Windows系統下YOLO動態鏈接庫的封裝和調用
Windows10+VS2015+OpenCV3.4.1+CUDA8.0+cuDNN8.0
參考教程 https://blog.csdn.net/stjuliet/article/details/87884976
承接上一篇文章所做工作，這篇文章進一步講述如何將YOLO封裝成動態鏈接庫以方便后續目標檢測時直接調用。
關於動態鏈接庫的介紹：
https://www.cnblogs.com/chechen/p/8676226.html
https://www.jianshu.com/p/458f87251b3d?tdsourcetag=s_pctim_aiomsg
step1 運行環境和前期准備
與上一篇文章所需環境完全一致，具體可參考：
https://blog.csdn.net/stjuliet/article/details/87731998
配置opecv3.4.1 cuda8.0以及配套cudnn
step2 編譯動態鏈接庫
1、下載Darknet源代碼：
https://github.com/AlexeyAB/darknet
2、
(1)下載解壓后，進入darknet-master->build->darknet目錄：
(2)打開yolo_cpp_dll.vcxproj文件，將具有CUDA的版本改成自己使用的版本（默認為10.0），一共有兩處，分別在55行和302行
自己電腦裝了cuda10和8,這里用8
(3)打開yolo_cpp_dll.sln文件，在屬性管理器中配置包含目錄、庫目錄、附加依賴項（和OpenCV環境配置一樣），特別注意要將CUDA設備中的Generation改成自己顯卡對應的計算能力（默認添加了35和75兩項，可能不是你的顯卡的計算能力，可以去英偉達顯卡官網查詢計算能力：https://developer.nvidia.com/cuda-gpus#collapseOne）
，否則接下來的生成會出錯。
(4)分別設置Debug/Release - x64，右鍵項目->生成，成功后在darknet-master\build\darknet\x64目錄下找到生成的yolo_cpp_dll.lib和yolo_cpp_dll.dll兩個文件。
step3 調用動態鏈接庫
一、至此所有准備工作已經完成，首先將調用所需的所有文件找出來：
1、動態鏈接庫（均在darknet-master\build\darknet\x64目錄下）
(1)yolo_cpp_dll.lib
(2)yolo_cpp_dll.dll
(3)pthreadGC2.dll
(4)pthreadVC2.dll
2、OpenCV庫（取決於使用debug還是release模式）
(1)opencv_world340d.dll
(2)opencv_world340.dll
如果是擴展庫需要
opencv_aruco341.lib opencv_bgsegm341.lib opencv_bioinspired341.lib opencv_calib3d341.lib opencv_ccalib341.lib opencv_core341.lib opencv_cudaarithm341.lib opencv_cudabgsegm341.lib opencv_cudacodec341.lib opencv_cudafeatures2d341.lib opencv_cudafilters341.lib opencv_cudaimgproc341.lib opencv_cudalegacy341.lib opencv_cudaobjdetect341.lib opencv_cudaoptflow341.lib opencv_cudastereo341.lib opencv_cudawarping341.lib opencv_cudev341.lib opencv_datasets341.lib opencv_dnn341.lib opencv_dnn_objdetect341.lib opencv_dpm341.lib opencv_face341.lib opencv_features2d341.lib opencv_flann341.lib opencv_fuzzy341.lib opencv_hfs341.lib opencv_highgui341.lib opencv_imgcodecs341.lib opencv_imgproc341.lib opencv_img_hash341.lib opencv_line_descriptor341.lib opencv_ml341.lib opencv_objdetect341.lib opencv_optflow341.lib opencv_phase_unwrapping341.lib opencv_photo341.lib opencv_plot341.lib opencv_reg341.lib opencv_rgbd341.lib opencv_saliency341.lib opencv_shape341.lib opencv_stereo341.lib opencv_stitching341.lib opencv_structured_light341.lib opencv_superres341.lib opencv_surface_matching341.lib opencv_text341.lib opencv_tracking341.lib opencv_video341.lib opencv_videoio341.lib opencv_videostab341.lib opencv_xfeatures2d341.lib opencv_ximgproc341.lib opencv_xobjdetect341.lib opencv_xphoto341.lib
3、YOLO模型文件（第一個文件在darknet-master\build\darknet\x64\data目錄下，第二個文件在darknet-master\build\darknet\x64目錄下，第三個文件需要自己下載，下載鏈接見前一篇文章）
(1)coco.names
(2)yolov3.cfg
(3)yolov3.weights
4、頭文件
(1)yolo_v2_class.hpp
頭文件包含了動態鏈接庫中具體的類的定義，調用時需要引用，這個文件在darknet-master\build\darknet目錄下的yolo_console_dll.sln中，將其復制到記事本保存成.hpp文件即可。
二、在VS2015中新建一個空項目，在源文件中添加main.cpp，將第一步中所有文件全部放入與main.cpp同路徑的文件夾中，並且放入一個目標檢測的測試視頻test0.mp4，在main.cpp中添加如下代碼：
#include <iostream>
#ifdef _WIN32
#define OPENCV
#define GPU
#endif
#include "yolo_v2_class.hpp" //引用動態鏈接庫中的頭文件
#include <opencv2/opencv.hpp>
#include "opencv2/highgui/highgui.hpp"
//#pragma comment(lib, "opencv_world340d.lib") //引入OpenCV鏈接庫
#pragma comment(lib, "yolo_cpp_dll.lib") //引入YOLO動態鏈接庫
//以下兩段代碼來自yolo_console_dll.sln
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 const max_width = (text_size.width > i.w + 2) ? text_size.width : (i.w + 2);
cv::rectangle(mat_img, cv::Point2f(std::max((int)i.x - 1, 0), std::max((int)i.y - 30, 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 - 10), cv::FONT_HERSHEY_COMPLEX_SMALL, 1.2, cv::Scalar(0, 0, 0), 2);
}
}
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);
}
}
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;
}
int main()
{
std::string names_file = "../../yolo權重/coco.names";
std::string cfg_file = "../../yolo權重/yolov3.cfg";
std::string weights_file = "../../yolo權重/yolov3.weights";
Detector detector(cfg_file, weights_file, 0); //初始化檢測器
//std::vector<std::string> obj_names = objects_names_from_file(names_file); //調用獲得分類對象名稱
//或者使用以下四行代碼也可實現讀入分類對象文件
std::vector<std::string> obj_names;
std::ifstream ifs(names_file.c_str());
std::string line;
while (getline(ifs, line)) obj_names.push_back(line);
//測試是否成功讀入分類對象文件
for (size_t i = 0; i < obj_names.size(); i++)
{
std::cout << obj_names[i] << std::endl;
}
cv::VideoCapture capture;
capture.open("DJI_0002.MP4");
if (!capture.isOpened())
{
printf("文件打開失敗");
}
cv::Mat frame;
while (true)
{
capture >> frame;
std::vector<bbox_t> result_vec = detector.detect(frame);
draw_boxes(frame, result_vec, obj_names);
cv::namedWindow("test", CV_WINDOW_NORMAL);
cv::imshow("test", frame);
cv::waitKey(3);
}
return 0;
}
工程配置
包含目錄
opencv
cuda
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v8.0\include F:\dongdong\0tool\navidia_cuda_opencv\opencv3.4.1\include F:\dongdong\0tool\navidia_cuda_opencv\opencv3.4.1\include\opencv2 F:\dongdong\0tool\navidia_cuda_opencv\opencv3.4.1\include\opencv
庫目錄
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v8.0\lib\x64 F:\dongdong\0tool\navidia_cuda_opencv\opencv3.4.1\x64\vc14\lib
輸入附加依賴項
增加 cuda
cublas.lib cuda.lib cudadevrt.lib cudart.lib cudart_static.lib nvcuvid.lib OpenCL.lib cudnn.lib
增加yolo
yolo_cpp_dll.lib
增加opencv
opencv_aruco341.lib opencv_bgsegm341.lib opencv_bioinspired341.lib opencv_calib3d341.lib opencv_ccalib341.lib opencv_core341.lib opencv_cudaarithm341.lib opencv_cudabgsegm341.lib opencv_cudacodec341.lib opencv_cudafeatures2d341.lib opencv_cudafilters341.lib opencv_cudaimgproc341.lib opencv_cudalegacy341.lib opencv_cudaobjdetect341.lib opencv_cudaoptflow341.lib opencv_cudastereo341.lib opencv_cudawarping341.lib opencv_cudev341.lib opencv_datasets341.lib opencv_dnn341.lib opencv_dnn_objdetect341.lib opencv_dpm341.lib opencv_face341.lib opencv_features2d341.lib opencv_flann341.lib opencv_fuzzy341.lib opencv_hfs341.lib opencv_highgui341.lib opencv_imgcodecs341.lib opencv_imgproc341.lib opencv_img_hash341.lib opencv_line_descriptor341.lib opencv_ml341.lib opencv_objdetect341.lib opencv_optflow341.lib opencv_phase_unwrapping341.lib opencv_photo341.lib opencv_plot341.lib opencv_reg341.lib opencv_rgbd341.lib opencv_saliency341.lib opencv_shape341.lib opencv_stereo341.lib opencv_stitching341.lib opencv_structured_light341.lib opencv_superres341.lib opencv_surface_matching341.lib opencv_text341.lib opencv_tracking341.lib opencv_video341.lib opencv_videoio341.lib opencv_videostab341.lib opencv_xfeatures2d341.lib opencv_ximgproc341.lib opencv_xobjdetect341.lib opencv_xphoto341.lib
預處理器
_CRT_SECURE_NO_WARNINGS _WINSOCK_DEPRECATED_NO_WARNINGS
工程配置完畢
4 配置代碼
代碼修改：
1包含yolo文件
#include "yolo_v2_class.hpp" //引用動態鏈接庫中的頭文件
由於找不到庫文件，把文件拷貝到工程main.cpp函數下
2修改權重文件路徑
上一層
再上一層
進入
為了省事也可以直接放在工程里同級目錄。
運行代碼
貼一張原來教程的作者圖
main測試代碼
、
#include <iostream>
#ifdef _WIN32
#define OPENCV
#define GPU
#endif
#include "yolo_v2_class.hpp" //引用動態鏈接庫中的頭文件
#include <opencv2/opencv.hpp>
#include "opencv2/highgui/highgui.hpp"
//#pragma comment(lib, "opencv_world340d.lib") //引入OpenCV鏈接庫
#pragma comment(lib, "yolo_cpp_dll.lib") //引入YOLO動態鏈接庫
//以下兩段代碼來自yolo_console_dll.sln
/*
輸入：
cv::Mat mat_img, 目標圖像
std::vector<bbox_t> result_vec, 所有目標框信息 位置 大小
std::vector<std::string> obj_names 所有目標名字列表
*/
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) { //遍歷目標框
cv::Scalar color = obj_id_to_color(i.obj_id);//根據目標框ID轉換顏色
cv::rectangle(mat_img, cv::Rect(i.x, i.y, i.w, i.h), color, 2); // 在圖像上畫目標框
if (obj_names.size() > i.obj_id) { //如果目標ID小於名字最大ID，證明事先賦予了名字
std::string obj_name = obj_names[i.obj_id]; //根據目標ID獲取名字，所以訓練的時候直接是分配ID了，根據ID在獲取名字
if (i.track_id > 0) obj_name += " - " + std::to_string(i.track_id);// 啥意思？如果有追蹤ID?? 加上編號？？
cv::Size const text_size = getTextSize(obj_name, cv::FONT_HERSHEY_COMPLEX_SMALL, 1.2, 2, 0);// 名字轉化為text
int const max_width = (text_size.width > i.w + 2) ? text_size.width : (i.w + 2);
//畫矩形
cv::rectangle(mat_img, cv::Point2f(std::max((int)i.x - 1, 0), std::max((int)i.y - 30, 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);
//畫文字
putText(mat_img, obj_name, cv::Point2f(i.x, i.y - 10), cv::FONT_HERSHEY_COMPLEX_SMALL, 1.2, cv::Scalar(0, 0, 0), 2);
}
}
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);
}
}
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;
}
int main()
{
std::string names_file = "../../yolo權重/coco.names";
std::string cfg_file = "../../yolo權重/yolov3.cfg";
std::string weights_file = "../../yolo權重/yolov3.weights";
Detector detector(cfg_file, weights_file, 0); //初始化檢測器
//std::vector<std::string> obj_names = objects_names_from_file(names_file); //調用獲得分類對象名稱
//或者使用以下四行代碼也可實現讀入分類對象文件
//將標簽名字從文件逐條讀取出來
std::vector<std::string> obj_names;
std::ifstream ifs(names_file.c_str());
std::string line;
while (getline(ifs, line)) obj_names.push_back(line);//讀取成功一條
//測試是否成功讀入分類對象文件
for (size_t i = 0; i < obj_names.size(); i++)
{
std::cout << obj_names[i] << std::endl; //輸出標簽名字
}
cv::VideoCapture capture;
capture.open("DJI_0002.MP4"); //打開測試視頻
if (!capture.isOpened())
{
printf("文件打開失敗");
}
cv::Mat frame;
while (true)
{
capture >> frame;
std::vector<bbox_t> result_vec = detector.detect(frame); // 檢測一幀，輸出目標框信息容器
draw_boxes(frame, result_vec, obj_names); // 目標圖像 所有目標檢測框 所有目標總分類名稱
cv::namedWindow("test", CV_WINDOW_NORMAL);
cv::imshow("test", frame);
cv::waitKey(3);
}
return 0;
}