Before you can make predictions using your trained model via the API, it must first be published. Publishing creates a stable, accessible endpoint that allows your Python code to send image data to the cloud-hosted model and retrieve predictions in return.
This step ensures that the model version you are calling is fixed and identifiable, making it suitable for repeated use, evaluation or deployment in live workflows.
Publishing Your Model
Under the Performance heading, select the iteration of trained model you wish to access.
Click the Publish button to make your model available via the API.
Once published, click Prediction URL, and a dialogue with relevant details will appear.
API Integration Options
If you are using an Image URL - where your images are hosted remotely online - you will need the Endpoint (POST request) URL given. Custom Vision provides you with:
If you are using a local image (either on your machine, or uploaded to a workspace in Azure ML Studio), the information is the same as above, but this will run your Custom Vision model on a local image rather than a remotely-hosted image.