YOLOv8 support

[tv-gc]

Hi!

I would like to ask if OpenCVForUnity will support YOLO version 8 and if so, when will the example be shared?

Cheers!

[dragan07-07]

it is already supported!

[tv-gc]

Oh really? I thought you had to make predictors and detectors for it because YoloV5 is on a different repository.

Is there references on how to use v8 that you could share?

Thanks!

[EnoxSoftware]

Sorry for the delay in responding. It seems that opencv's dnn module is capable of inferring yolov8 models in onnx format. https://github.com/opencv/opencv/pull/24611

We have a YOLOv5 example repository on our OpenCVForUnity GitHub, but that example is not applicable to yolov8 models. The data structure of the inference results is different between yolov5 and yolov8, so you need to implement yolov8-specific code. We have created a new repository containing examples of yolov8 object detection models. Please try it out. https://github.com/EnoxSoftware/YOLOv8WithOpenCVForUnityExample

[EnoxSoftware]

Examples of Segmentation and Classification models have also been added!

[dragan07-07]

Tnx it is very useful!

[wimvanhenden-tool]

It seems when using CUDA backend with dnn, the returned bounding boxes are all 0. Using inference on CPU works fine. That seems to be an OpenCV 4.2.8 problem.

[EnoxSoftware]

@wimvanhenden-tool This problem seems to have been reported in the opencv repository. https://github.com/opencv/opencv/issues/23977

[namas191297]

You can use the onnx models on my repository: https://github.com/namas191297/yolov8-segmentation-end2end-onnxruntime

They contain ONNX models for segmentation and bounding boxes for YOLOv8 without the need of any post-processing or YOLOv8 related code