C++ TensorRT yolov8

目录

效果

项目

代码

下载

效果

C++ TensorRT yolov8

项目

电脑环境

处理器：AMD Ryzen 7 7735H with Radeon Graphics 3.20 GHz
内存：16.0 GB 
显卡：NVIDIA GeForce RTX 4060 Laptop GPU
操作系统：Windows 10 企业版
opencv-4.8.1
CUDA12.4
TensorRT-8.6.1.6
VS2022

包含目录

库目录

附件依赖项

cublas.lib
cublasLt.lib
cuda.lib
cudadevrt.lib
cudart.lib
cudart_static.lib
cudnn.lib
cudnn64_8.lib
cudnn_adv_infer.lib
cudnn_adv_infer64_8.lib
cudnn_adv_train.lib
cudnn_adv_train64_8.lib
cudnn_cnn_infer.lib
cudnn_cnn_infer64_8.lib
cudnn_cnn_train.lib
cudnn_cnn_train64_8.lib
cudnn_ops_infer.lib
cudnn_ops_infer64_8.lib
cudnn_ops_train.lib
cudnn_ops_train64_8.lib
cufft.lib
cufftw.lib
cufilt.lib
curand.lib
cusolver.lib
cusolverMg.lib
cusparse.lib
nppc.lib
nppial.lib
nppicc.lib
nppidei.lib
nppif.lib
nppig.lib
nppim.lib
nppist.lib
nppisu.lib
nppitc.lib
npps.lib
nvblas.lib
nvJitLink.lib
nvJitLink_static.lib
nvjpeg.lib
nvml.lib
nvptxcompiler_static.lib
nvrtc-builtins_static.lib
nvrtc.lib
nvrtc_static.lib
OpenCL.lib
nvinfer.lib
nvinfer_dispatch.lib
nvinfer_lean.lib
nvinfer_plugin.lib
nvinfer_vc_plugin.lib
nvonnxparser.lib
nvparsers.lib
opencv_world481.lib

代码

#define _CRT_SECURE_NO_DEPRECATE

#include
#include
#include
#include
#include
#include
#include "NvInfer.h"

std::vectorstring> labels;
float score_threshold = 0.3f;
float nms_threshold = 0.5f;
int input_h = 640;
int input_w = 640;

std::string lable_path = "model/lable.txt";
std::string engin_path = "model/yolov8n.engine";
std::string video_path = "test/VID_2K.mp4";

NvinferStruct* p = nullptr;
Logger logger;

int w = 0;
int h = 0;
float x_factor = 0;
float y_factor = 0;
std::vector input_image;

float* output_data = nullptr;
size_t output_size = 0;

double preprocessTime = 0;
double inferTime = 0;
double postprocessTime = 0;
double totalTime = 0;
double detFps = 0;

int init() {

    std::ifstream lable_file(lable_path);
    if (!lable_file.is_open())
    {
        std::cerr << "Error opening file: " << lable_path << std::endl;
        return -1;
    }
    std::string line;
    while (std::getline(lable_file, line))
    {
        if (!line.empty())
        {
            labels.push_back(line);
        }
    }
    lable_file.close();

    // 以二进制方式读取文件
    std::ifstream engin_file(engin_path.data(), std::ios::binary);
    if (!engin_file.good()) {
        std::cerr << "文件无法打开，请确定文件是否可用！" << std::endl;
        return -1;
    }
    size_t size = 0;
    engin_file.seekg(0, engin_file.end);    // 将读指针从文件末尾开始移动0个字节
    size = engin_file.tellg();    // 返回读指针的位置，此时读指针的位置就是文件的字节数
    engin_file.seekg(0, engin_file.beg);    // 将读指针从文件开头开始移动0个字节
    char* modelStream = new char[size];
    engin_file.read(modelStream, size);
    engin_file.close();// 关闭文件

    //创建推理核心结构体，初始化变量
    p = new NvinferStruct();

    //初始化反序列化引擎
    p->runtime = nvinfer1::createInferRuntime(logger);

    // 初始化推理引擎
    p->engine = p->runtime->deserializeCudaEngine(modelStream, size);

    // 创建上下文
    p->context = p->engine->createExecutionContext();
    int numNode = p->engine->getNbBindings();

    // 创建gpu数据缓冲区
    p->dataBuffer = new void* [numNode];

    delete[] modelStream;

    for (int i = 0; i < numNode; i++) {
        nvinfer1::Dims dims = p->engine->getBindingDimensions(i);
        nvinfer1::DataType type = p->engine->getBindingDataType(i);
        std::vector shape(dims.d, dims.d + dims.nbDims);
        size_t size = std::accumulate(dims.d, dims.d + dims.nbDims, 1, std::multiplies());
        switch (type)
        {
        case nvinfer1::DataType::kINT32:
        case nvinfer1::DataType::kFLOAT: size *= 4; break;  // 明确为类型 float
        case nvinfer1::DataType::kHALF: size *= 2; break;
        case nvinfer1::DataType::kBOOL:
        case nvinfer1::DataType::kINT8:
        default:break;
        }
        cudaMalloc(&(p->dataBuffer[i]), size);
    }

    output_size = 1 * (labels.size() + 4) * 8400;;
    output_data = new float[output_size];

    return 0;
}

void preprocess(cv::Mat& frame) {
    //前处理
    w = frame.cols;
    h = frame.rows;
    int max = std::max(h, w);
    cv::Mat image = cv::Mat::zeros(cv::Size(max, max), CV_8UC3);
    cv::Rect roi(0, 0, w, h);
    frame.copyTo(image(roi));
    x_factor = image.cols / static_cast(input_w);
    y_factor = image.rows / static_cast(input_h);
    cv::resize(image, image, cv::Size(input_w, input_h));
    cv::cvtColor(image, image, cv::COLOR_BGR2RGB);
    std::vector bgrChannels(3);
    cv::split(image, bgrChannels);
    for (int c = 0; c < 3; c++)
    {
        bgrChannels[c].convertTo(bgrChannels[c], CV_32FC1, 1.0 / 255.0);
    }
    std::vector input_image;
    int image_area = input_h * input_w;
    input_image.clear();
    input_image.resize(3 * image_area);
    size_t single_chn_size = image_area * sizeof(float);
    memcpy(input_image.data(), (float*)bgrChannels[0].data, single_chn_size);
    memcpy(input_image.data() + image_area, (float*)bgrChannels[1].data, single_chn_size);
    memcpy(input_image.data() + image_area * 2, (float*)bgrChannels[2].data, single_chn_size);

    cudaMemcpy(p->dataBuffer[0], input_image.data(), input_image.size() * sizeof(float), cudaMemcpyHostToDevice);
}

void postprocess(std::vector& detectionResult) {


    cudaMemcpy(output_data, p->dataBuffer[1], output_size * sizeof(float), cudaMemcpyDeviceToHost);

    cv::Mat dout(labels.size() + 4, 8400, CV_32F, output_data);
    cv::Mat det_output = dout.t();

    std::vector boxes;
    std::vector classIds;
    std::vector confidences;

    for (int i = 0; i < det_output.rows; i++)
    {
        cv::Mat classes_scores = det_output.row(i).colRange(4, labels.size() + 4);
        cv::Point classIdPoint;
        double score;
        cv::minMaxLoc(classes_scores, 0, &score, 0, &classIdPoint);

        if (score > score_threshold)
        {
            float cx = det_output.at(i, 0);
            float cy = det_output.at(i, 1);
            float ow = det_output.at(i, 2);
            float oh = det_output.at(i, 3);
            int x = static_cast((cx - 0.5 * ow) * x_factor);
            int y = static_cast((cy - 0.5 * oh) * y_factor);
            int width = static_cast(ow * x_factor);
            int height = static_cast(oh * y_factor);

            // 坐标值安全校验
            if (x < 0)x = 0;
            if (y < 0)y = 0;

            if (x > w)x = w;
            if (y > h)y = h;

            if (x + width > w)width = w - x;
            if (y + height > h)height = h - y;

            cv::Rect box;
            box.x = x;
            box.y = y;
            box.width = width;
            box.height = height;

            boxes.push_back(box);
            classIds.push_back(classIdPoint.x);
            confidences.push_back(score);
        }
    }

    std::vector indexes;
    cv::dnn::NMSBoxes(boxes, confidences, score_threshold, nms_threshold, indexes);


    for (size_t i = 0; i < indexes.size(); i++)
    {
        int index = indexes[i];
        detresult box(labels[classIds[index]], classIds[index], confidences[index], boxes[index]);
        detectionResult.push_back(box);
    }

}

void draw(cv::Mat& frame, std::vector& detectionResult) {

    for (size_t i = 0; i < detectionResult.size(); ++i)
    {
        detresult box = detectionResult[i];
        cv::rectangle(frame, box.rect, cv::Scalar(0, 0, 255), 2);
        std::string label = box.className + ":" + cv::format("%.2f", box.confidence);
        putText(frame, label, cv::Point(box.rect.x, box.rect.y - 5), cv::FONT_HERSHEY_SIMPLEX, 1, cv::Scalar(0, 255, 0), 2);
    }

    // 绘制时间
    putText(frame, "preprocessTime:" + std::to_string(preprocessTime * 1000) + "ms", cv::Point(10, 30), cv::FONT_HERSHEY_SIMPLEX, 1, cv::Scalar(0, 255, 0), 2);
    putText(frame, "inferTime:" + std::to_string(inferTime * 1000) + "ms", cv::Point(10, 70), cv::FONT_HERSHEY_SIMPLEX, 1, cv::Scalar(0, 255, 0), 2);
    putText(frame, "postprocessTime:" + std::to_string(postprocessTime * 1000) + "ms", cv::Point(10, 110), cv::FONT_HERSHEY_SIMPLEX, 1, cv::Scalar(0, 255, 0), 2);
    putText(frame, "totalTime:" + std::to_string(totalTime * 1000) + "ms", cv::Point(10, 150), cv::FONT_HERSHEY_SIMPLEX, 1, cv::Scalar(0, 255, 0), 2);
    putText(frame, "detFps:" + std::to_string(detFps), cv::Point(10, 190), cv::FONT_HERSHEY_SIMPLEX, 1, cv::Scalar(0, 255, 0), 2);

    cv::imshow("detresult", frame);

}

void destroy() {

    delete output_data;
    delete p->dataBuffer;
    p->context->destroy();
    p->engine->destroy();
    p->runtime->destroy();
    delete p;
}

int main()
{
    init();

    //cv::Mat frame = cv::imread(img_path);
    //cv::imshow("1.jpg", frame);
    //cv::waitKey(0);

    cv::VideoCapture capture(video_path);
    // 检查视频是否成功打开
    if (!capture.isOpened())
    {
        std::cout << "无法读取视频文件" << std::endl;
        return -1;
    }

    double fps = capture.get(cv::CAP_PROP_FPS);
    int width = static_cast(capture.get(cv::CAP_PROP_FRAME_WIDTH));
    int height = static_cast(capture.get(cv::CAP_PROP_FRAME_HEIGHT));

    cv::Mat frame;
    while (true)
    {
        bool success = capture.read(frame); // 读取一帧数据
        // 检查是否成功读取帧
        if (!success)
        {
            std::cout << "读取完毕" << std::endl;
            break;
        }

        double start = (double)cv::getTickCount();
        preprocess(frame);
        preprocessTime = ((double)cv::getTickCount() - start) / cv::getTickFrequency();

        //推理
        start = (double)cv::getTickCount();
        p->context->executeV2(p->dataBuffer);
        inferTime = ((double)cv::getTickCount() - start) / cv::getTickFrequency();

        //后处理
        start = (double)cv::getTickCount();
        std::vector detectionResult;
        postprocess(detectionResult);
        postprocessTime = ((double)cv::getTickCount() - start) / cv::getTickFrequency();

        totalTime = preprocessTime + inferTime + postprocessTime;
        detFps = (1 / (totalTime));

        //绘制、显示
        cv::namedWindow("detresult", cv::WINDOW_NORMAL); // cv::WINDOW_NORMAL允许用户调整窗口大小
        cv::resizeWindow("detresult", width / 2, height / 2); // 设置窗口的宽度和高度
        draw(frame, detectionResult);

        if (cv::waitKey(1) == 27) // 通过按下ESC键退出循环
        {
            break;
        }
    }

    destroy();

    cv::destroyAllWindows();
    getchar();

    return 0;
}

  1. #define _CRT_SECURE_NO_DEPRECATE
  2.  
  3. #include <iostream>
  4. #include <opencv2/opencv.hpp>
  5. #include <opencv2/imgproc.hpp>
  6. #include <opencv2/highgui.hpp>
  7. #include <fstream>
  8. #include <numeric>
  9. #include "NvInfer.h"
  10.  
  11. std::vector<std::string> labels;
  12. float score_threshold = 0.3f;
  13. float nms_threshold = 0.5f;
  14. int input_h = 640;
  15. int input_w = 640;
  16.  
  17. std::string lable_path = "model/lable.txt";
  18. std::string engin_path = "model/yolov8n.engine";
  19. std::string video_path = "test/VID_2K.mp4";
  20.  
  21. NvinferStruct* p = nullptr;
  22. Logger logger;
  23.  
  24. int w = 0;
  25. int h = 0;
  26. float x_factor = 0;
  27. float y_factor = 0;
  28. std::vector<float> input_image;
  29.  
  30. float* output_data = nullptr;
  31. size_t output_size = 0;
  32.  
  33. double preprocessTime = 0;
  34. double inferTime = 0;
  35. double postprocessTime = 0;
  36. double totalTime = 0;
  37. double detFps = 0;
  38.  
  39. int init() {
  40.  
  41. 	std::ifstream lable_file(lable_path);
  42. 	if (!lable_file.is_open())
  43. 	{
  44. 		std::cerr << "Error opening file: " << lable_path << std::endl;
  45. 		return -1;
  46. 	}
  47. 	std::string line;
  48. 	while (std::getline(lable_file, line))
  49. 	{
  50. 		if (!line.empty())
  51. 		{
  52. 			labels.push_back(line);
  53. 		}
  54. 	}
  55. 	lable_file.close();
  56.  
  57. 	// 以二进制方式读取文件
  58. 	std::ifstream engin_file(engin_path.data(), std::ios::binary);
  59. 	if (!engin_file.good()) {
  60. 		std::cerr << "文件无法打开，请确定文件是否可用！" << std::endl;
  61. 		return -1;
  62. 	}
  63. 	size_t size = 0;
  64. 	engin_file.seekg(0, engin_file.end);	// 将读指针从文件末尾开始移动0个字节
  65. 	size = engin_file.tellg();	// 返回读指针的位置，此时读指针的位置就是文件的字节数
  66. 	engin_file.seekg(0, engin_file.beg);	// 将读指针从文件开头开始移动0个字节
  67. 	char* modelStream = new char[size];
  68. 	engin_file.read(modelStream, size);
  69. 	engin_file.close();// 关闭文件
  70.  
  71. 	//创建推理核心结构体，初始化变量
  72. 	p = new NvinferStruct();
  73.  
  74. 	//初始化反序列化引擎
  75. 	p->runtime = nvinfer1::createInferRuntime(logger);
  76.  
  77. 	// 初始化推理引擎
  78. 	p->engine = p->runtime->deserializeCudaEngine(modelStream, size);
  79.  
  80. 	// 创建上下文
  81. 	p->context = p->engine->createExecutionContext();
  82. 	int numNode = p->engine->getNbBindings();
  83.  
  84. 	// 创建gpu数据缓冲区
  85. 	p->dataBuffer = new void* [numNode];
  86.  
  87. 	delete[] modelStream;
  88.  
  89. 	for (int i = 0; i < numNode; i++) {
  90. 		nvinfer1::Dims dims = p->engine->getBindingDimensions(i);
  91. 		nvinfer1::DataType type = p->engine->getBindingDataType(i);
  92. 		std::vector<int> shape(dims.d, dims.d + dims.nbDims);
  93. 		size_t size = std::accumulate(dims.d, dims.d + dims.nbDims, 1, std::multiplies<size_t>());
  94. 		switch (type)
  95. 		{
  96. 		case nvinfer1::DataType::kINT32:
  97. 		case nvinfer1::DataType::kFLOAT: size *= 4; break;  // 明确为类型 float
  98. 		case nvinfer1::DataType::kHALF: size *= 2; break;
  99. 		case nvinfer1::DataType::kBOOL:
  100. 		case nvinfer1::DataType::kINT8:
  101. 		default:break;
  102. 		}
  103. 		cudaMalloc(&(p->dataBuffer[i]), size);
  104. 	}
  105.  
  106. 	output_size = 1 * (labels.size() + 4) * 8400;;
  107. 	output_data = new float[output_size];
  108.  
  109. 	return 0;
  110. }
  111.  
  112. void preprocess(cv::Mat& frame) {
  113. 	//前处理
  114. 	w = frame.cols;
  115. 	h = frame.rows;
  116. 	int max = std::max(h, w);
  117. 	cv::Mat image = cv::Mat::zeros(cv::Size(max, max), CV_8UC3);
  118. 	cv::Rect roi(0, 0, w, h);
  119. 	frame.copyTo(image(roi));
  120. 	x_factor = image.cols / static_cast<float>(input_w);
  121. 	y_factor = image.rows / static_cast<float>(input_h);
  122. 	cv::resize(image, image, cv::Size(input_w, input_h));
  123. 	cv::cvtColor(image, image, cv::COLOR_BGR2RGB);
  124. 	std::vector<cv::Mat> bgrChannels(3);
  125. 	cv::split(image, bgrChannels);
  126. 	for (int c = 0; c < 3; c++)
  127. 	{
  128. 		bgrChannels[c].convertTo(bgrChannels[c], CV_32FC1, 1.0 / 255.0);
  129. 	}
  130. 	std::vector<float> input_image;
  131. 	int image_area = input_h * input_w;
  132. 	input_image.clear();
  133. 	input_image.resize(3 * image_area);
  134. 	size_t single_chn_size = image_area * sizeof(float);
  135. 	memcpy(input_image.data(), (float*)bgrChannels[0].data, single_chn_size);
  136. 	memcpy(input_image.data() + image_area, (float*)bgrChannels[1].data, single_chn_size);
  137. 	memcpy(input_image.data() + image_area * 2, (float*)bgrChannels[2].data, single_chn_size);
  138.  
  139. 	cudaMemcpy(p->dataBuffer[0], input_image.data(), input_image.size() * sizeof(float), cudaMemcpyHostToDevice);
  140. }
  141.  
  142. void postprocess(std::vector<detresult>& detectionResult) {
  143.  
  144.  
  145. 	cudaMemcpy(output_data, p->dataBuffer[1], output_size * sizeof(float), cudaMemcpyDeviceToHost);
  146.  
  147. 	cv::Mat dout(labels.size() + 4, 8400, CV_32F, output_data);
  148. 	cv::Mat det_output = dout.t();
  149.  
  150. 	std::vector<cv::Rect> boxes;
  151. 	std::vector<int> classIds;
  152. 	std::vector<float> confidences;
  153.  
  154. 	for (int i = 0; i < det_output.rows; i++)
  155. 	{
  156. 		cv::Mat classes_scores = det_output.row(i).colRange(4, labels.size() + 4);
  157. 		cv::Point classIdPoint;
  158. 		double score;
  159. 		cv::minMaxLoc(classes_scores, 0, &score, 0, &classIdPoint);
  160.  
  161. 		if (score > score_threshold)
  162. 		{
  163. 			float cx = det_output.at<float>(i, 0);
  164. 			float cy = det_output.at<float>(i, 1);
  165. 			float ow = det_output.at<float>(i, 2);
  166. 			float oh = det_output.at<float>(i, 3);
  167. 			int x = static_cast<int>((cx - 0.5 * ow) * x_factor);
  168. 			int y = static_cast<int>((cy - 0.5 * oh) * y_factor);
  169. 			int width = static_cast<int>(ow * x_factor);
  170. 			int height = static_cast<int>(oh * y_factor);
  171.  
  172. 			// 坐标值安全校验
  173. 			if (x < 0)x = 0;
  174. 			if (y < 0)y = 0;
  175.  
  176. 			if (x > w)x = w;
  177. 			if (y > h)y = h;
  178.  
  179. 			if (x + width > w)width = w - x;
  180. 			if (y + height > h)height = h - y;
  181.  
  182. 			cv::Rect box;
  183. 			box.x = x;
  184. 			box.y = y;
  185. 			box.width = width;
  186. 			box.height = height;
  187.  
  188. 			boxes.push_back(box);
  189. 			classIds.push_back(classIdPoint.x);
  190. 			confidences.push_back(score);
  191. 		}
  192. 	}
  193.  
  194. 	std::vector<int> indexes;
  195. 	cv::dnn::NMSBoxes(boxes, confidences, score_threshold, nms_threshold, indexes);
  196.  
  197.  
  198. 	for (size_t i = 0; i < indexes.size(); i++)
  199. 	{
  200. 		int index = indexes[i];
  201. 		detresult box(labels[classIds[index]], classIds[index], confidences[index], boxes[index]);
  202. 		detectionResult.push_back(box);
  203. 	}
  204.  
  205. }
  206.  
  207. void draw(cv::Mat& frame, std::vector<detresult>& detectionResult) {
  208.  
  209. 	for (size_t i = 0; i < detectionResult.size(); ++i)
  210. 	{
  211. 		detresult box = detectionResult[i];
  212. 		cv::rectangle(frame, box.rect, cv::Scalar(0, 0, 255), 2);
  213. 		std::string label = box.className + ":" + cv::format("%.2f", box.confidence);
  214. 		putText(frame, label, cv::Point(box.rect.x, box.rect.y - 5), cv::FONT_HERSHEY_SIMPLEX, 1, cv::Scalar(0, 255, 0), 2);
  215. 	}
  216.  
  217. 	// 绘制时间
  218. 	putText(frame, "preprocessTime:" + std::to_string(preprocessTime * 1000) + "ms", cv::Point(10, 30), cv::FONT_HERSHEY_SIMPLEX, 1, cv::Scalar(0, 255, 0), 2);
  219. 	putText(frame, "inferTime:" + std::to_string(inferTime * 1000) + "ms", cv::Point(10, 70), cv::FONT_HERSHEY_SIMPLEX, 1, cv::Scalar(0, 255, 0), 2);
  220. 	putText(frame, "postprocessTime:" + std::to_string(postprocessTime * 1000) + "ms", cv::Point(10, 110), cv::FONT_HERSHEY_SIMPLEX, 1, cv::Scalar(0, 255, 0), 2);
  221. 	putText(frame, "totalTime:" + std::to_string(totalTime * 1000) + "ms", cv::Point(10, 150), cv::FONT_HERSHEY_SIMPLEX, 1, cv::Scalar(0, 255, 0), 2);
  222. 	putText(frame, "detFps:" + std::to_string(detFps), cv::Point(10, 190), cv::FONT_HERSHEY_SIMPLEX, 1, cv::Scalar(0, 255, 0), 2);
  223.  
  224. 	cv::imshow("detresult", frame);
  225.  
  226. }
  227.  
  228. void destroy() {
  229.  
  230. 	delete output_data;
  231. 	delete p->dataBuffer;
  232. 	p->context->destroy();
  233. 	p->engine->destroy();
  234. 	p->runtime->destroy();
  235. 	delete p;
  236. }
  237.  
  238. int main()
  239. {
  240. 	init();
  241.  
  242. 	//cv::Mat frame = cv::imread(img_path);
  243. 	//cv::imshow("1.jpg", frame);
  244. 	//cv::waitKey(0);
  245.  
  246. 	cv::VideoCapture capture(video_path);
  247. 	// 检查视频是否成功打开
  248. 	if (!capture.isOpened())
  249. 	{
  250. 		std::cout << "无法读取视频文件" << std::endl;
  251. 		return -1;
  252. 	}
  253.  
  254. 	double fps = capture.get(cv::CAP_PROP_FPS);
  255. 	int width = static_cast<int>(capture.get(cv::CAP_PROP_FRAME_WIDTH));
  256. 	int height = static_cast<int>(capture.get(cv::CAP_PROP_FRAME_HEIGHT));
  257.  
  258. 	cv::Mat frame;
  259. 	while (true)
  260. 	{
  261. 		bool success = capture.read(frame); // 读取一帧数据
  262. 		// 检查是否成功读取帧
  263. 		if (!success)
  264. 		{
  265. 			std::cout << "读取完毕" << std::endl;
  266. 			break;
  267. 		}
  268.  
  269. 		double start = (double)cv::getTickCount();
  270. 		preprocess(frame);
  271. 		preprocessTime = ((double)cv::getTickCount() - start) / cv::getTickFrequency();
  272.  
  273. 		//推理
  274. 		start = (double)cv::getTickCount();
  275. 		p->context->executeV2(p->dataBuffer);
  276. 		inferTime = ((double)cv::getTickCount() - start) / cv::getTickFrequency();
  277.  
  278. 		//后处理
  279. 		start = (double)cv::getTickCount();
  280. 		std::vector<detresult> detectionResult;
  281. 		postprocess(detectionResult);
  282. 		postprocessTime = ((double)cv::getTickCount() - start) / cv::getTickFrequency();
  283.  
  284. 		totalTime = preprocessTime + inferTime + postprocessTime;
  285. 		detFps = (1 / (totalTime));
  286.  
  287. 		//绘制、显示
  288. 		cv::namedWindow("detresult", cv::WINDOW_NORMAL); // cv::WINDOW_NORMAL允许用户调整窗口大小
  289. 		cv::resizeWindow("detresult", width / 2, height / 2); // 设置窗口的宽度和高度
  290. 		draw(frame, detectionResult);
  291.  
  292. 		if (cv::waitKey(1) == 27) // 通过按下ESC键退出循环
  293. 		{
  294. 			break;
  295. 		}
  296. 	}
  297.  
  298. 	destroy();
  299.  
  300. 	cv::destroyAllWindows();
  301. 	getchar();
  302.  
  303. 	return 0;
  304. }

下载

源码下载

天天代码码天天
微信公众号
.NET 人工智能实践
注：本文转载自blog.csdn.net的天天代码码天天的文章"https://lw112190.blog.csdn.net/article/details/143225616"。版权归原作者所有，此博客不拥有其著作权，亦不承担相应法律责任。如有侵权，请联系我们删除。
复制链接
相关推荐
发表评论

评论记录：

未查询到任何数据！