The AMD WinML (vx_winml) is an OpenVX module that implements a mechanism to access WinML functionality as OpenVX kernels. These kernels can be accessed from within OpenVX framework using OpenVX API call vxLoadKernels(context, "vx_winml").
The WinML extension will allow developers to import a pre-trained ONNX model into an OpenVX graph and add hundreds of different pre & post-processing vision / generic / user-defined functions, available in OpenVX and OpenCV interop, to the input and output of the neural net model. This will allow developers to build an end to end application for inference.
The following is a list of WinML functions that have been included in the vx_winml module.
onnxToMivisionX com.winml.onnx_to_mivisionx
convertImageToTensor com.winml.convert_image_to_tensor
getTopKLabels com.winml.get_top_k_labels
NOTE: For the list of OpenVX API calls for WinML-interop refer include/vx_ext_winml.h
- Windows 10, version
1809or later - Windows SDK, build
17763or later - Visual Studio 2017, version
15.7.4or later- Visual Studio extension for C++/WinRT
- Install the latest AMD drivers
- Install OpenCL SDK
- Install OpenCV 3.4
- Set
OpenCV_DIRenvironment variable toOpenCV/buildfolder - Add
%OpenCV_DIR%\x64\vc14\binor%OpenCV_DIR%\x64\vc15\binto yourPATH
- Set
- Use
amd_openvx_extensions/amd_winml.slnto build for x64 platform
This utility can be used to test and verify the ONNX model on the Windows platform. If the ONNX model is supported by this utility, the amd_winml extension can import the ONNX model and add other OpenVX nodes for pre & post-processing in a single OpenVX graph to run efficient inference.
NOTE: Samples are available
Samples to run inference on a single image and a live camera is provided in the samples folder.