Experiments for object detection in real-time, using YOLO v8.

Pretrained Yolo models can be found e.g. here. Since most modern Yolo models are implemented in PyTorch, the usual format you find pretrained models in is .pt. With ultralytics python package, you can easily convert these .pt models to e.g. OpenCV DNN compatible .onnx format.

from ultralytics import YOLO

model = YOLO("yolov8m.pt")

model.export(format="onnx", opset=12, imgsz=[640, 640])

The ultralytics python package will also automatically attempt to download a suitable .pt file if not found at the specified path.

For C++, OpenCV is required.
Easiest way to install that is by utilizing vcpkg package manager.
vcpkg install opencv4, for cuda support: vcpkg install opencv4[cuda]
For higher performance it's recommended to install the CUDA version supposing you have an Nvidia GPU so that the model can run on the GPU instead of the CPU.
Visual Studio 2022 project files are included. CMake file is on the way.

Python can be run either with OpenCV or Ultralytics. For OpenCV, you need to install the opencv-python and numpy. For Ultralytics, you need to install the ultralytics package. For the prototypes you can simply run pip install -r requirements.txt in the respective folder as well.

Yolo v8 XL, 30ms inference result on RTX 3070 Ti Laptop, Laptop Camera Feed (real-time)