AI Builder - Creating a Custom Image Recognition Model
Science & Technology
Introduction
In this article, we will explore how to set up a model for Power Automate's AI Builder, specifically focusing on creating an image recognition model. Follow along as we step through the process of detecting custom objects in images.
Getting Started with AI Builder
Once logged into Power Automate, navigate to the AI Builder section in the left menu and select "Models." Here, you’ll find a list of your existing models. To create a new model, click on "Build a model," selecting the option to detect custom objects in images.
AI Builder provides a wizard-like interface that guides you through each step. It helps define what you wish to identify and provides training recommendations based on the inputs and examples you provide. Click the "Get Started" button to begin.
Defining Your Model
For image recognition, you can define multiple objects to search for within your images. It’s important to note that a minimum of 15 examples per object is required for effective training. The wizard will prompt you to specify the domain; for this example, we will leave it set to "Common Objects."
Once in the object definition step, specify which objects you are identifying. For instance, if you are working with tabletop gaming dice, you could define eight different objects to detect various sizes.
If preferred, you can alternatively connect a database to retrieve this information. After defining the objects, click next.
Preparing Images for Training
You will need to upload images that contain examples of each object. These images should encompass different angles, backgrounds, lighting conditions, and colors. The good news is that you can use a single image that contains multiple objects, as long as the objects are distinct.
To add images, click "Add images" and choose your data source—local device, SharePoint, or Blob storage. After selecting your images, click upload. Keep in mind that images must be in JPEG, PNG, or BMP format and cannot exceed 6 MB in size.
AI Builder recommends using a minimum of 50 images per object for the best results, but you can initially start with the minimum requirement of 15 images.
Once the images are uploaded, you can click "Done" to proceed. Note that you can add more images at any time, making the process flexible.
Tagging Your Images
Next, you will need to identify the objects within each of your uploaded images. By dragging a box over each object, you will label it accordingly. As you progress, you’ll see a tally indicating how many images you have tagged for each object, helping you keep track of your progress.
If you misidentify an object, or wish to remove an image from your list, options to edit or delete are readily available. When you’re finished tagging, click "Done tagging."
Training Your Model
With your images tagged, you’re ready to proceed to the training phase. Click "Next" to see a summary of the objects and images you’ve prepared. Before training, ensure you’ve named your model appropriately.
Hit the "Train" button and allow some time for the model to process the images. Depending on the volume of images and objects, training can take a few minutes to several hours. Once complete, your model will be marked as "Trained."
However, you must publish the model before utilizing it in applications. Navigate back to the model's detail page and select "Publish." After publishing, your model is ready for deployment.
Testing the Model
It’s best practice to test your model with new images not included in the training set. Upload your test images and observe how the model performs in identifying the objects. Remember that the model's accuracy will be directly related to how well it was trained with diverse examples.
Once tested, you can integrate the model into a Power Automate flow, making it accessible via mobile or a web interface. The default output will showcase a JSON object listing the detected objects, their confidence scores, and their coordinates within the image.
The foundation of the application can grew from these initial capabilities, allowing for further enhancements and integrations.
Conclusion
AI Builder serves as an excellent starting point for developing custom image recognition models. With careful preparation and thorough testing, you can leverage its power to create sophisticated applications.
Keyword
- AI Builder
- Image Recognition
- Custom Objects
- Training Model
- Power Automate
- Data Source
- Tagging Images
- Publish Model
- Test Model
- Deployment
FAQ
Q1: What is AI Builder?
A1: AI Builder is a feature within Power Automate that allows users to create and deploy custom AI models, including image recognition models.
Q2: How many example images are needed for each object?
A2: A minimum of 15 images is required for effective training, but it is recommended to use at least 50 images per object for better accuracy.
Q3: Can I upload images from different sources?
A3: Yes, images can be uploaded from local devices, SharePoint, or Blob storage.
Q4: How long does model training take?
A4: Model training duration may vary from a few minutes to several hours, depending on the number of images and objects included.
Q5: How do I test my trained model?
A5: You can test your model by uploading new images not used in training, and observing how many objects it can correctly identify.
