Practice 3

include/detectedobject.h

@@ -7,6 +7,7 @@ struct DetectedObject
     int Right;
     int Top;
     int Bottom;
-    int uuid;
+    double score;
+	int classid;
     std::string classname;
 };

include/detector.h

@@ -17,3 +17,18 @@ class Detector
 public:
     virtual vector<DetectedObject> Detect(Mat image) = 0 {}
 };
+class DnnDetector : public Detector
+{
+	string modelPath;
+	string configPath;
+	string labelPath;
+	int width;
+	int height;
+	double scale;
+	Scalar mean;
+	bool swapRB;
+	Net net;
+public:
+	DnnDetector(string _modelPath, string _configPath, string _labelPath, int _width, int _height, double _scale, Scalar _mean, bool _swapRB);
+	vector<DetectedObject> Detect (Mat frame);
+}; 

samples/practice3_YANA.cpp

@@ -0,0 +1,78 @@
+#include <string>
+#include <iostream>
+
+#include <opencv2/core.hpp>
+#include <opencv2/highgui.hpp>
+#include <opencv2/dnn.hpp>
+
+#include "detector.h"
+
+using namespace std;
+using namespace cv;
+using namespace cv::dnn;
+
+const char* cmdAbout =
+"This is an empty application that can be treated as a template for your "
+"own doing-something-cool applications.";
+
+const char* cmdOptions =
+"{ i image                              |        | image to process                  }"
+"{ w  width                             |        | image width for classification    }"
+"{ h  heigth                            |        | image heigth fro classification   }"
+"{ model_path                           |        | path to model                     }"
+"{ config_path                          |        | path to model configuration       }"
+"{ label_path                           |        | path to class labels              }"
+"{ scale                                |        |                                   }"
+"{ mean                                 |        | vector of mean model values       }"
+"{ swap                                 |        | swap R and B channels. TRUE|FALSE }"
+"{ h ? help usage                       |        | print help message                }";
+
+
+int main(int argc, const char** argv) {
+	// Parse command line arguments.
+	CommandLineParser parser(argc, argv, cmdOptions);
+	parser.about(cmdAbout);
+
+	// If help option is given, print help message and exit.
+	if (parser.get<bool>("help")) {
+		parser.printMessage();
+		return 0;
+	}
+
+	// Load image and init parameters
+	String imgName(parser.get<String>("image"));
+	Mat image = imread(imgName);
+
+	String modelPath(parser.get<String>("model_path"));
+	String configPath(parser.get<String>("config_path"));
+	String labelPath(parser.get<String>("label_path"));
+	int width = parser.get<int>("width");
+	int heigth = parser.get<int>("heigth");
+	double scale = parser.get<double>("scale");
+	Scalar mean = parser.get<Scalar>("mean");
+	bool swap = parser.get<bool>("swap");
+
+	String label[21] = { "background", "aeroplane", "bicycle", "bird",
+	  "boat","bottle", "bus",  "car", "cat", "chair", "cow",
+	  "diningtable",  "dog", "horse", "motorbike", "person",
+	  "pottedplant", "sheep",  "sofa", "train", "tvmonitor" };
+
+	//Image detection
+	Detector* a = new DnnDetector(modelPath, configPath, labelPath, width, heigth, scale, mean, swap);
+	vector<DetectedObject> objects = a->Detect(image);
+
+	//Show result
+	for (int i = 0; i < objects.size(); i++)
+	{
+		Point point1(objects[i].Left, objects[i].Bottom);
+		Point point2(objects[i].Right, objects[i].Top);
+		rectangle(image, point1, point2, Scalar(0, 0, 255), 3);
+		string text = "Class: " + label[objects[i].classid] + " Confidence: " + std::to_string(objects[i].score);
+		putText(image, text, Point(objects[i].Left, objects[i].Bottom - 10), FONT_HERSHEY_SIMPLEX, 1, Scalar(0, 0, 0));
+	}
+	namedWindow("Objects", WINDOW_NORMAL);
+	imshow("Objects", image);
+	waitKey(0);
+
+	return 0;
+}

src/detector.cpp

@@ -1 +1,40 @@
-#include "detector.h"
+#include "detector.h"
+
+DnnDetector::DnnDetector(string _modelPath, string _configPath, string _labelPath, int _width, int _height, double _scale, Scalar _mean, bool _swapRB)
+{
+	modelPath = _modelPath;
+	configPath = _configPath;
+	labelPath = _labelPath;
+	width = _width;
+	height = _height;
+	scale = _scale;
+	mean = _mean;
+	swapRB = _swapRB;
+	net = readNet(_modelPath, _configPath);
+	net.setPreferableBackend(DNN_BACKEND_OPENCV);
+	net.setPreferableTarget(DNN_TARGET_CPU);
+};
+vector<DetectedObject> DnnDetector::Detect(Mat frame)
+{
+	Mat inputTensor;
+	blobFromImage(frame, inputTensor, scale, Size(width, height), mean, swapRB, false, CV_32F);
+	net.setInput(inputTensor);
+	Mat prob = net.forward();
+	prob = prob.reshape(1, 1);
+	int n = prob.cols / 7;
+	prob = prob.reshape(1, n);
+	vector<DetectedObject> objects;
+	int cols = frame.cols;
+	int rows = frame.rows;
+	for (int i = 0; i < n; i++) {
+		DetectedObject tmp;
+		tmp.classid = prob.at<float>(i, 1);
+		tmp.score = prob.at<float>(i, 2);
+		tmp.Left = prob.at<float>(i, 3) * cols;
+		tmp.Bottom = prob.at <float>(i, 4) * rows;
+		tmp.Right = prob.at<float>(i, 5) * cols;
+		tmp.Top = prob.at<float>(i, 6) * rows;
+		objects.push_back(tmp);
+		return objects;
+	}
+};