YOLOv8 deployment

the project code heavily borrowed from: https://github.com/UNeedCryDear/yolov8-opencv-onnxruntime-cpp and https://github.com/triple-Mu/ncnn-examples This is the C++ deployment implementation of yolov8, using OpenCV, the df module of yolov8 detection branch in the post-processing is implemented in the post-processing, which can be deployed on the Horizon X3 board. To train or convert the yolov8 model, you can refer to the implementation of triple-mu with branch triplemu/x3pi

Implemented

• detection
• instance segmentation
• pose estimation

requirements: opencv-dnn:

OpenCV >= 4.5.5

build and run

• build:

  mkdir build
  cd build && cmake .. && make -j8
  

• run

  ./build/YOLOv8 seg_dfl
  

When writing post-processing, always be careful to use netron to view the model structure

• seg without df branch, c++ details to see yolov8_seg_dfl.cpp

• det without df branch, c++ details to see yolov8_det_dfl.cpp

• pose without df branch, c++ details to see yolov8_pose_dfl.cpp

• df branch in det model