OpenVINO inference for yolov5 in cpp.
- openvino 2021.4
- opencv
You can follow https://docs.openvino.ai/2021.4/openvino_docs_install_guides_installing_openvino_linux.html to install the above two libraries.
cd yolov5
python3 export.py --weights yolov5s.pt --img 384 640 --train --simplify --include onnx openvino
option:
--img: openvino network inference image size(height, width)
You can adjust --img as small as possible and keep it a multiple of 16, so as to decrease computation and increase inference speed. For example, the height and width of origin image are 720 and 1280. If --img is set to 384 640, the image will be resized to 360*640 and then enlarged to 384*640. If --img is set to 640 640, the image will be resized to 360*640 and then enlarged to 640*640. The first one is prefered because of its fast inference speed.
first set up environment
source /opt/intel/openvino_2021/bin/setupvars.sh
source /opt/intel/openvino_2021/opencv/setupvars.sh
git clone https://github.com/wangziren1/yolov5-OpenVINO.git
cd yolov5-OpenVINO
mkdir build && cd build
cmake ..
make
./demo
The result.jpg will be saved in build directory.