Recently we announced that Windows 10 now includes the ability to run Open Neural Network Exchange (ONNX) models natively with hardware acceleration .
This announcement now brings 100s of millions of Windows devices, ranging from IoT edge devices to HoloLens to 2-in-1s and desktop PCs, into the ONNX ecosystem.
Yesterday we released the Windows 10 RS4 preview to HoloLens so this now allows Data scientists and developers creating AI models will be able to deploy their innovations to this large user base. From an academic perspective I have lots of HoloLens developers wanting to building apps that are use AI models offline to deliver more powerful and engaging experiences.
So what is ONNX?
ONNX is an open source model representation for interoperability and innovation in the AI ecosystem.
Microsoft actually helped start ONNX in September 2017, and with support from many other companies , ONNX v1 launched in December 2017.
Thanks to ONNX-ML, Windows supports both classic machine learning and deep learning, enabling a spectrum of AI models and scenarios. Developers can obtain ONNX models to include in their apps in several ways:
-
Create and train ONNX models using any of the many
frameworks
that currently support generating ONNX models.
-
Convert models from various sources including SciKit-Learn and CoreML into ONNX models by using
ONNXML tools
.
- Obtain pre-trained ONNX models from the growing ONNX model zoo .
You can learn more about how to integrate ONNX models into Windows apps here .
If your interested in adding ONNX to your HoloLens Apps see my colleagues Mike Taulty blogs
First Experiment with Image Classification on Windows ML from UWP
Second Experiment with Image Classification on Windows ML from UWP (on HoloLens)
Third Experiment with Image Classification on Windows ML from UWP (on HoloLens in Unity)