/hl2_detection

Apache License 2.0Apache-2.0

hl2_detection

Configure a project

Steps to run app on HoloLens 2:

  1. Clone the repository
  2. Checkout to alpha branch. This branch uses a Barracuda library, which will be replaced with Sentis.
  3. Open Unity Hub, then Open->Add project from disk and open a MachineLearningPlayground directory inside this repository.
  4. There should a MRTK Project Configurator window appear. Select Unity OpenXR plugin (recommended) option, the project will configure.
  5. The Settings window will appear. You can close it, same to a MRTK Project Configurator window.
  6. In a Unity Editor topbar select File->Build Settings... . In a Platform select a Universal Windows Platform, on the right change architecture to ARM 64 and click on Switch platform button on the bottom. Wait for a project to reconfigure.
  7. In a Unity Editor in a Project Explorer view (bottom) select the Assets, and then Scenes directory. Open SampleScene.unity file. The scene will load.
  8. From a Hierarchy panel select MixedRealityPlayspace->Main Camera object. In a bottom of Inspector panel (right side of editor) there is a Detection script. The model property is missing, since you have to provide a model by yourself. Follow the next steps to prepare a YOLOv8 model.

Prepare YOLOv8 model

Install the ultralytics repo according to manual in: https://github.com/ultralytics/ultralytics . Then, export a weights of YOLOv8 Nano model to .onnx format with opset level of 9. In the following example we use an image input size of 160 by 160 pixels.

Export script:

yolo export model=yolov8n.pt format=onnx opset=9 imgsz=160

As a result, the yolov8n.onnx was generated in a working directory.

You can also use this notebook to export and download a model.

When exported, copy a .onnx file to a Assets/Weights* directory.

Build and deploy on HL2

  1. Drag and drop a exported neural network weights file to a Model property in Unity Editor.
  2. In top bar select File->Build Settings->Build. Create a folder to place a build files and start a build process.
  3. After finished, open a build directory, open a .sln file with a project, make sure to pick a proper build configuration, architecture and device, then build and deploy solution.

Versions

We have used a following versions of software:

  1. Unity Editor 2021.3.20f1
  2. Microsoft Visual Studio 2022 17.5.1
  3. Barracuda 3.0.0

Dependencies

This project uses a https://github.com/EnoxSoftware/HoloLensCameraStream plugin to gather camera feed and calculate a pose of bounding boxes in a 3D space.