++C<\/strong><\/p>\n#include <opencv2\/opencv.hpp>\n#include <fstream>\n\/\/ Namespaces.\nusing namespace cv;\nusing namespace std;\nusing namespace cv::dnn;<\/pre>\nPython<\/strong><\/p>\nimport cv2\nimport numpy as np<\/pre>\n\u00a0<\/p>\n
\u06f2-\u06f3-\u06f4 \u062a\u0639\u0631\u06cc\u0641 \u067e\u0627\u0631\u0627\u0645\u062a\u0631\u0647\u0627\u06cc \u0633\u0631\u0627\u0633\u0631\u06cc<\/strong><\/p>\n\u067e\u0627\u0631\u0627\u0645\u062a\u0631 \u0647\u0627\u06cc \u062b\u0627\u0628\u062a INPUT_WIDTH \u0648 INPUT_HEIGHT \u0628\u0631\u0627\u06cc \u062a\u0639\u06cc\u06cc\u0646 \u0627\u0646\u062f\u0627\u0632\u0647 blob \u0647\u0633\u062a\u0646\u062f \u06a9\u0647 BLOB \u0645\u062e\u0641\u0641 Binary Large Object \u0627\u0633\u062a \u06a9\u0647 \u062e\u0648\u062f \u062f\u0631 \u0648\u0627\u0642\u0639 \u0634\u0627\u0645\u0644 \u062f\u0627\u062f\u0647 \u0647\u0627\u06cc \u062e\u0627\u0645 \u0642\u0627\u0628\u0644 \u062e\u0648\u0627\u0646\u062f\u0646 \u0627\u0633\u062a. \u062a\u0635\u0648\u06cc\u0631 \u0628\u0627\u06cc\u062f \u0628\u0647 \u06cc\u06a9 blob \u062a\u0628\u062f\u06cc\u0644 \u0634\u0648\u062f \u062a\u0627 \u0634\u0628\u06a9\u0647 \u0628\u062a\u0648\u0627\u0646 \u0622\u0646 \u0631\u0627 \u067e\u0631\u062f\u0627\u0632\u0634 \u06a9\u0631\u062f. \u062f\u0631 \u0627\u06cc\u0646 \u0645\u0648\u0631\u062f \u0645\u0627\u060c \u06cc\u06a9 \u0634\u06cc\u0621 \u0622\u0631\u0627\u06cc\u0647 \u06f4 \u0628\u0639\u062f\u06cc \u0628\u0627 \u0627\u0646\u062f\u0627\u0632\u0647 \u06cc\u0627 \u0634\u06a9\u0644 (\u06f1,\u06f3,\u06f6\u06f4\u06f0,\u06f6\u06f4\u06f0) \u0627\u0633\u062a.<\/p>\n
\u00a0<\/p>\n
\n- SCORE_THRESHOLD : \u0628\u0631\u0627\u06cc \u0641\u06cc\u0644\u062a\u0631 \u06a9\u0631\u062f\u0646 \u06a9\u0644\u0627\u0633 \u0628\u0627 \u0627\u062d\u062a\u0645\u0627\u0644 \u06a9\u0645 \u0627\u0645\u062a\u06cc\u0627\u0632\u0627\u062a.<\/li>\n
- NMS_THRESHOLD : \u0628\u0631\u0627\u06cc \u062d\u0630\u0641 \u06a9\u0627\u062f\u0631\u0647\u0627\u06cc \u0645\u062d\u062f\u0648\u062f\u06a9\u0646\u0646\u062f\u0647 \u0647\u0645\u067e\u0648\u0634\u0627\u0646\u06cc.<\/li>\n
- CONFIDENCE_THRESHOLD : \u0628\u0631\u0627\u06cc \u0641\u06cc\u0644\u062a\u0631 \u062a\u0634\u062e\u06cc\u0635 \u0647\u0627\u06cc \u0628\u0627 \u0627\u062d\u062a\u0645\u0627\u0644 \u06a9\u0645.<\/li>\n<\/ul>\n
\u00a0<\/p>\n
\u062f\u0631 \u062d\u06cc\u0646 \u0645\u0631\u0648\u0631 \u06a9\u062f\u060c \u062f\u0631 \u0645\u0648\u0631\u062f \u0627\u06cc\u0646 \u067e\u0627\u0631\u0627\u0645\u062a\u0631\u0647\u0627 \u0628\u06cc\u0634\u062a\u0631 \u0628\u062d\u062b \u062e\u0648\u0627\u0647\u06cc\u0645 \u06a9\u0631\u062f.<\/p>\n
\u062a\u0648\u062c\u0647<\/strong>: \u0628\u0631\u062e\u0644\u0627\u0641 ++C\u060c \u0645\u0642\u0627\u062f\u06cc\u0631 \u0627\u0646\u062f\u0627\u0632\u0647 \u0648\u0631\u0648\u062f\u06cc \u062f\u0631 \u067e\u0627\u06cc\u062a\u0648\u0646 \u0646\u0645\u06cc \u062a\u0648\u0627\u0646\u0646\u062f \u0627\u0632 \u0646\u0648\u0639 float \u0628\u0627\u0634\u0646\u062f.<\/p>\n\u00a0<\/p>\n
++C<\/strong><\/p>\n\/\/ Constants.\nconst float INPUT_WIDTH = 640.0;\nconst float INPUT_HEIGHT = 640.0;\nconst float SCORE_THRESHOLD = 0.5;\nconst float NMS_THRESHOLD = 0.45;\nconst float CONFIDENCE_THRESHOLD = 0.45;\n\n\/\/ Text parameters.\nconst float FONT_SCALE = 0.7;\nconst int FONT_FACE = FONT_HERSHEY_SIMPLEX;\nconst int THICKNESS = 1;\n\n\/\/ Colors.\nScalar BLACK = Scalar(0,0,0);\nScalar BLUE = Scalar(255, 178, 50);\nScalar YELLOW = Scalar(0, 255, 255);\nScalar RED = Scalar(0,0,255);<\/pre>\nPython<\/strong><\/p>\n# Constants.\nINPUT_WIDTH = 640\nINPUT_HEIGHT = 640\nSCORE_THRESHOLD = 0.5\nNMS_THRESHOLD = 0.45\nCONFIDENCE_THRESHOLD = 0.45\n\n# Text parameters.\nFONT_FACE = cv2.FONT_HERSHEY_SIMPLEX\nFONT_SCALE = 0.7\nTHICKNESS = 1\n\n# Colors.\nBLACK = (0,0,0)\nBLUE = (255,178,50)\nYELLOW = (0,255,255)<\/pre>\n\u00a0<\/p>\n
\u06f3-\u06f3-\u06f4- \u062a\u0646\u0638\u06cc\u0645 \u0628\u0631\u0686\u0633\u0628<\/strong><\/p>\n\u062a\u0627\u0628\u0639 draw_label \u0646\u0627\u0645 \u06a9\u0644\u0627\u0633\u200c\u0647\u0627 \u0631\u0627 \u06a9\u0647 \u062f\u0631 \u06af\u0648\u0634\u0647 \u0633\u0645\u062a \u0686\u067e \u0628\u0627\u0644\u0627\u06cc \u06a9\u0627\u062f\u0631 \u0645\u062d\u062f\u0648\u062f\u06a9\u0646\u0646\u062f\u0647\u060c \u062d\u0627\u0634\u06cc\u0647\u200c\u0646\u0648\u06cc\u0633\u06cc \u0645\u06cc \u06a9\u0646\u062f \u06a9\u0647 \u06a9\u062f \u0646\u0633\u0628\u062a\u0627\u064b \u0633\u0627\u062f\u0647 \u0627\u06cc \u062f\u0627\u0631\u062f. \u0631\u0634\u062a\u0647 \u0645\u062a\u0646 \u060c \u0628\u0647 \u0639\u0646\u0648\u0627\u0646 \u0628\u0631\u0686\u0633\u0628 \u0628\u0627 \u062a\u0627\u0628\u0639 ()<\/strong>OpenCV getTextSize<\/strong>\u00a0\u062f\u0631 \u0622\u0631\u06af\u0648\u0645\u0627\u0646 \u0627\u0631\u0633\u0627\u0644 \u0645\u06cc \u0634\u0648\u062f \u062a\u0627 \u0627\u0646\u062f\u0627\u0632\u0647 \u06a9\u0627\u062f\u0631 \u0645\u062d\u062f\u0648\u062f\u06a9\u0646\u0646\u062f\u0647 \u0631\u0627 \u0628\u0631\u0645\u06cc \u06af\u0631\u062f\u0627\u0646\u062f. \u0627\u06cc\u0646 \u0645\u0642\u0627\u062f\u06cc\u0631 \u0627\u0628\u0639\u0627\u062f \u0628\u0631\u0627\u06cc \u062a\u0631\u0633\u06cc\u0645 \u06cc\u06a9 \u0645\u0633\u062a\u0637\u06cc\u0644 \u067e\u0633 \u0632\u0645\u06cc\u0646\u0647 \u0633\u06cc\u0627\u0647 \u0645\u0648\u0631\u062f \u0627\u0633\u062a\u0641\u0627\u062f\u0647 \u0642\u0631\u0627\u0631 \u0645\u06cc \u06af\u06cc\u0631\u062f \u06a9\u0647 \u0628\u0631\u0686\u0633\u0628 \u062a\u0648\u0633\u0637 \u062a\u0627\u0628\u0639 ()putText \u067e\u0631\u062f\u0627\u0632\u0634 \u0645\u06cc \u0634\u0648\u062f.<\/p>\n\u00a0<\/p>\n
++C<\/strong><\/p>\nvoid draw_label(Mat& input_image, string label, int left, int top)\n{\n \/\/ Display the label at the top of the bounding box.\n int baseLine;\n Size label_size = getTextSize(label, FONT_FACE, FONT_SCALE, THICKNESS, &baseLine);\n top = max(top, label_size.height);\n \/\/ Top left corner.\n Point tlc = Point(left, top);\n \/\/ Bottom right corner.\n Point brc = Point(left + label_size.width, top + label_size.height + baseLine);\n \/\/ Draw white rectangle.\n rectangle(input_image, tlc, brc, BLACK, FILLED);\n \/\/ Put the label on the black rectangle.\n putText(input_image, label, Point(left, top + label_size.height), FONT_FACE, FONT_SCALE, YELLOW, THICKNESS);\n}<\/pre>\nPython<\/strong><\/p>\ndef draw_label(im, label, x, y):\n \"\"\"Draw text onto image at location.\"\"\"\n # Get text size.\n text_size = cv2.getTextSize(label, FONT_FACE, FONT_SCALE, THICKNESS)\n dim, baseline = text_size[0], text_size[1]\n # Use text size to create a BLACK rectangle.\n cv2.rectangle(im, (x,y), (x + dim[0], y + dim[1] + baseline), (0,0,0), cv2.FILLED);\n # Display text inside the rectangle.\n cv2.putText(im, label, (x, y + dim[1]), FONT_FACE, FONT_SCALE, YELLOW, THICKNESS, cv2.LINE_AA)<\/pre>\n\u00a0<\/p>\n
\u06f4-\u06f3-\u06f4- \u067e\u06cc\u0634 \u067e\u0631\u062f\u0627\u0632\u0634<\/strong><\/p>\n\u062a\u0627\u0628\u0639 pre-process \u0634\u0628\u06a9\u0647 \u0648 \u062a\u0635\u0648\u06cc\u0631 \u0631\u0627 \u0628\u0647 \u0639\u0646\u0648\u0627\u0646 \u0622\u0631\u06af\u0648\u0645\u0627\u0646 \u0648\u0631\u0648\u062f\u06cc \u0645\u06cc \u06af\u06cc\u0631\u062f. \u062f\u0631 \u0627\u0628\u062a\u062f\u0627 \u062a\u0635\u0648\u06cc\u0631 \u0628\u0647 \u06cc\u06a9 blob \u0648 \u0633\u067e\u0633 \u0628\u0647 \u0639\u0646\u0648\u0627\u0646 \u0648\u0631\u0648\u062f\u06cc \u0634\u0628\u06a9\u0647 \u062a\u0646\u0638\u06cc\u0645 \u0645\u06cc \u0634\u0648\u062f. \u062a\u0627\u0628\u0639 ()getUnconnectedOutLayerNames \u0646\u0627\u0645 \u0644\u0627\u06cc\u0647 \u0647\u0627\u06cc \u062e\u0631\u0648\u062c\u06cc \u0631\u0627 \u0627\u0631\u0627\u0626\u0647 \u0645\u06cc \u06a9\u0646\u062f \u06a9\u0647 \u062f\u0627\u0631\u0627\u06cc \u0648\u06cc\u0698\u06af\u06cc \u0647\u0627\u06cc \u062a\u0645\u0627\u0645 \u0644\u0627\u06cc\u0647 \u0647\u0627 \u0627\u0633\u062a \u0648 \u0627\u0632 \u0637\u0631\u06cc\u0642 \u0622\u0646 \u0647\u0627 \u062a\u0635\u0648\u06cc\u0631 \u0627\u0632 \u0634\u0628\u06a9\u0647 \u0639\u0628\u0648\u0631 \u06a9\u0631\u062f\u0647 \u062a\u0627 \u062a\u0634\u062e\u06cc\u0635 \u0647\u0627 \u062d\u0627\u0635\u0644 \u06af\u0631\u062f\u062f. \u062f\u0631\u0646\u0647\u0627\u06cc\u062a\u060c \u067e\u0633 \u0627\u0632 \u067e\u0631\u062f\u0627\u0632\u0634\u060c \u0646\u062a\u0627\u06cc\u062c \u062a\u0634\u062e\u06cc\u0635 \u0631\u0627 \u0628\u0631\u0645\u06cc \u06af\u0631\u062f\u0627\u0646\u062f.<\/p>\n
\u00a0<\/p>\n
++C<\/strong><\/p>\nvector<Mat> pre_process(Mat &input_image, Net &net)\n{\n \/\/ Convert to blob.\n Mat blob;\n blobFromImage(input_image, blob, 1.\/255., Size(INPUT_WIDTH, INPUT_HEIGHT), Scalar(), true, false);\n\n net.setInput(blob);\n\n \/\/ Forward propagate.\n vector<Mat> outputs;\n net.forward(outputs, net.getUnconnectedOutLayersNames());\n\n return outputs;\n}<\/pre>\nPython<\/strong><\/p>\ndef pre_process(input_image, net):\n # Create a 4D blob from a frame.\n blob = cv2.dnn.blobFromImage(input_image, 1\/255, (INPUT_WIDTH, INPUT_HEIGHT), [0,0,0], 1, crop=False)\n\n # Sets the input to the network.\n net.setInput(blob)\n\n # Run the forward pass to get output of the output layers.\n outputs = net.forward(net.getUnconnectedOutLayersNames())\n return outputs<\/pre>\n\u00a0<\/p>\n
\u06f5-\u06f3-\u06f4- \u067e\u0633 \u067e\u0631\u062f\u0627\u0632\u0634<\/strong><\/p>\n\u062f\u0631 \u062a\u0627\u0628\u0639 \u0642\u0628\u0644\u06cc pre_process\u060c \u0646\u062a\u0627\u06cc\u062c \u062a\u0634\u062e\u06cc\u0635 \u0631\u0627 \u0628\u0647 \u0639\u0646\u0648\u0627\u0646 \u06cc\u06a9 \u0634\u06cc \u06cc\u0627 object \u062f\u0631\u06cc\u0627\u0641\u062a \u0645\u06cc \u06a9\u0646\u06cc\u0645 \u06a9\u0647 \u0628\u0631\u0627\u06cc \u067e\u0631\u062f\u0627\u0632\u0634 \u0628\u06cc\u0634\u062a\u0631 \u0628\u0627\u06cc\u062f \u0628\u0627\u0632\u0628\u06cc\u0646\u06cc \u0634\u0648\u062f. \u0642\u0628\u0644 \u0627\u0632 \u0628\u062d\u062b \u0628\u06cc\u0634\u062a\u0631 \u062f\u0631 \u0645\u0648\u0631\u062f \u06a9\u062f\u060c \u0627\u062c\u0627\u0632\u0647 \u062f\u0647\u06cc\u062f \u0634\u06a9\u0644 \u0627\u06cc\u0646 \u0634\u06cc \u0648 \u0645\u062d\u062a\u0648\u0627\u06cc \u0622\u0646 \u0631\u0627 \u0628\u0628\u06cc\u0646\u06cc\u0645.<\/p>\n
\u0634\u06cc\u0621 \u0628\u0631\u06af\u0634\u062a\u06cc \u06cc\u06a9 \u0622\u0631\u0627\u06cc\u0647 \u062f\u0648 \u0628\u0639\u062f\u06cc \u0627\u0633\u062a. \u062e\u0631\u0648\u062c\u06cc \u0628\u0633\u062a\u06af\u06cc \u0628\u0647 \u0627\u0646\u062f\u0627\u0632\u0647 \u0648\u0631\u0648\u062f\u06cc \u062f\u0627\u0631\u062f. \u0628\u0647 \u0639\u0646\u0648\u0627\u0646 \u0645\u062b\u0627\u0644\u060c \u0628\u0627 \u0627\u0646\u062f\u0627\u0632\u0647 \u0648\u0631\u0648\u062f\u06cc \u067e\u06cc\u0634 \u0641\u0631\u0636 \u06f6\u06f4\u06f0\u060c \u06cc\u06a9 \u0622\u0631\u0627\u06cc\u0647 \u062f\u0648 \u0628\u0639\u062f\u06cc \u0628\u0627 \u0627\u0646\u062f\u0627\u0632\u0647 \u06f2\u06f5\u06f2\u06f0\u06f0\u00d7\u06f8\u06f5 (\u0631\u062f\u06cc\u0641\u200c\u0647\u0627 \u0648 \u0633\u062a\u0648\u0646\u200c\u0647\u0627) \u062f\u0631\u06cc\u0627\u0641\u062a \u0645\u06cc \u06a9\u0646\u06cc\u0645. \u0631\u062f\u06cc\u0641 \u0647\u0627 \u062a\u0639\u062f\u0627\u062f \u0634\u0646\u0627\u0633\u0627\u06cc\u06cc \u0647\u0627 \u0631\u0627 \u0646\u0634\u0627\u0646 \u0645\u06cc \u062f\u0647\u0646\u062f. \u0628\u0646\u0627\u0628\u0631\u0627\u06cc\u0646 \u0647\u0631 \u0628\u0627\u0631 \u06a9\u0647 \u0634\u0628\u06a9\u0647 \u0627\u062c\u0631\u0627 \u0645\u06cc \u0634\u0648\u062f\u060c \u06f2\u06f5\u06f2\u06f0\u06f0 \u06a9\u0627\u062f\u0631 \u0645\u062d\u062f\u0648\u062f \u06a9\u0646\u0646\u062f\u0647 \u067e\u06cc\u0634 \u0628\u06cc\u0646\u06cc \u0645\u06cc \u0634\u0648\u062f. \u0647\u0631 \u06a9\u0627\u062f\u0631 \u062f\u0627\u0631\u0627\u06cc \u06cc\u06a9 \u0622\u0631\u0627\u06cc\u0647 \u06cc\u06a9 \u0628\u0639\u062f\u06cc \u0627\u0632 \u06f8\u06f5 \u0648\u0631\u0648\u062f\u06cc \u0627\u0633\u062a \u06a9\u0647 \u06a9\u06cc\u0641\u06cc\u062a \u062a\u0634\u062e\u06cc\u0635 \u0631\u0627 \u0646\u0634\u0627\u0646 \u0645\u06cc \u062f\u0647\u062f. \u0627\u06cc\u0646 \u0627\u0637\u0644\u0627\u0639\u0627\u062a \u0628\u0631\u0627\u06cc \u0641\u06cc\u0644\u062a\u0631 \u06a9\u0631\u062f\u0646 \u062a\u0634\u062e\u06cc\u0635 \u0647\u0627\u06cc \u0645\u0648\u0631\u062f \u0646\u0638\u0631 \u06a9\u0627\u0641\u06cc \u0627\u0633\u062a.<\/p>\n\n\n
![\"\u062a\u0634\u062e\u06cc\u0635](\"https:\/\/shahaab-co.com\/mag\/wp-content\/uploads\/2022\/05\/\u062a\u0634\u062e\u06cc\u0635-\u0628\u0627-YOLOv5.jpg\" "\\"\\"")
<\/figure><\/div>\n\n\n\u062f\u0648 \u062c\u0627\u06cc\u06af\u0627\u0647 \u0627\u0648\u0644 \u060c \u0645\u062e\u062a\u0635\u0627\u062a \u0646\u0631\u0645\u0627\u0644 \u0634\u062f\u0647 \u0645\u0631\u06a9\u0632 \u06a9\u0627\u062f\u0631 \u0634\u0646\u0627\u0633\u0627\u06cc\u06cc \u0634\u062f\u0647 \u0647\u0633\u062a\u0646\u062f \u0648 \u0633\u067e\u0633 \u0639\u0631\u0636 \u0648 \u0627\u0631\u062a\u0641\u0627\u0639 \u0646\u0631\u0645\u0627\u0644 \u0634\u062f\u0647. \u0634\u0627\u062e\u0635 \u06f4 \u062f\u0627\u0631\u0627\u06cc \u0627\u0645\u062a\u06cc\u0627\u0632 \u0627\u0637\u0645\u06cc\u0646\u0627\u0646 \u0627\u0633\u062a \u06a9\u0647 \u0627\u062d\u062a\u0645\u0627\u0644 \u0634\u0646\u0627\u0633\u0627\u06cc\u06cc \u06cc\u06a9 \u0634\u0626 \u0631\u0627 \u0646\u0634\u0627\u0646 \u0645\u06cc \u062f\u0647\u062f. \u06f8\u06f0 \u0648\u0631\u0648\u062f\u06cc \u0632\u06cc\u0631 \u0627\u0645\u062a\u06cc\u0627\u0632\u0627\u062a \u06a9\u0644\u0627\u0633 \u06f8\u06f0 \u0634\u06cc \u0627\u0632 \u0645\u062c\u0645\u0648\u0639\u0647 \u062f\u0627\u062f\u0647 COCO 2017 \u0631\u0627 \u0646\u0634\u0627\u0646 \u0645\u06cc \u062f\u0647\u062f \u06a9\u0647 \u0645\u062f\u0644 \u0628\u0631 \u0631\u0648\u06cc \u0622\u0646 \u0647\u0627 \u0622\u0645\u0648\u0632\u0634 \u062f\u0627\u062f\u0647 \u0634\u062f\u0647 \u0627\u0633\u062a.<\/p>\n
\u0648\u0627\u0642\u0639\u06cc\u062a \u062c\u0627\u0644\u0628<\/strong>: \u0645\u062c\u0645\u0648\u0639\u0647 \u062f\u0627\u062f\u0647 COCO 2017 \u062f\u0631 \u0645\u062c\u0645\u0648\u0639 \u06f9\u06f1 \u0634\u06cc \u062f\u0627\u0631\u062f. \u0628\u0627 \u0627\u06cc\u0646 \u062d\u0627\u0644\u060c \u06f1\u06f1 \u0634\u06cc \u0647\u0646\u0648\u0632 \u0628\u0631\u0686\u0633\u0628 \u0646\u062f\u0627\u0631\u0646\u062f.<\/p>\n\u0627\u0644\u0641)<\/strong> \u0641\u06cc\u0644\u062a\u0631 \u062a\u0634\u062e\u06cc\u0635 \u0647\u0627\u06cc \u062f\u0631\u0633\u062a<\/p>\n\u062f\u0631 \u062d\u06cc\u0646 \u0628\u0627\u0632 \u06a9\u0631\u062f\u0646\u060c \u0628\u0627\u06cc\u062f \u0645\u0631\u0627\u0642\u0628 \u0634\u06a9\u0644 \u0647\u0627 \u0628\u0627\u0634\u06cc\u0645. \u0628\u0627 OpenCV-Python 4.5.5\u060c \u06cc\u06a9 tuple \u0627\u0632 \u06cc\u06a9 \u0622\u0631\u0627\u06cc\u0647 \u0633\u0647 \u0628\u0639\u062f\u06cc \u0628\u0647 \u0627\u0646\u062f\u0627\u0632\u0647 \u06f1x row x column<\/strong> \u0627\u0633\u062a. \u0628\u0627\u06cc\u062f row x column<\/strong> \u0628\u0627\u0634\u062f. \u0628\u0646\u0627\u0628\u0631\u0627\u06cc\u0646\u060c \u0622\u0631\u0627\u06cc\u0647 \u0627\u0632 \u0634\u0627\u062e\u0635 \u0635\u0641\u0631 \u0642\u0627\u0628\u0644 \u062f\u0633\u062a\u0631\u0633\u06cc \u0627\u0633\u062a. \u0627\u06cc\u0646 \u0645\u0648\u0631\u062f \u062f\u0631 ++C\u00a0 \u0631\u0639\u0627\u06cc\u062a \u0646\u0645\u06cc \u0634\u0648\u062f.<\/p>\n\u0634\u0628\u06a9\u0647 \u0645\u062e\u062a\u0635\u0627\u062a \u062e\u0631\u0648\u062c\u06cc \u0631\u0627 \u0628\u0631 \u0627\u0633\u0627\u0633 \u0627\u0646\u062f\u0627\u0632\u0647 \u0648\u0631\u0648\u062f\u06cc blob\u060c \u06cc\u0639\u0646\u06cc \u06f6\u06f4\u06f0 \u062a\u0648\u0644\u06cc\u062f \u0645\u06cc \u06a9\u0646\u062f. \u0628\u0646\u0627\u0628\u0631\u0627\u06cc\u0646\u060c \u0645\u062e\u062a\u0635\u0627\u062a \u0628\u0627\u06cc\u062f \u062f\u0631 \u0641\u0627\u06a9\u062a\u0648\u0631\u0647\u0627\u06cc \u062a\u063a\u06cc\u06cc\u0631 \u0627\u0646\u062f\u0627\u0632\u0647 \u0636\u0631\u0628 \u0634\u0648\u062f \u062a\u0627 \u062e\u0631\u0648\u062c\u06cc \u0648\u0627\u0642\u0639\u06cc \u0628\u0647 \u062f\u0633\u062a \u0622\u06cc\u062f. \u0645\u0631\u0627\u062d\u0644 \u0632\u06cc\u0631 \u0628\u0631\u0627\u06cc \u0628\u0627\u0632\u06a9\u0631\u062f\u0646 \u062a\u0634\u062e\u06cc\u0635 \u0647\u0627 \u0627\u0646\u062c\u0627\u0645 \u0645\u06cc \u06af\u06cc\u0631\u062f:<\/p>\n
\u00a0<\/p>\n
\n- \u062d\u0644\u0642\u0647 \u0627\u0632 \u0637\u0631\u06cc\u0642 \u062a\u0634\u062e\u06cc\u0635<\/li>\n
- \u0641\u06cc\u0644\u062a\u0631 \u062a\u0634\u062e\u06cc\u0635 \u0647\u0627\u06cc \u062f\u0631\u0633\u062a<\/li>\n
- \u062f\u0631\u06cc\u0627\u0641\u062a \u0634\u0627\u062e\u0635 \u0628\u0647\u062a\u0631\u06cc\u0646 \u0646\u0645\u0631\u0647 \u06a9\u0644\u0627\u0633<\/li>\n
- \u062d\u0630\u0641 \u062a\u0634\u062e\u06cc\u0635 \u0647\u0627\u06cc\u06cc \u06a9\u0647 \u0627\u0645\u062a\u06cc\u0627\u0632 \u06a9\u0644\u0627\u0633 \u0634\u0627\u0646 \u0627\u0632 \u0645\u0642\u062f\u0627\u0631 \u0622\u0633\u062a\u0627\u0646\u0647 \u06a9\u0645\u062a\u0631 \u0627\u0633\u062a<\/li>\n<\/ol>\n
++C<\/strong><\/p>\nMat post_process(Mat &input_image, vector<Mat> &outputs, const vector<string> &class_name)\n{\n \/\/ Initialize vectors to hold respective outputs while unwrapping detections.\n vector<int> class_ids;\n vector<float> confidences;\n vector<Rect> boxes;\n \/\/ Resizing factor.\n float x_factor = input_image.cols \/ INPUT_WIDTH;\n float y_factor = input_image.rows \/ INPUT_HEIGHT;\n float *data = (float *)outputs[0].data;\n const int dimensions = 85;\n \/\/ \u06f2\u06f5\u06f2\u06f0\u06f0 for default size 640.\n const int rows = 25200;\n \/\/ Iterate through 25200 detections.\n for (int i = 0; i < rows; ++i)\n {\n float confidence = data[4];\n \/\/ Discard bad detections and continue.\n if (confidence >= CONFIDENCE_THRESHOLD)\n {\n float * classes_scores = data + 5;\n \/\/ Create a 1x85 Mat and store class scores of 80 classes.\n Mat scores(1, class_name.size(), CV_32FC1, classes_scores);\n \/\/ Perform minMaxLoc and acquire the index of best class score.\n Point class_id;\n double max_class_score;\n minMaxLoc(scores, 0, &max_class_score, 0, &class_id);\n \/\/ Continue if the class score is above the threshold.\n if (max_class_score > SCORE_THRESHOLD)\n {\n \/\/ Store class ID and confidence in the pre-defined respective vectors.\n confidences.push_back(confidence);\n class_ids.push_back(class_id.x);\n \/\/ Center.\n float cx = data[0];\n float cy = data[1];\n \/\/ Box dimension.\n float w = data[2];\n float h = data[3];\n \/\/ Bounding box coordinates.\n int left = int((cx - 0.5 * w) * x_factor);\n int top = int((cy - 0.5 * h) * y_factor);\n int width = int(w * x_factor);\n int height = int(h * y_factor);\n \/\/ Store good detections in the boxes vector.\n boxes.push_back(Rect(left, top, width, height));\n }\n }\n \/\/ Jump to the next row.\n data += 85;\n }<\/pre>\nPython<\/strong><\/p>\ndef post_process(input_image, outputs):\n # Lists to hold respective values while unwrapping.\n class_ids = []\n confidences = []\n boxes = []\n # Rows.\n rows = outputs[0].shape[1]\n image_height, image_width = input_image.shape[:2]\n # Resizing factor.\n x_factor = image_width \/ INPUT_WIDTH\n y_factor = image_height \/ INPUT_HEIGHT\n # Iterate through detections.\n for r in range(rows):\n row = outputs[0][0][r]\n confidence = row[4]\n # Discard bad detections and continue.\n if confidence >= CONFIDENCE_THRESHOLD:\n classes_scores = row[5:]\n # Get the index of max class score.\n class_id = np.argmax(classes_scores)\n # Continue if the class score is above threshold.\n if (classes_scores[class_id] > SCORE_THRESHOLD):\n confidences.append(confidence)\n class_ids.append(class_id)\n cx, cy, w, h = row[0], row[1], row[2], row[3]\n left = int((cx - w\/2) * x_factor)\n top = int((cy - h\/2) * y_factor)\n width = int(w * x_factor)\n height = int(h * y_factor)\n box = np.array([left, top, width, height])\n boxes.append(box)<\/pre>\n\u00a0<\/p>\n
\u0628) <\/strong>\u062d\u0630\u0641 \u06a9\u0627\u062f\u0631 \u0647\u0627\u06cc \u0647\u0645\u067e\u0648\u0634\u0627\u0646<\/p>\n\u067e\u0633 \u0627\u0632 \u0641\u06cc\u0644\u062a\u0631 \u06a9\u0631\u062f\u0646 \u062a\u0634\u062e\u06cc\u0635 \u0647\u0627\u06cc \u062f\u0631\u0633\u062a\u060c \u00a0\u06a9\u0627\u062f\u0631 \u0647\u0627\u06cc \u0645\u062d\u062f\u0648\u062f\u06a9\u0646\u0646\u062f\u0647 \u0645\u0648\u0631\u062f \u0646\u0638\u0631 \u0631\u0627 \u062e\u0648\u0627\u0647\u06cc\u0645 \u062f\u0627\u0634\u062a. \u0628\u0627 \u0627\u06cc\u0646 \u062d\u0627\u0644\u060c \u0645\u0645\u06a9\u0646 \u0627\u0633\u062a \u0686\u0646\u062f\u06cc\u0646 \u06a9\u0627\u062f\u0631 \u0645\u062d\u062f\u0648\u062f \u06a9\u0646\u0646\u062f\u0647 \u0647\u0645\u067e\u0648\u0634\u0627\u0646\u06cc \u0628\u0647 \u0634\u06a9\u0644 \u0632\u06cc\u0631 \u0648\u062c\u0648\u062f \u062f\u0627\u0634\u062a\u0647 \u0628\u0627\u0634\u062f.<\/p>\n\n\n
![\"\u062d\u0630\u0641](\"https:\/\/shahaab-co.com\/mag\/wp-content\/uploads\/2022\/05\/\u062d\u0630\u0641-\u06a9\u0627\u062f\u0631-\u0647\u0627\u06cc-\u0645\u062d\u0635\u0648\u0631-\u06a9\u0646\u0646\u062f\u0647-\u0647\u0645\u067e\u0648\u0634\u0627\u0646.jpg\" "\\"\\"")
<\/figure><\/div>\n\n\n\u06a9\u0647 \u0627\u06cc\u0646 \u0645\u0634\u06a9\u0644 \u0628\u0627 \u0627\u0646\u062c\u0627\u0645 Non-Maximum Suppression \u062d\u0644 \u0645\u06cc \u0634\u0648\u062f. \u062a\u0627\u0628\u0639 ()NMSBoxes \u0644\u06cc\u0633\u062a\u06cc \u0627\u0632 \u06a9\u0627\u062f\u0631 \u0647\u0627 \u0631\u0627 \u0645\u06cc \u06af\u06cc\u0631\u062f\u060c (\u0627\u0634\u062a\u0631\u0627\u06a9 \u0628\u0631 \u0627\u062c\u062a\u0645\u0627\u0639)IOU \u0631\u0627 \u0645\u062d\u0627\u0633\u0628\u0647 \u0645\u06cc \u06a9\u0646\u062f\u060c \u0648 \u062a\u0635\u0645\u06cc\u0645 \u0645\u06cc \u06af\u06cc\u0631\u062f \u06a9\u0647 \u06a9\u0627\u062f\u0631 \u0647\u0627 \u0631\u0627 \u0648\u0627\u0628\u0633\u062a\u0647 \u0628\u0647 NMS_THRESHOLD \u0646\u06af\u0647 \u062f\u0627\u0631\u062f. \u0628\u0631\u0627\u06cc \u06a9\u0633\u0628 \u0627\u0637\u0644\u0627\u0639\u0627\u062a \u0628\u06cc\u0634\u062a\u0631\u060c \u0645\u0642\u0627\u0644\u0647 \u062f\u0631 \u0645\u0648\u0631\u062f NMS \u0631\u0627 \u0628\u0631\u0631\u0633\u06cc \u06a9\u0646\u06cc\u062f.<\/p>\n
\u00a0<\/strong><\/p>\n++C<\/strong><\/p>\n \/\/ Perform Non-Maximum Suppression and draw predictions.\n vector<int> indices;\n NMSBoxes(boxes, confidences, SCORE_THRESHOLD, NMS_THRESHOLD, indices);\n for (int i = 0; i < indices.size(); i++)\n {\n int idx = indices[i];\n Rect box = boxes[idx];\n int left = box.x;\n int top = box.y;\n int width = box.width;\n int height = box.height;\n \/\/ Draw bounding box.\n rectangle(input_image, Point(left, top), Point(left + width, top + height), BLUE, 3*THICKNESS);\n \/\/ Get the label for the class name and its confidence.\n string label = format(\"%.2f\", confidences[idx]);\n label = class_name[class_ids[idx]] + \":\" + label;\n \/\/ Draw class labels.\n draw_label(input_image, label, left, top);\n }\n return input_image;\n}<\/pre>\nPython<\/strong><\/p>\n# Perform non maximum suppression to eliminate redundant, overlapping boxes with lower confidences.\n indices = cv2.dnn.NMSBoxes(boxes, confidences, CONFIDENCE_THRESHOLD, NMS_THRESHOLD)\n for i in indices:\n box = boxes[i]\n left = box[0]\n top = box[1]\n width = box[2]\n height = box[3] \n # Draw bounding box. \n cv2.rectangle(input_image, (left, top), (left + width, top + height), BLUE, 3*THICKNESS)\n # Class label. \n label = \"{}:{:.2f}\".format(classes[class_ids[i]], confidences[i]) \n # Draw label. \n draw_label(input_image, label, left, top)\n return input_image<\/pre>\n