AI-102 Question 153
Select 2You have developed a custom vision model using Azure Custom Vision to classify images of different types of fruits. You want to make the model accessible via an API for integration with a mobile application. Which steps must you take to publish the model and make it available for use?
- A
Train the model and set it to the 'Published' state using the Custom Vision portal or SDK.
- B
Export the model to a supported format, such as ONNX, and upload it to Azure Blob Storage.
- C
Obtain the prediction endpoint URL and prediction key from the Custom Vision portal.
- D
Enable the 'Always On' setting in the App Service used to deploy the model.
- E
Deploy the model to an edge device using Azure IoT Edge modules.
Show answer and explanation
Correct answers: A, C
Explanation
To publish a custom vision model and make it accessible via an API, you must first train the model and set its state to 'Published' in the Custom Vision portal or through the SDK. Once published, you can retrieve the prediction endpoint URL and prediction key to use the model in integrations, such as a mobile application. Exporting the model or deploying it to an edge device is not required for API accessibility, and enabling 'Always On' is relevant for maintaining App Service uptime but not a publishing requirement.
- A. Correct.
Correct. To make a custom vision model accessible, you must first train it and publish it via the Custom Vision portal or SDK, which makes it available for use through an API.
- B. Incorrect.
Incorrect. Exporting the model is not required for API accessibility; it is only necessary if you plan to use the model offline or on a local device.
- C. Correct.
Correct. After publishing the model, you need the prediction endpoint URL and key to interact with the model using the API.
- D. Incorrect.
Incorrect. While the 'Always On' setting ensures the App Service remains active, it is not a required step for publishing the model in the context of Custom Vision.
- E. Incorrect.
Incorrect. Deploying to an edge device is an optional step if you want to use the model offline or in a constrained environment, but it is not necessary to make the model accessible via an API.