A brief introduction to ONNX and WinML

To solve the challenges we have outlined in the introduction of this chapter, the ecosystem has introduced the concept of Edge computing. Instead of delegating every operation to the cloud, machine learning models are evaluated directly on a local device, without needing an internet connection. The cloud still plays a critical role, but as a companion; for example, if the recognition rate of an anomaly is under a certain threshold, the application might choose to send the video feed to a cloud service for deeper analysis.

Windows 10 has introduced a platform called WinML to enable any Windows device to become part of the Edge ecosystem. Thanks to this feature, you can evaluate machine learning models directly on your device, using the computing power offered by the CPU and GPU of your machine. WinML exposes a set of APIs that you can integrate into your Windows applications so that you can evaluate a model in your application and process the...