{"id":38875,"date":"2024-11-02T01:39:43","date_gmt":"2024-11-02T01:39:43","guid":{"rendered":"http:\/\/atmokpo.com\/w\/?p=38875"},"modified":"2024-11-26T06:43:06","modified_gmt":"2024-11-26T06:43:06","slug":"yolo-%eb%aa%a8%eb%8d%b8%ec%9d%98-%eb%8b%a4%ec%96%91%ed%95%9c-%ec%9d%91%ec%9a%a9-%ec%96%bc%ea%b5%b4-%ea%b2%80%ec%b6%9c-%ec%b0%a8%eb%9f%89-%eb%b2%88%ed%98%b8%ed%8c%90-%ec%9d%b8%ec%8b%9d-%eb%93%b1","status":"publish","type":"post","link":"https:\/\/atmokpo.com\/w\/38875\/","title":{"rendered":"YOLO \ubaa8\ub378\uc758 \ub2e4\uc591\ud55c \uc751\uc6a9, \uc5bc\uad74 \uac80\ucd9c, \ucc28\ub7c9 \ubc88\ud638\ud310 \uc778\uc2dd \ub4f1\uc758 \ud2b9\uc815 \uac1d\uccb4 \uac80\ucd9c \uc0ac\ub840"},"content":{"rendered":"

<\/p>\n

YOLO (You Only Look Once)\ub294 \uac1d\uccb4 \ud0d0\uc9c0(Object Detection) \ubd84\uc57c\uc5d0\uc11c \uac00\uc7a5 \uc778\uae30 \uc788\ub294 \uc54c\uace0\ub9ac\uc998 \uc911 \ud558\ub098\uc785\ub2c8\ub2e4. YOLO\ub294 \uc774\ubbf8\uc9c0\ub098 \ub3d9\uc601\uc0c1\uc5d0\uc11c \uac1d\uccb4\ub97c \uc2e4\uc2dc\uac04\uc73c\ub85c \uac80\ucd9c\ud560 \uc218 \uc788\ub294 \ub2a5\ub825 \ub355\ubd84\uc5d0 \ub9ce\uc740 \uc5f0\uad6c \ubc0f \uc0c1\uc5c5\uc801 \uc751\uc6a9\uc5d0\uc11c \uc0ac\uc6a9\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4. \uc774\ubc88 \uae00\uc5d0\uc11c\ub294 YOLO \ubaa8\ub378\uc744 \uc774\uc6a9\ud55c \ub2e4\uc591\ud55c \uc751\uc6a9 \ubc0f \uc0ac\ub840\uc5d0 \ub300\ud574 \uc54c\uc544\ubcf4\uaca0\uc2b5\ub2c8\ub2e4. \ud2b9\ud788, \uc5bc\uad74 \uac80\ucd9c\uacfc \ucc28\ub7c9 \ubc88\ud638\ud310 \uc778\uc2dd \uc0ac\ub840\ub97c \uc911\uc2ec\uc73c\ub85c \ub17c\uc758\ud558\uaca0\uc2b5\ub2c8\ub2e4.<\/p>\n

YOLO \ubaa8\ub378 \uac1c\uc694<\/h2>\n

YOLO\ub294 \ud2b9\uc815 \uac1d\uccb4\ub97c \uac80\ucd9c\ud558\uae30 \uc704\ud574 \ub2e8\uc77c \uc2e0\uacbd\ub9dd\uc744 \uc0ac\uc6a9\ud558\uc5ec \uc774\ubbf8\uc9c0\ub97c \uaca9\uc790\ub85c \ub098\ub204\uace0 \uac01 \uaca9\uc790\uc5d0\uc11c \uac1d\uccb4\uc758 \uc704\uce58 \ubc0f \ud074\ub798\uc2a4 \ud655\ub960\uc744 \uc608\uce21\ud569\ub2c8\ub2e4. YOLO\ub294 \ub2e4\ub978 \uac1d\uccb4 \ud0d0\uc9c0 \uc54c\uace0\ub9ac\uc998\ubcf4\ub2e4 \ub6f0\uc5b4\ub09c \uc18d\ub3c4\uc640 \uc815\ud655\uc131\uc744 \uc81c\uacf5\ud569\ub2c8\ub2e4. YOLOv3, YOLOv4, YOLOv5 \ub4f1 \uc5ec\ub7ec \ubc84\uc804\uc774 \uc788\uc73c\uba70, \uac01 \ubc84\uc804\uc740 \uc815\ud655\ub3c4 \ubc0f \uc18d\ub3c4 \uba74\uc5d0\uc11c \uac1c\uc120\ub418\uc5c8\uc2b5\ub2c8\ub2e4.<\/p>\n

1. \uc5bc\uad74 \uac80\ucd9c<\/h2>\n

\uc5bc\uad74 \uac80\ucd9c\uc740 \ub2e4\uc591\ud55c \uc560\ud50c\ub9ac\ucf00\uc774\uc158\uc5d0\uc11c \uc0ac\uc6a9\ub429\ub2c8\ub2e4. \uc608\ub97c \ub4e4\uc5b4, \ubcf4\uc548 \uc2dc\uc2a4\ud15c, \uc0ac\uc9c4 \ucd2c\uc601, \uc0ac\uc6a9\uc790 \uacbd\ud5d8 \uac1c\uc120 \ub4f1\uc774 \uc788\uc2b5\ub2c8\ub2e4. YOLO \ubaa8\ub378\uc740 \uc5bc\uad74 \uac80\ucd9c\uc5d0\uc11c \ub192\uc740 \uc131\ub2a5\uc744 \ubcf4\uc5ec\uc8fc\uba70, \uc544\ub798\uc640 \uac19\uc774 \uc5bc\uad74 \uac80\ucd9c\uc744 \uad6c\ud604\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n

1.1 YOLO\ub97c \uc774\uc6a9\ud55c \uc5bc\uad74 \uac80\ucd9c \uad6c\ud604<\/h3>\n

YOLO \ubaa8\ub378\uc744 \uc0ac\uc6a9\ud558\uc5ec \uc5bc\uad74\uc744 \uac80\ucd9c\ud558\uae30 \uc704\ud574 \uba3c\uc800 YOLO \ubaa8\ub378\uc744 \ub2e4\uc6b4\ub85c\ub4dc\ud558\uace0 \ud544\uc694\ud55c \ub77c\uc774\ube0c\ub7ec\ub9ac\ub97c \uc124\uce58\ud574\uc57c \ud569\ub2c8\ub2e4.<\/p>\n

import cv2\nimport numpy as np\n\n# YOLO \ubaa8\ub378 \uc124\uc815 \ubc0f \uac00\uc911\uce58 \ubd88\ub7ec\uc624\uae30\nnet = cv2.dnn.readNet(\"yolov3.weights\", \"yolov3.cfg\")\nlayer_names = net.getLayerNames()\noutput_layers = [layer_names[i[0] - 1] for i in net.getUnconnectedOutLayers()]\n\n# \uc774\ubbf8\uc9c0 \ub85c\ub4dc\nimg = cv2.imread(\"image.jpg\")\nheight, width, _ = img.shape\n\n# \uc774\ubbf8\uc9c0Blob \uc0dd\uc131\nblob = cv2.dnn.blobFromImage(img, 0.00392, (416, 416), (0, 0, 0), True, crop=False)\nnet.setInput(blob)\nouts = net.forward(output_layers)\n\n# \uac80\ucd9c\ub41c \uc5bc\uad74 \uadf8\ub9ac\uae30\nclass_ids = []\nconfidences = []\nboxes = []\nfor out in outs:\n    for detection in out:\n        scores = detection[5:]\n        class_id = np.argmax(scores)\n        confidence = scores[class_id]\n        if confidence > 0.5:\n            # \uac1d\uccb4\uc758 \uacbd\uacc4 \uc0c1\uc790\n            center_x = int(detection[0] * width)\n            center_y = int(detection[1] * height)\n            w = int(detection[2] * width)\n            h = int(detection[3] * height)\n\n            # \uc0c1\uc790 \uc88c\ud45c\n            x = int(center_x - w \/ 2)\n            y = int(center_y - h \/ 2)\n\n            boxes.append([x, y, w, h])\n            confidences.append(float(confidence))\n            class_ids.append(class_id)\n\n# \ube44\ucd5c\ub300 \uc5b5\uc81c(NMS) \uc801\uc6a9\nindexes = cv2.dnn.NMSBoxes(boxes, confidences, 0.5, 0.4)\n\n# \uacb0\uacfc \ud45c\uc2dc\nfor i in range(len(boxes)):\n    if i in indexes:\n        x, y, w, h = boxes[i]\n        label = str(classes[class_ids[i]])\n        cv2.rectangle(img, (x, y), (x + w, y + h), (255, 0, 0), 2)\n        cv2.putText(img, label, (x, y + 30), cv2.FONT_HERSHEY_PLAIN, 3, (255, 0, 0), 3)\n\ncv2.imshow(\"Image\", img)\ncv2.waitKey(0)\ncv2.destroyAllWindows()<\/code><\/pre>\n
1.2 YOLO \ubaa8\ub378 \ud6c8\ub828\uc744 \uc704\ud55c \ub370\uc774\ud130\uc14b \uc900\ube44<\/h3>\n
\uc5bc\uad74 \uac80\ucd9c\uc744 \uc704\ud574\uc11c\ub294 \ub77c\ubca8\uc774 \uc8fc\uc5b4\uc9c4 \ub370\uc774\ud130\uc14b\uc774 \ud544\uc694\ud569\ub2c8\ub2e4. \ub370\uc774\ud130\uc14b\uc740 \uc774\ubbf8\uc9c0\uc640 \ud568\uaed8 \uc5bc\uad74\uc758 \uc704\uce58 \uc815\ubcf4\ub97c \ud3ec\ud568\ud574\uc57c \ud558\uba70, \uc77c\ubc18\uc801\uc73c\ub85c Pascal VOC, COCO \ud615\uc2dd\uc73c\ub85c \ud45c\uc2dc\ub429\ub2c8\ub2e4. \uc608\ub97c \ub4e4\uc5b4, OpenCV \ub77c\uc774\ube0c\ub7ec\ub9ac\ub97c \uc0ac\uc6a9\ud558\uc5ec \uc5bc\uad74 \uc778\uc2dd\uc774 \ud544\uc694\ud55c \uc774\ubbf8\uc9c0\uc5d0\uc11c \ub77c\ubca8\uc744 \ubd80\uc5ec\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n
2. \ucc28\ub7c9 \ubc88\ud638\ud310 \uc778\uc2dd<\/h2>\n
\ucc28\ub7c9 \ubc88\ud638\ud310 \uc778\uc2dd\uc740 \uc8fc\ucc28 \uad00\ub9ac, \uad50\ud1b5 \ub2e8\uc18d \ub4f1\uc758 \ubd84\uc57c\uc5d0\uc11c \uc911\uc694\ud55c \uc5ed\ud560\uc744 \ud569\ub2c8\ub2e4. YOLO \ubaa8\ub378\uc740 \ucc28\ub7c9 \ubc88\ud638\ud310 \uac80\ucd9c\uc744 \ud1b5\ud574 \ub9e4\uc6b0 \ub192\uc740 \uc131\ub2a5\uc744 \ubcf4\uc5ec\uc90d\ub2c8\ub2e4. \ub2e4\uc74c\uc740 \ucc28\ub7c9 \ubc88\ud638\ud310 \uc778\uc2dd\uc744 \uc704\ud55c YOLO\ub97c \uc0ac\uc6a9\ud558\ub294 \uc608\uc81c\uc785\ub2c8\ub2e4.<\/p>\n
2.1 \ucc28\ub7c9 \ubc88\ud638\ud310 \uc778\uc2dd \uae30\ubcf8 \uad6c\ud604<\/h3>\n
\ucc28\ub7c9 \ubc88\ud638\ud310 \uc778\uc2dd\uc744 \uad6c\ud604\ud558\uae30 \uc704\ud574 \uc544\ub798\uc640 \uac19\uc774 YOLO \ubaa8\ub378\uc744 \uc0ac\uc6a9\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n
# \ucc28\ub7c9 \ubc88\ud638\ud310 \ub370\uc774\ud130\uc14b \ubd88\ub7ec\uc624\uae30\nlabels = open(\"obj.names\").read().strip().split(\"\\n\")\n\n# YOLO \ucd08\uae30\ud654\nnet = cv2.dnn.readNet(\"yolov3.weights\", \"yolov3.cfg\")\nlayer_names = net.getLayerNames()\noutput_layers = [layer_names[i[0] - 1] for i in net.getUnconnectedOutLayers()]\n\n# \uc774\ubbf8\uc9c0 \ub85c\ub4dc\nimg = cv2.imread(\"car.jpg\")\nheight, width, _ = img.shape\n\n# Blob \uc0dd\uc131 \ubc0f \ub124\ud2b8\uc6cc\ud06c\uc5d0 \uc4f0\uae30\nblob = cv2.dnn.blobFromImage(img, 0.00392, (416, 416), (0, 0, 0), True, crop=False)\nnet.setInput(blob)\nouts = net.forward(output_layers)\n\n# \ubc88\ud638\ud310 \uac80\ucd9c \ubc0f \uc2dc\uac01\ud654\nclass_ids = []\nconfidences = []\nboxes = []\nfor out in outs:\n    for detection in out:\n        scores = detection[5:]\n        class_id = np.argmax(scores)\n        confidence = scores[class_id]\n        if confidence > 0.5:  # \uc2e0\ub8b0\ub3c4 \uae30\uc900 \uc124\uc815\n            center_x = int(detection[0] * width)\n            center_y = int(detection[1] * height)\n            w = int(detection[2] * width)\n            h = int(detection[3] * height)\n\n            x = int(center_x - w \/ 2)\n            y = int(center_y - h \/ 2)\n\n            boxes.append([x, y, w, h])\n            confidences.append(float(confidence))\n            class_ids.append(class_id)\n\nindexes = cv2.dnn.NMSBoxes(boxes, confidences, 0.5, 0.4)\n\nfor i in range(len(boxes)):\n    if i in indexes:\n        x, y, w, h = boxes[i]\n        label = str(labels[class_ids[i]])\n        cv2.rectangle(img, (x, y), (x + w, y + h), (0, 255, 0), 2)\n        cv2.putText(img, label, (x, y + 30), cv2.FONT_HERSHEY_PLAIN, 3, (0, 255, 0), 3)\n\ncv2.imshow(\"Image\", img)\ncv2.waitKey(0)\ncv2.destroyAllWindows()<\/code><\/pre>\n
2.2 \ucc28\ub7c9 \ubc88\ud638\ud310 \ub370\uc774\ud130\uc14b \uc900\ube44<\/h3>\n
\ucc28\ub7c9 \ubc88\ud638\ud310 \uc778\uc2dd\uc744 \uc704\ud574\uc11c\ub294 \ub2e4\uc591\ud55c \ucc28\ub7c9\uc758 \ubc88\ud638\ud310 \uc774\ubbf8\uc9c0\ub97c \ud3ec\ud568\ud558\ub294 \ub370\uc774\ud130\uc14b\uc774 \ud544\uc694\ud569\ub2c8\ub2e4. \uac01 \uc774\ubbf8\uc9c0\ub294 \ucc28\ub7c9 \ubc88\ud638\ud310\uc5d0 \ub300\ud55c \uc704\uce58\uc640 \ub77c\ubca8 \uc815\ubcf4\ub97c \ud3ec\ud568\ud574\uc57c \ud558\ub294\ub370, \uc77c\ubc18\uc801\uc73c\ub85c \ud615\uc2dd\uc740 JSON\uc774\ub098 XML \ud615\ud0dc\ub85c \uc800\uc7a5\ud569\ub2c8\ub2e4.<\/p>\n
3. YOLO\uc758 \ubc1c\uc804\uacfc \uc751\uc6a9<\/h2>\n
YOLO\uc758 \ucd5c\uc2e0 \ubc84\uc804\ub4e4\uc740 \uc774\ubbf8\uc9c0\ub97c \ucc98\ub9ac\ud558\ub294 \uc18d\ub3c4\ub97c \ud06c\uac8c \ud5a5\uc0c1\uc2dc\ucf30\uc73c\uba70, \ub354 \uc815\ubc00\ud55c \uac1d\uccb4 \ud0d0\uc9c0\ub97c \uc9c0\uc6d0\ud558\uac8c \ub418\uc5c8\uc2b5\ub2c8\ub2e4. YOLOv4\uc640 YOLOv5\ub294 \ud2b9\ud788 \uc131\ub2a5\uc774 \ud5a5\uc0c1\ub418\uc5b4 \uc2e4\uc2dc\uac04 \uac1d\uccb4 \ud0d0\uc9c0 \ubd84\uc57c\uc5d0\uc11c \ub9ce\uc774 \uc0ac\uc6a9\ub418\uace0 \uc788\uc2b5\ub2c8\ub2e4. \uc774\ub7ec\ud55c \ubc1c\uc804\uc740 \ub2e4\uc591\ud55c \uc0b0\uc5c5\uc5d0 \uc751\uc6a9\ub418\uace0 \uc788\ub294\ub370, \ubcf4\uc548, \uc2a4\ud3ec\uce20 \ubd84\uc11d, HTML \ub0b4 \uace0\ud654\uc9c8 \ube44\ub514\uc624 \ubd84\uc11d \ub4f1\uc758 \ubd84\uc57c\uc5d0\uc11c \ud65c\uc6a9\ub418\uace0 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n
4. \uacb0\ub860<\/h2>\n
YOLO \ubaa8\ub378\uc740 \uac1d\uccb4 \ud0d0\uc9c0 \ubd84\uc57c\uc5d0\uc11c \ub9e4\uc6b0 \uc720\uc6a9\ud55c \ub3c4\uad6c\uc785\ub2c8\ub2e4. \ub2e4\uc591\ud55c \uc751\uc6a9 \ud504\ub85c\uadf8\ub7a8\uc5d0\uc11c \ud6a8\uacfc\uc801\uc73c\ub85c \uc0ac\uc6a9\ub420 \uc218 \uc788\uc73c\uba70, \ud2b9\ud788 \uc5bc\uad74 \uac80\ucd9c \ubc0f \ucc28\ub7c9 \ubc88\ud638\ud310 \uc778\uc2dd\uacfc \uac19\uc740 \ud2b9\uc815 \uac1d\uccb4 \uac80\ucd9c\uc5d0 \ub192\uc740 \uc131\ub2a5\uc744 \ubcf4\uc5ec\uc8fc\uace0 \uc788\uc2b5\ub2c8\ub2e4. \uc6b0\ub9ac\uac00 \ud0d0\uad6c\ud55c \uc0ac\ub840\ub294 YOLO\uc758 \uac00\ub2a5\uc131\uc744 \ubcf4\uc5ec\uc8fc\uba70, \ud5a5\ud6c4 \uc5f0\uad6c \ubc0f \uac1c\ubc1c\uc5d0\uc11c \ub354 \ub9ce\uc740 \ubc1c\uc804\uc744 \uc774\ub8f0 \uc218 \uc788\uc744 \uac83\uc785\ub2c8\ub2e4.<\/p>\n
\ucc38\uace0 \ubb38\ud5cc<\/h2>\n