In the previous recipe, we implemented an object classifier using GPU and an NVIDIA Jetson Nano. There are other types of GPU-enabled devices. These range from the NVIDIA TX2, which can be put on a drone to do real-time analysis of pipelines, to industrial PCs running GPUs and using computer vision to perform analyses on workplace safety. In this recipe, we are going to train and add to an existing image classification model by adding our own images to it.

Challenges that the IoT faces include over-the-air (OTA) updates and fleet management. IoT Edge is a conceptual framework that solves this. In OTA updates, Docker containers are used as an update mechanism. The underlying systems can be updated without having to worry about complete device failure. If an update does not work, the system can be rolled back because container failures do not affect the main OS and the Docker daemon can perform the update and roll back.

In this recipe, we are going...