenge Mode |
|---|
| Environment | Azure Portal, Azure Custom Vision Portal |
|---|
| Completed | 2026 |
|---|
Overview
Azure Custom Vision enables training of custom image classification and object detection models without requiring deep machine learning expertise. In this lab, I acted as an AI engineer for a supermarket chain and built an object detection model to identify bell peppers and carrots in images for inventory sorting. I uploaded and manually tagged training images in the Custom Vision portal, trained the model using Quick Training, and tested it against new images — reviewing predicted tags, bounding boxes, and probability scores at adjustable confidence thresholds.
What I Did
- Signed in to the Azure Custom Vision portal using existing Azure portal credentials and accepted the Terms of Service
- Downloaded the Foods.zip training and testing image set from the lab's GitHub repository and extracted the files locally
- Created a new Object Detection project named "Food Detection" using the General (A1) domain
- Uploaded all bell pepper training images, then manually tagged each pepper in every image by hovering to generate a bounding box and entering the tag "bell pepper"
- Uploaded all carrot training images and tagged each carrot in every image with the tag "carrot", using the Tagged/Untagged toggle to track progress
- Clicked Train and selected Quick Training, then waited for the model to finish training
- Used Quick Test to submit two images from the Testing Images folder, reviewing the detected objects highlighted with bounding boxes and their predicted tags and probability scores in the Predictions panel
- Adjusted the Predicted Object Threshold slider on both test images to observe how confidence thresholds filter detections, demonstrating the trade-off between precision and recall
← Back to Pluralsight Labs