This project implements the YOLO (You Only Look Once) detection model made by Ultralytics in Unity 6. It uses the YOLO IX model—with plans to incorporate newer versions—and leverages the Unity Sentis package for ONNX model manipulation. Through GPU acceleration via ComputeShaders, the application enables real-time object detection from images, videos, or camera feeds with a configurable confidence and IoU (Intersection over Unon) thresholds. The recognized objects are limited to those included in the COCO (Common Objects in Context) dataset.
My motivation for creating this application stems from a previous job where we considered integrating real-time object detection from a HoloLens 2 camera stream. Unfortunately, short deadlines prevented that feature from becoming a reality. However, my fascination with the concept led me to develop this project on my own.