To use Microsoft's custom vision service, you will require an Azure subscription. Then you will need to spin up a new custom vision project. There is a free tier for testing out small models and a paid tier for larger models and serving models at scale. After creating the custom vision project in the Azure portal, you will see two new projects in the resource group. The first will be for training, and the second will have a -prediction label appended to the name, which will be used for the predictions. 

Then you will require images of what you are classifying. In our case, we are identifying beverages in an environment with lead and carcinogen exposure. If you have completed the previous recipe, you will have a camera capturing images at 1 second intervals. To make an object detection model in cognitive services, you will need at least 30 images of each thing you are trying to classify. More images will improve accuracy. To get good accuracy, you...