Create Object Detection Model inside the Power Platform | Power Apps — AI Builder

Amit Prajapati
5 min readApr 25, 2020

In this blog, we are going to see how to create object detection model which can be used in PowerApps or MS Flow / PowerAutomate

Step 1: Log in to

Select the PowerApps, If PowerApps is not visible then click on All Apps then you will able to see the PowerApps.

Office Portal

Step 2: Expand the AI Builder Section and click on Build Section in PowerApps.

Note: Please ensure that you are select the correct instance.

Power Apps | AI Builder

Step 3: Click on the Object Detection Model

We are going to create the Object Detection Model which can be used to created PowerApps or In MS Flow / Automate.

Build Model — Object Detection

Step 4: Name the AI Model and click on Create.

Create a Fruit Detection Model

Step 5: We will select the Domain of Model so we will go with Common Object. And click on Next.

Select the Domain of Model

Step 6: Before moving forward, we will download the data which will be used to train and test the Model. Kaggle is the best source for the data to train the machine learning model.

We will require the fruit images as we are designing the Fruit Detection Model, search fruit images and download the Dataset given in the screenshot.

Kaggle — the source of Dataset
Search Fruit Images

Select the Fruit Image for Object Detection

Fruit Images for Object Detection

Click on the Download

Download — Fruit Images

Step 7: After downloading the Zip file extract it. You will see the following two folders — train_zip and test_zip respectively.

Downloaded Fruit Image Zip File

Now, we will open the train_zip folder and you can see that there will be four categories of images

  • Apple
  • Banana
  • Orange
  • Mixed [Apple, Banana, Orange]
Fruit Images under the Train Set

Step 8: Let move back to the PowerApps Platform, 2nd step in the creation of Object Model is to define the object that we are going to detect. Here, we have three objects — Apple, Banana & Orange and click on next to move further steps.

Choose Object for your model to detect
Add Images

Step 9: We will require a minimum 15 images of each category to train our Object Detection Model.

Now, we will click on Add Images and select Upload from local storage.

Upload images from the local storage

Step 10: Select the images. Once all images are upload click on Close.

Click on Next

Step 11: Now, we are in the most important phase of the training where we provide the Tag or Label to the Images which we have uploaded.

Image Tagging Phase

Click on the uploaded image and select the area where the object is present. Once you will select the area you will get the option to select the object is that present in the selection area.

Tagged Image by Selecting the areas on Image

If an image is not suitable for the model to remove that particular image click on the “Don’t use Image ”. Click on remove to remove the image from the model data set.

Remove images if not suitable for the training model

Step 12: Once you are done will the tagging or labeling the image click on the Done tagging.

Note: Please ensure that you have a more tagged image for each model so that your model will work accurately. The more the label data more the accurate your machine learning model.

Step 13: Click on the Train to start the training your model based on tagged images.

Click Train to train the model

Click on Go to Model

Go to model

It will take around 5–10 min to train based on your complexity of images, model and objects.

Step 14: Publish the model.

Publish Trained Model

Once you model finished with done and you have published the Model. You can use that model to in Power App or MS Flow. We are going to see that in the next part of this blog.



Amit Prajapati

I am Technical Consultant from Mumbai, India. I like to contribute my knowledge to the community and good social cause.