0xSofea / OpenVINO_CPP_YOLOX-YOLOV5-YOLOV8-YOLOV9

C++ inference of YOLOX/YOLOV5/YOLOV8/YOLOV9 for OpenVINO,Support float32、float16 and int8.

YOLOX-YOLOV5-YOLOV8-YOLOV9-OpenVINO in C++

This tutorial includes a C++ inference of YOLOX/YOLOV5/YOLOV8/YOLOV9 for OpenVINO.

Install OpenVINO Toolkit

Please visit Openvino Homepage for more details.

Environment

OpenVINO_2022.3.0
OpenCV-4.6.0

Convert model

1. Export ONNX model

2. Convert ONNX to OpenVINO

   Install requirements for convert tool

   pip install openvino-dev

   Convert ONNX into the OpenVINO IR

   FP32

   mo -m <onnx_model_path>  --output_dir <MODEL_DIR> 

   FP16

   mo -m <onnx_model_path>  --output_dir <MODEL_DIR> --compress_to_fp16

   INT8 Quantization with POT

   pot -q default -m <ir_model_xml> -w <ir_model_bin> --engine simplified --data-source <data_dir> --output-dir <output_dir_name> --direct-dump --name <int8_model_name>

   visit Openvino POT for more details.

Build

Linux

mkdir build
cd build
cmake ..
make

Demo

c++

./detect  <...><...> 

scrennshot