YOLO(You Only Look Once)\ub294 \uac1d\uccb4 \uac10\uc9c0 \ubd84\uc57c\uc5d0\uc11c \ub110\ub9ac \uc0ac\uc6a9\ub418\ub294 \ub525\ub7ec\ub2dd \ubaa8\ub378\uc785\ub2c8\ub2e4. \ud2b9\ud788, YOLO\ub294 \uc774\ubbf8\uc9c0 \uc804\uccb4\ub97c \ud55c \ubc88\uc5d0 \ucc98\ub9ac\ud558\uc5ec \ube60\ub978 \uc18d\ub3c4\ub85c \uac1d\uccb4\ub97c \uac10\uc9c0\ud560 \uc218 \uc788\ub2e4\ub294 \uc7a5\uc810\uc774 \uc788\uc2b5\ub2c8\ub2e4. \ubcf8 \uae00\uc5d0\uc11c\ub294 YOLO \ubaa8\ub378\uc758 \uc804\uc774 \ud559\uc2b5\uc744 \ud1b5\ud55c \uc131\ub2a5 \ud5a5\uc0c1 \ubc29\ubc95\uacfc \ub370\uc774\ud130 \uc99d\uac15 \uae30\ubc95\uc744 \ud65c\uc6a9\ud55c \ubc29\ubc95\uc5d0 \ub300\ud574 \uc790\uc138\ud788 \ub2e4\ub8e8\uaca0\uc2b5\ub2c8\ub2e4.<\/p>\n
\uc804\uc774 \ud559\uc2b5\uc774\ub780 \uae30\uc874\uc5d0 \ud559\uc2b5\ub41c \ubaa8\ub378\uc744 \uae30\ubc18\uc73c\ub85c \uc0c8\ub85c\uc6b4 \uc791\uc5c5\uc5d0 \ud65c\uc6a9\ud558\ub294 \uae30\ubc95\uc785\ub2c8\ub2e4. \uc77c\ubc18\uc801\uc73c\ub85c \ub300\ub7c9\uc758 \ub370\uc774\ud130\ub85c \ud559\uc2b5\ub41c \ubaa8\ub378\uc740 \ub9ce\uc740 \ud2b9\uc9d5\uc744 \ucd94\ucd9c\ud574 \ub193\uc2b5\ub2c8\ub2e4. \uc774\ub7ec\ud55c \ubaa8\ub378\uc744 \uc0ac\uc6a9\ud558\uc5ec \ube44\uc2b7\ud55c \uc791\uc5c5\uc744 \ud558\ub294 \ub370 \ud544\uc694\ud55c \uc2dc\uac04\uc744 \uc904\uc774\uace0, \uc801\uc740 \ub370\uc774\ud130\ub85c\ub3c4 \uc88b\uc740 \uc131\ub2a5\uc744 \uc5bb\uc744 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n
YOLO \ubaa8\ub378\uc740 \ud2b9\ubcc4\ud788 \ub9ce\uc740 \ub370\uc774\ud130\uc14b\uc5d0\uc11c \ud559\uc2b5\ub418\uc5b4 \uc788\uc73c\uba70, \uc774 \ubaa8\ub378\uc744 \ubc14\ud0d5\uc73c\ub85c \uc804\uc774 \ud559\uc2b5\uc744 \uc218\ud589\ud558\uba74 \uc0c8\ub85c\uc6b4 \ub370\uc774\ud130\uc14b\uc5d0 \ub300\ud574\uc11c\ub3c4 \ub192\uc740 \uc815\ud655\ub3c4\ub97c \uc5bb\uc744 \uc218 \uc788\uc2b5\ub2c8\ub2e4. \uc5ec\uae30\uc5d0\uc11c\ub294 YOLOv5\ub97c \uc0ac\uc6a9\ud55c \uc804\uc774 \ud559\uc2b5\uc758 \uc608\uc81c\ub97c \ubcf4\uc5ec\ub4dc\ub9ac\uaca0\uc2b5\ub2c8\ub2e4.<\/p>\n
!git clone https:\/\/github.com\/ultralytics\/yolov5\n%cd yolov5\n!pip install -r requirements.txt\n<\/code><\/pre>\n\uc774\uc81c YOLOv5 \ub808\ud3ec\uc9c0\ud1a0\ub9ac\ub97c \ubcf5\uc81c\ud558\uace0 \ud544\uc218 \ub77c\uc774\ube0c\ub7ec\ub9ac\ub97c \uc124\uce58\ud588\uc2b5\ub2c8\ub2e4. \ub2e4\uc74c \ub2e8\uacc4\ub294 \ub370\uc774\ud130\uc14b\uc744 \uc900\ube44\ud558\ub294 \uac83\uc785\ub2c8\ub2e4. \uc804\uc774 \ud559\uc2b5\uc744 \uc704\ud574\uc11c\ub294 \ub450 \uac00\uc9c0 \ud30c\uc77c\uc774 \ud544\uc694\ud569\ub2c8\ub2e4:<\/p>\n
\n- \uc774\ubbf8\uc9c0 \ub370\uc774\ud130: \uac1d\uccb4 \uac10\uc9c0\uc5d0 \uc0ac\uc6a9\ub420 \uc774\ubbf8\uc9c0\ub4e4.<\/li>\n
- \ub77c\ubca8 \ud30c\uc77c: \uac01 \uc774\ubbf8\uc9c0\uc5d0 \ub300\ud55c bounding box \uc815\ubcf4\uac00 \uc788\ub294 \ud14d\uc2a4\ud2b8 \ud30c\uc77c.<\/li>\n<\/ul>\n
\ub2e4\uc74c\uc73c\ub85c, \ub370\uc774\ud130\uc14b\uc744 \uc900\ube44\ud558\ub294 \uacfc\uc815\uc744 \ubcf4\uc5ec\ub4dc\ub9ac\uaca0\uc2b5\ub2c8\ub2e4. \uac01 \ud074\ub798\uc2a4\ubcc4\ub85c \ud3f4\ub354\ub97c \ub2ec\ub9ac\ud558\uc5ec \uc774\ubbf8\uc9c0\ub97c \uc815\ub9ac\ud558\uace0, \ud574\ub2f9 \ud3f4\ub354 \ub0b4\uc5d0 `labels` \ud3f4\ub354\ub97c \ub9cc\ub4e4\uc5b4 \uadf8\uc5d0 \ub300\ud55c \ud14d\uc2a4\ud2b8 \ub77c\ubca8 \ud30c\uc77c\uc744 \ub9cc\ub4e4\uc5b4\uc57c \ud569\ub2c8\ub2e4.<\/p>\n
import os\n\n# \ub370\uc774\ud130\uc14b \ud3f4\ub354 \uad6c\uc870 \uc124\uc815\ndataset_path = 'path\/to\/your\/dataset'\nos.makedirs(os.path.join(dataset_path, 'images\/train'), exist_ok=True)\nos.makedirs(os.path.join(dataset_path, 'images\/val'), exist_ok=True)\nos.makedirs(os.path.join(dataset_path, 'labels\/train'), exist_ok=True)\nos.makedirs(os.path.join(dataset_path, 'labels\/val'), exist_ok=True)\n<\/code><\/pre>\n\ubaa8\ub4e0 \ub370\uc774\ud130\uac00 \uc900\ube44\ub418\uc5c8\ub2e4\uba74, YOLOv5\uc5d0\uc11c \uc774 \ub370\uc774\ud130\uc14b\uc744 \uc0ac\uc6a9\ud560 \uc218 \uc788\ub3c4\ub85d \uad6c\uc131 \ud30c\uc77c\uc744 \uc791\uc131\ud574\uc57c \ud569\ub2c8\ub2e4.<\/p>\n
data_config = \"\"\"\ntrain: ..\/dataset\/images\/train\nval: ..\/dataset\/images\/val\n\nnc: 2\nnames: ['class1', 'class2'] # \uc0ac\uc6a9\ud560 \ud074\ub798\uc2a4 \uc774\ub984\n\"\"\"\n\nwith open('data.yaml', 'w') as f:\n f.write(data_config)\n<\/code><\/pre>\nGPU \ud658\uacbd\uc5d0\uc11c YOLO \ubaa8\ub378 \ud6c8\ub828<\/h2>\n
\uc774\uc81c \ub370\uc774\ud130\uc14b\uacfc \uad6c\uc131 \ud30c\uc77c\uc774 \uc900\ube44\ub410\uc73c\ub2c8, YOLOv5 \ubaa8\ub378\uc744 \ud6c8\ub828\ud560 \uc900\ube44\uac00 \ub418\uc5c8\uc2b5\ub2c8\ub2e4. GPU\ub97c \ud65c\uc6a9\ud560 \uc218 \uc788\ub2e4\uba74 \ud6c8\ub828 \uc18d\ub3c4\ub97c \ud06c\uac8c \uac1c\uc120\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4. \ud6c8\ub828\uc740 \ub2e4\uc74c\uacfc \uac19\uc774 \uc9c4\ud589\ud569\ub2c8\ub2e4.<\/p>\n
!python train.py --img 640 --batch 16 --epochs 50 --data data.yaml --weights yolov5s.pt\n<\/code><\/pre>\n\uc5ec\uae30\uc11c --img<\/code>\ub294 \uc774\ubbf8\uc9c0\uc758 \ud06c\uae30, --batch<\/code>\ub294 \ubc30\uce58 \ud06c\uae30, --epochs<\/code>\ub294 \ud6c8\ub828 \uc5d0\ud3ec\ud06c \uc218\ub97c \ub098\ud0c0\ub0c5\ub2c8\ub2e4. --weights<\/code>\ub294 \ubbf8\ub9ac \ud559\uc2b5\ub41c YOLOv5s \ubaa8\ub378\uc758 \uac00\uc911\uce58\ub97c \uc0ac\uc6a9\ud558\uaca0\ub2e4\ub294 \uc758\ubbf8\uc785\ub2c8\ub2e4. \uc774 \uacfc\uc815\uc744 \ud1b5\ud574 \ubaa8\ub378\uc774 \uc0c8\ub85c\uc6b4 \ub370\uc774\ud130\uc14b\uc5d0 \ub300\ud574 \ud559\uc2b5\uc744 \uc9c4\ud589\ud569\ub2c8\ub2e4.<\/p>\n\ub370\uc774\ud130 \uc99d\uac15(Data Augmentation) \uae30\ubc95<\/h2>\n
\ub370\uc774\ud130 \uc99d\uac15\uc740 \ubaa8\ub378\uc758 \uc77c\ubc18\ud654 \ub2a5\ub825\uc744 \ud5a5\uc0c1\uc2dc\ud0a4\uae30 \uc704\ud574 \uc6d0\ubcf8 \ub370\uc774\ud130\uc5d0 \ub2e4\uc591\ud55c \ubcc0\ud615\uc744 \uc8fc\uc5b4 \uc0c8\ub85c\uc6b4 \ub370\uc774\ud130\ub97c \ub9cc\ub4e4\uc5b4 \ub0b4\ub294 \uae30\ubc95\uc785\ub2c8\ub2e4. \ub370\uc774\ud130 \uc99d\uac15 \uae30\ubc95\uc73c\ub85c\ub294 \ud68c\uc804, \uc0c9\uc0c1 \ubcc0\ud654, \ud06c\uae30 \uc870\uc815, \uc790\ub974\uae30 \ub4f1\uc774 \uc788\uc2b5\ub2c8\ub2e4. YOLOv5\ub294 \uc774\ubbf8 \uc774\ub7ec\ud55c \uc99d\uac15 \uae30\ubc95\uc744 \uae30\ubcf8\uc801\uc73c\ub85c \ud3ec\ud568\ud558\uace0 \uc788\uc9c0\ub9cc, \ubcf4\ub2e4 \ub098\uc740 \uc131\ub2a5\uc744 \uc704\ud574 \ucd94\uac00\uc801\uc778 \uae30\ubc95\uc744 \uc801\uc6a9\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n
\ud30c\uc774\uc36c\uc73c\ub85c \uc9c1\uc811 \ub370\uc774\ud130 \uc99d\uac15\ud558\uae30<\/h3>\n
\ub2e4\uc74c\uc740 OpenCV\uc640 albumentations \ub77c\uc774\ube0c\ub7ec\ub9ac\ub97c \uc0ac\uc6a9\ud558\uc5ec \ub370\uc774\ud130\ub97c \uc99d\uac15\ud558\ub294 \uc608\uc81c\uc785\ub2c8\ub2e4.<\/p>\n
import cv2\nimport albumentations as A\n\ndef augment_image(image):\n transform = A.Compose([\n A.RandomCrop(width=450, height=450),\n A.HorizontalFlip(p=0.5),\n A.RandomBrightnessContrast(p=0.2),\n A.Rotate(limit=40, p=0.5)\n ])\n \n augmented = transform(image=image)\n return augmented['image']\n\n# \uc608\uc2dc \uc774\ubbf8\uc9c0 \ubd88\ub7ec\uc624\uae30 \ubc0f \uc99d\uac15\nimage = cv2.imread('path\/to\/image.jpg')\naugmented_image = augment_image(image)\n\ncv2.imwrite('path\/to\/augmented_image.jpg', augmented_image)\n<\/code><\/pre>\n\uc704 \ucf54\ub4dc\ub294 \uc8fc\uc5b4\uc9c4 \uc774\ubbf8\uc9c0\ub97c \ub2e4\uc591\ud55c \ubc29\ubc95\uc73c\ub85c \uc99d\uac15\ud558\uc5ec \uc0c8\ub85c\uc6b4 \uc774\ubbf8\uc9c0\ub97c \uc0dd\uc131\ud569\ub2c8\ub2e4. \uc774\ub7f0 \uc2dd\uc73c\ub85c \ub370\uc774\ud130\ub97c \uc99d\uac15\ud558\uba74 \ubaa8\ub378\uc774 \ub2e4\uc591\ud55c \uc0c1\ud669\uc5d0\uc11c \ub354 \uc798 \uc77c\ubc18\ud654\ub420 \uc218 \uc788\ub3c4\ub85d \ub3c4\uc640\uc90d\ub2c8\ub2e4.<\/p>\n
\uacb0\uacfc \ud3c9\uac00 \ubc0f \uc131\ub2a5 \ud5a5\uc0c1 \ud655\uc778<\/h2>\n
\ubaa8\ub378 \ud6c8\ub828\uc774 \uc644\ub8cc\ub418\uba74, \uc131\ub2a5\uc744 \ud3c9\uac00\ud558\uc5ec \uc804\uc774 \ud559\uc2b5 \ubc0f \ub370\uc774\ud130 \uc99d\uac15\uc758 \uc601\ud5a5\uc744 \ud655\uc778\ud560 \uc218 \uc788\ub294 \uacfc\uc815\uc774 \ud544\uc694\ud569\ub2c8\ub2e4. \uc77c\ubc18\uc801\uc73c\ub85c mAP(mean Average Precision) \uc9c0\ud45c\ub97c \uc0ac\uc6a9\ud558\uc5ec \uc131\ub2a5\uc744 \uce21\uc815\ud569\ub2c8\ub2e4. \uc544\ub798 \ucf54\ub4dc\ub294 YOLOv5\uc5d0\uc11c \uc81c\uacf5\ud558\ub294 \uac80\uc99d \uae30\ub2a5\uc744 \uc0ac\uc6a9\ud558\uc5ec \ubaa8\ub378 \uc131\ub2a5\uc744 \ud3c9\uac00\ud558\ub294 \ubc29\ubc95\uc785\ub2c8\ub2e4.<\/p>\n
!python val.py --weights runs\/train\/exp\/weights\/best.pt --data data.yaml --img 640\n<\/code><\/pre>\n\ud6c8\ub828\ub41c \ubaa8\ub378\uc744 \uc0ac\uc6a9\ud558\uc5ec \uac80\uc99d\uc744 \uc218\ud589\ud558\uace0, mAP \uac12\uc744 \uae30\uc900\uc73c\ub85c \uc774\uc804\uc758 \ubaa8\ub378\uacfc \uc131\ub2a5 \ud5a5\uc0c1\uc744 \ube44\uad50\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n
\uacb0\ub860<\/h3>\n
\uc774\ubc88 \uac15\uc88c\uc5d0\uc11c\ub294 YOLO \ubaa8\ub378\uc758 \uc804\uc774 \ud559\uc2b5\uacfc \ub370\uc774\ud130 \uc99d\uac15 \uae30\ubc95\uc744 \ud1b5\ud574 \uc131\ub2a5\uc744 \ud5a5\uc0c1\uc2dc\ud0ac \uc218 \uc788\ub294 \ubc29\ubc95\uc744 \uc54c\uc544\ubcf4\uc558\uc2b5\ub2c8\ub2e4. YOLOv5\ub97c \ud1b5\ud55c \uc804\uc774 \ud559\uc2b5\uc740 \uc801\uc740 \uc591\uc758 \ub370\uc774\ud130\ub85c\ub3c4 \ud6a8\uacfc\uc801\uc778 \ubaa8\ub378\uc744 \ud559\uc2b5\ud558\uac8c \ud574\uc8fc\uba70, \ub370\uc774\ud130 \uc99d\uac15 \uae30\ubc95\uc740 \ubaa8\ub378\uc758 \uc77c\ubc18\ud654 \ub2a5\ub825\uc744 \uc99d\uac00\uc2dc\ucf1c \ubcf4\ub2e4 \uacac\uace0\ud55c \ubaa8\ub378\uc744 \ub9cc\ub4e4 \uc218 \uc788\uac8c \ub3c4\uc640\uc90d\ub2c8\ub2e4.<\/p>\n
\uc774 \ub450 \uac00\uc9c0 \uae30\ubc95\uc744 \uc801\uc808\ud788 \ud65c\uc6a9\ud558\uba74 \ub2e4\uc591\ud55c \ucef4\ud4e8\ud130 \ube44\uc804 \ubb38\uc81c\ub97c \ud574\uacb0\ud558\ub294 \ub370 \ud070 \ub3c4\uc6c0\uc774 \ub420 \uac83\uc785\ub2c8\ub2e4. \uc55e\uc73c\ub85c\ub3c4 YOLO\ub97c \ud65c\uc6a9\ud55c \ub2e4\uc591\ud55c \uc0ac\ub840\uc640 \uae30\ubc95\uc5d0 \ub300\ud574 \uc9c0\uc18d\uc801\uc73c\ub85c \ud0d0\uad6c\ud574 \ubcf4\uc2dc\uae30 \ubc14\ub78d\ub2c8\ub2e4.<\/p>\n
<\/body><\/p>\n","protected":false},"excerpt":{"rendered":"