Microsoft's custom vision website allows a user to create a Docker file consisting of learned machine learning code. By uploading pictures to the website and classifying them, the machine learning algorithms behind the scenes learn what a particular object is. In the following diagram, you can see that a series of different faces are passed to the custom vision website:
The resulting Docker file may then be deployed to the edge device—in this case, a Raspberry Pi running Azure IoT Edge.
The simplicity of the process means that the user does not have to be a data scientist or programmer to create a learned dataset. An excellent tutorial showing Azure custom vision for IoT Edge may be viewed at https://github.com/Azure-Samples/Custom-vision-service-iot-edge-raspberry-pi.