"One rainy afternoon I was struggling to find a couple of demos to show high school students "real-world" applications of Machine Learning image classification & detection. Then, I reached across my desk to get an M&M - image detection, and count the number of M&Ms left in the bowl. Shortly afterward, In the distance I could hear my son playing with his Lego on the living room floor - image classification, which Lego brick had I stood on last night. The Windows 10 IoT Core client can be remotely configured (via Azure IoT Central Device Twin Settings) to either an Azure Cognitive Services image detection or classification model. The client application post-processes the model results and uploads them to Azure IoT Central for presentation. Useful for tracking my M&M consumption and making sure my M&M consumption KPIs were achieved. Training the image detection and classification models was a bit tedious, for a detection model each tag needed at least 15 images so for 6 different colors of M&M I needed at least 90 training images. Every so often the camera image contrast settings would get messed up and the model wasn't able to process the "washed-out" images. I'm looking at trialing robust outdoor cameras (for a student project counting wildfowl on a pond), generating a compact model and running it on the device (to allow the application to run disconnected out of wireless coverage), and initiating the image processing in an Azure function to reduce the amount of data traffic to/from the device."