Using deep learning models for image classification directly on the HoloLens 2.
Using the EfficientNetB0
model, trained on the ImageNet
1000 class dataset, for image classification. Model inference is run directly on the HoloLens 2 using its onboard CPU.
(224, 224)
for online inferenceUniversal Windows Platform
, select HoloLens
for target device, and ARM64
as the target platformonnx-models\model.onnx
file to the Builds\HoloLens-2-Machine-Learning\Assets
folderContent
field has its boolean value set to True
. This enables the ONNX
model to be loaded at runtime from the Visual Studio assets folderRelease
mode for ARM64
and deploy to the HoloLens 2 to testpredicted label
, its associated probability
and the inference time
in milliseconds