yolo_onnx2engine源码 包含相关参数和解析

Posted 2023-03-02 爱吃油淋鸡的莫何

yolo_onnx2engine源码 包含相关参数和解析

文章目录

• • yolo_onnx2engine源码 包含相关参数和解析
  • • 注释：
    • • CMakeLists.txt文件
      • yolo_onnx2engine.cpp源码中包含相关参数需要记得修改
    • 注：
    • • 1 static const int anchor_output_num这个值的计算方法
      • • (1) 若输入图片尺寸为640
        • (2) 若输入图片尺寸为960
      • 2 yolo_onnx2engine.cpp
      • 3 CMakeLists.txt

注释：

CMakeLists.txt文件

中涉及 cuda 和 TensoRT 的相关路径需要匹配当前系统

yolo_onnx2engine.cpp源码中包含相关参数需要记得修改

如下

40-44行：

static const int INPUT_H = 960;
static const int INPUT_W = 960;
static const int cls_num = 15;
static const unsigned int maxBatchSize = 2;
static const int anchor_output_num = 56700;  

49-50行：代码 onnx 文件的路径（注意路径需是绝对路径，如果是相对路径的话，会有问题）

const char* onnx_path = "/home/onnx2engine/yolov5_5.0/model_onnx/yolov5s.onnx";
const char* TrtSaveFileName = "/home/onnx2engine/yolov5_5.0/model_onnx/yolov5s.engine";

注：

1 static const int anchor_output_num这个值的计算方法

// 不同输入尺寸anchor:640-->25200 | 960-->56700
2^5 = 32 

(1) 若输入图片尺寸为640

640/32 = 20
20   40   80
20和40和80分别对应3个尺寸的anchor框，每个尺寸的anchor框有3个大小
（20*20+40*40+80*80）*3 = 25200

(2) 若输入图片尺寸为960

960/32 = 30
30   60   120
30和60和120分别对应3个尺寸的anchor框，每个尺寸的anchor框有3个大小
（30*30+60*60+120*120）*3 = 56700

2 yolo_onnx2engine.cpp

 yolo_onnx2engine.cpp
 ref1 : https://blog.csdn.net/weixin_43863869/article/details/124614334
 ref2 : https://www.cnblogs.com/tangjunjun/p/16639361.html
#include "NvInfer.h"
#include "cuda_runtime_api.h"
#include <fstream>
#include <iostream>
#include <map>
#include <sstream>
#include <vector>
#include <chrono>
#include <cmath>
#include <cassert>

#include<opencv2/core/core.hpp>
#include<opencv2/highgui/highgui.hpp>
#include <opencv2/opencv.hpp>


// onnx转换头文件
#include "NvOnnxParser.h"
using namespace nvonnxparser;
using namespace nvinfer1;
using namespace std;


#define CHECK(status) \\
    do\\
    \\
        auto ret = (status);\\
        if (ret != 0)\\
        \\
            std::cerr << "Cuda failure: " << ret << std::endl;\\
            abort();\\
        \\
     while (0)


// stuff we know about the network and the input/output blobs
static const int INPUT_H = 960;
static const int INPUT_W = 960;
static const int cls_num = 15;
static const unsigned int maxBatchSize = 2;
static const int anchor_output_num = 56700;   // 不同输入尺寸anchor:640-->25200 | 960-->56700
static const int OUTPUT_SIZE = 1* anchor_output_num *(cls_num+5); //1000 * sizeof(Detection) / sizeof(float) + 1;

const char* INPUT_BLOB_NAME = "images";
const char* OUTPUT_BLOB_NAME = "output";
const char* onnx_path = "/home/mec/hlj/onnx2engine/yolov5_5.0/model_onnx/yolov5s.onnx";
const char* TrtSaveFileName = "/home/mec/hlj/onnx2engine/yolov5_5.0/model_onnx/yolov5s.engine";


//static Logger gLogger;
//构建Logger
class Logger : public ILogger

    void log(Severity severity, const char* msg) noexcept override
    
        // suppress info-level messages
        if (severity <= Severity::kWARNING)
            std::cout << msg << std::endl;
    
 gLogger;


void saveToTrtModel(const char * TrtSaveFileName,IHostMemory*trtModelStream)
	
		std::ofstream out(TrtSaveFileName, std::ios::binary);
		if (!out.is_open())
		
		std::cout << "打开文件失败!" <<std:: endl;
		
		out.write(reinterpret_cast<const char*>(trtModelStream->data()), trtModelStream->size());
		out.close();
	


// Creat the engine using only the API and not any parser.
ICudaEngine* createEngine(unsigned int maxBatchSize, IBuilder* builder, IBuilderConfig* config)

    
    /*等价于*bulider.createNetwork(),通过Ibulider定义的
    名为creatNetwork()方法，创建INetworkDefinition的对象，ntework这个指针指向这个对象*/ 
    INetworkDefinition* network = builder->createNetworkV2(1U); //此处重点1U为OU就有问题

    //创建解析器
    IParser* parser = createParser(*network, gLogger);
    parser->parseFromFile(onnx_path, static_cast<int32_t>(ILogger::Severity::kWARNING));
    //解析有错误将返回
    for (int32_t i = 0; i < parser->getNbErrors(); ++i) 
     
        std::cout << parser->getError(i)->desc() << std::endl; 
    
    std::cout << "successfully parse the onnx model" << std::endl;

    // Build engine
    builder->setMaxBatchSize(maxBatchSize);
    config->setMaxWorkspaceSize(1 << 20);
    //config->setFlag(nvinfer1::BuilderFlag::kFP16); // 设置精度计算
    //config->setFlag(nvinfer1::BuilderFlag::kINT8);

    //通过Ibuilder类的buildCudaEngine()方法创建IcudaEngine对象，
    ICudaEngine* engine = builder->buildEngineWithConfig(*network, *config);
    std::cout << "successfully  convert onnx to  engine！！！ " << std::endl;

    //销毁
    network->destroy();
    parser->destroy();

    return engine;



void APIToModel(unsigned int maxBatchSize, IHostMemory** modelStream)


    // Create builder//创建构建器(即指向Ibuilder类型对象的指针)
    IBuilder* builder = createInferBuilder(gLogger);
    IBuilderConfig* config = builder->createBuilderConfig();

    // Create model to populate the network, then set the outputs and create an engine
    ICudaEngine* engine = createEngine(maxBatchSize, builder, config);

    assert(engine != nullptr);

    // Serialize the engine
    //将引擎序列化，保存到文件中
    (*modelStream) = engine->serialize();

    // Close everything down
    engine->destroy();
    builder->destroy();
    config->destroy();
    saveToTrtModel(TrtSaveFileName, *modelStream);



int get_trtengine() 

    IHostMemory* modelStream nullptr ;

    APIToModel(maxBatchSize, &modelStream);   
    assert(modelStream != nullptr);

    std::ofstream p("./best.engine", std::ios::binary);
    if (!p)
    
        std::cerr << "could not open plan output file" << std::endl;
        return -1;
    
    p.write(reinterpret_cast<const char*>(modelStream->data()), modelStream->size());
    modelStream->destroy();

    return 0;




int main(int argc, char** argv)
 
    // cout<<"argv = "<<argv<<endl;
    // cout<<"argv[1] = "<<argv[1]<<endl;

    get_trtengine();

    return 0;

3 CMakeLists.txt

cmake_minimum_required(VERSION 2.6)

project(yolo)

add_definitions(-std=c++11)

option(CUDA_USE_STATIC_CUDA_RUNTIME OFF)
set(CMAKE_CXX_STANDARD 11)
set(CMAKE_BUILD_TYPE Debug)

include_directories($PROJECT_SOURCE_DIR/include)
# include and link dirs of cuda and tensorrt, you need adapt them if yours are different
# cuda
include_directories(/usr/local/cuda-11.6/include)
link_directories(/usr/local/cuda-11.6/lib64)
# tensorrt
include_directories(/home/package/TensorRT-8.2.5.1/include/)
link_directories(/home/package/TensorRT-8.2.5.1/lib/)

include_directories(/home/package/TensorRT-8.2.5.1/samples/common/)
#link_directories(/home/package/TensorRT-8.2.5.1/lib/stubs/)

# opencv
find_package(OpenCV REQUIRED)
include_directories($OpenCV_INCLUDE_DIRS)
add_executable(yolo $PROJECT_SOURCE_DIR/yolo_onnx2engine.cpp)
target_link_libraries(yolo nvinfer)
target_link_libraries(yolo cudart)
target_link_libraries(yolo $OpenCV_LIBS)
#如果onnx2engine则需要如下库
target_link_libraries(yolo /home/package/TensorRT-8.2.5.1/lib/stubs/libnvonnxparser.so)

add_definitions(-O2 -pthread)