Custom Vision

Object Detection AI Vision

Now that you have created your object detection project, it's time to upload training images and create bounding boxes around the objects you want to detect.

Step 1: Upload Training Images

As you have done in the classification steps, click Add Images, and upload the training images from the fruitbowl_dataset/Training folder on GitHub. Make sure to select additional images from the /Object_Detection subfolder.

Note
For training a Custom Vision object detection model, you will require at least 15 images, with at least 15 instances of each tag present across all images in the training dataset.

Step 2: Create Bounding Boxes and Tags

Begin tagging these images one at a time, drawing bounding boxes over each type of fruit in the image. The tags you should create should be:

  • banana
  • green_apple
  • red_apple
  • orange
  • basket
Fruit basket image with bounding boxes drawn around each fruit and object tag panel

Drawing individual bounding boxes around each fruit and applying appropriate tags

Note
It's important to draw individual bounding boxes around each visible fruit when tagging the training images. Avoid drawing a single box around a cluster of items if you can, as this may lead the model to misclassify overlapping or partially obscured fruits.
Basket of oranges with individual bounding boxes drawn around each orange

Example of proper individual bounding box placement around each orange

Step 3: Verify Tag Counts

Check that you have at least 15 instances of each tag by filtering by Tagged:

Custom Vision interface showing tagged images with tag counts displayed in the sidebar

Custom Vision interface showing tag counts - ensure you have at least 15 instances of each tag