Windows Machine Learning

User avatar
binhlq
Posts: 378
Joined: Thu Apr 16, 2020 3:56 am

ONNX Conversion (ONNXMLTools)

Post by binhlq »

ONNX Conversion (ONNXMLTools)

  • 05/26/2021
  • 2 minutes to read

ONNXMLTools enables you to convert models from different machine learning toolkits into ONNX.

Installation and use instructions are available at the ONNXMLTools GitHub repo.

Support

Currently, the following toolkits are supported.

Pytorch also has a build-in ONNX exporter. Check here for further details.


User avatar
binhlq
Posts: 378
Joined: Thu Apr 16, 2020 3:56 am

Windows Machine Learning samples

Post by binhlq »

Windows Machine Learning samples

  • 10/14/2021
  • 2 minutes to read
    • Image

The Windows-Machine-Learning repository on GitHub contains sample applications that demonstrate how to use Windows Machine Learning, as well as tools that help verify models and troubleshoot issues during development.

Samples

The following sample applications are available on GitHub.

SAMPLES Name Description
AdapterSelection (Win32 C++) A desktop application that demonstrates how to choose a specific device adapter for running your model.
BatchSupport Shows how to bind and evaluate batches of inputs with Windows ML.
Custom Operator Sample (Win32 C++) A desktop application that defines multiple custom CPU operators. One of these is a debug operator that you can integrate into your own workflow.
Custom Tensorization (Win32 C++) Shows how to tensorize an input image by using the Windows ML APIs on both the CPU and GPU.
Custom Vision (UWP C#) Shows how to train an ONNX model in the cloud using Custom Vision, and integrate it into an application with Windows ML.
Emoji8 (UWP C#) Shows how you can use Windows ML to power a fun emotion-detecting application.
FNS Style Transfer (UWP C#) Uses the FNS-Candy style transfer model to re-style images or video streams.
MNIST (UWP C#/C++) Corresponds to Tutorial: Create a Windows Machine Learning UWP application (C#). Start from a basis and work through the tutorial, or run the completed project.
NamedDimensionOverrides Demonstrates how to override named dimensions to concrete values in order to optimize model performance.
PlaneIdentifier (UWP C#, WPF C#) Uses a pre-trained machine learning model, generated using the Custom Vision service on Azure, to detect if the given image contains a specific object: a plane.
RustSqueezeNet Rust projection of WinRT using SqueezeNet.
SqueezeNet Object Detection (Win32 C++, UWP C#/JavaScript, .NET5, .NETCORE) Uses SqueezeNet, a pre-trained machine learning model, to detect the predominant object in an image selected by the user from a file.
SqueezeNet Object Detection (Azure IoT Edge on Windows, C#) This is a sample module showing how to run Windows ML inferencing in an Azure IoT Edge module running on Windows. Images are supplied by a connected camera, inferenced against the SqueezeNet model, and sent to IoT Hub.
StreamFromResource Shows how to take an embedded resource that contains an ONNX model and convert it to a stream that can be passed to the LearningModel constructor.
StyleTransfer (C#) A UWP app which performs style transfer on user-provided input images or web camera streams.
winml_tracker (ROS C++) A ROS (Robot Operating System) node which uses Windows ML to track people (or other objects) in camera frames.
https://docs.microsoft.com/en-us/window ... ml/samples

Post Reply

Return to “AI Technology”