IceVision → OpenVINO → Unity Tutorial

⚠️ ARCHIVED REPOSITORY ⚠️

This tutorial code will no longer receive updates as the IceVision library no longer appears in development. For the latest and improved version of this tutorial that uses PyTorch directly, please visit the new tutorial repositories linked below:

• pytorch-yolox-object-detection-tutorial-code
• barracuda-inference-yolox-demo

---

Tutorial Links

• Part 1: Train a YOLOX model using IceVision and export it to OpenVINO.
• Part 2: Create a dynamic link library (DLL) file in Visual Studio to perform object detection with a YOLOX model using OpenVINO.
• Part 3: Perform object detection in a Unity project with OpenVINO.
• Follow up: Use ONNX Runtime and DirectML instead of OpenVINO.

Demo Video

HaGRID_demo.mp4

Training Code

Jupyter Notebook	Colab	Kaggle
GitHub Repository

Note: Training on the free GPU tier for Google Colab takes approximately 11 minutes per epoch, while training on the free GPU tier for Kaggle Notebooks takes around 15 minutes per epoch.

Kaggle Datasets

• HaGRID Sample 30k 384p
• HaGRID Sample 120k 384p

Reference Images

Class	Image
call
dislike
fist
four
like
mute
ok
one
palm
peace
peace_inverted
rock
stop
stop_inverted
three
three2
two_up
two_up_inverted