{"id":29564,"date":"2024-10-28T01:55:28","date_gmt":"2024-10-28T01:55:28","guid":{"rendered":"http:\/\/atmokpo.com\/w\/?p=29564"},"modified":"2024-11-26T06:52:15","modified_gmt":"2024-11-26T06:52:15","slug":"%ed%97%88%ea%b9%85%ed%8e%98%ec%9d%b4%ec%8a%a4-%ed%8a%b8%eb%a0%8c%ec%8a%a4%ed%8f%ac%eb%a8%b8-%ed%99%9c%ec%9a%a9%ea%b0%95%ec%a2%8c-clip-%ea%b8%b0%eb%b0%98-%ec%82%ac%ec%a0%84%ed%95%99%ec%8a%b5-%eb%aa%a8","status":"publish","type":"post","link":"https:\/\/atmokpo.com\/w\/29564\/","title":{"rendered":"\ud5c8\uae45\ud398\uc774\uc2a4 \ud2b8\ub80c\uc2a4\ud3ec\uba38 \ud65c\uc6a9\uac15\uc88c, CLIP \uae30\ubc18 \uc0ac\uc804\ud559\uc2b5 \ubaa8\ub378 \ubd88\ub7ec\uc624\uae30"},"content":{"rendered":"

<\/p>\n

1. \uc11c\ub860<\/h2>\n

\n \ucd5c\uadfc \uc778\uacf5\uc9c0\ub2a5 \ubd84\uc57c\uc5d0\uc11c\uc758 \ubc1c\uc804\uc740 \ub208\ubd80\uc2e0 \uc18d\ub3c4\ub85c \uc9c4\ud589\ub418\uace0 \uc788\uc2b5\ub2c8\ub2e4. \ud2b9\ud788, \ucef4\ud4e8\ud130 \ube44\uc804\uacfc \uc790\uc5f0\uc5b4 \ucc98\ub9ac \ubd84\uc57c\uc5d0\uc11c\ub294
\n \ub525\ub7ec\ub2dd \ubaa8\ub378\uc774 \ub180\ub77c\uc6b4 \uc131\ub2a5\uc744 \ubcf4\uc5ec\uc8fc\uace0 \uc788\uc2b5\ub2c8\ub2e4. \uadf8 \uc911\uc5d0\uc11c\ub3c4 CLIP(Contrastive Language\u2013Image Pre-Training) \ubaa8\ub378\uc740
\n \ud14d\uc2a4\ud2b8\uc640 \uc774\ubbf8\uc9c0\ub97c \ub3d9\uc2dc\uc5d0 \uc774\ud574\ud558\uace0 \ucc98\ub9ac\ud560 \uc218 \uc788\ub294 \uac15\ub825\ud55c \ubaa8\ub378\ub85c \uc8fc\ubaa9\ubc1b\uace0 \uc788\uc2b5\ub2c8\ub2e4. \ubcf8 \uac15\uc88c\uc5d0\uc11c\ub294 \ud5c8\uae45\ud398\uc774\uc2a4\uc758
\n Transformers \ub77c\uc774\ube0c\ub7ec\ub9ac\ub97c \ud65c\uc6a9\ud558\uc5ec CLIP \uae30\ubc18 \uc0ac\uc804\ud559\uc2b5 \ubaa8\ub378\uc744 \ubd88\ub7ec\uc624\uace0 \uc0ac\uc6a9\ud558\ub294 \ubc29\ubc95\uc5d0 \ub300\ud574 \uc0c1\uc138\ud788 \uc124\uba85\ud558\uaca0\uc2b5\ub2c8\ub2e4.\n <\/p>\n

2. CLIP \ubaa8\ub378\uc758 \uac1c\ub150<\/h2>\n

\n CLIP \ubaa8\ub378\uc740 OpenAI\uc5d0\uc11c \ubc1c\ud45c\ud55c \ubaa8\ub378\ub85c, \ud14d\uc2a4\ud2b8\uc640 \uc774\ubbf8\uc9c0\ub97c \uc5f0\uacb0\ud558\uc5ec \uc774\ud574\ud560 \uc218 \uc788\ub3c4\ub85d \ud6c8\ub828\ub418\uc5c8\uc2b5\ub2c8\ub2e4.
\n \uc774 \ubaa8\ub378\uc740 \ub300\uaddc\ubaa8 \ub370\uc774\ud130\uc14b\uc5d0\uc11c \ud14d\uc2a4\ud2b8\uc640 \uc774\ubbf8\uc9c0 \uc30d\uc744 \ud559\uc2b5\ud558\uc5ec, \uc8fc\uc5b4\uc9c4 \uc774\ubbf8\uc9c0\uc5d0 \ub300\ud55c \uc124\uba85\uc744 \uc0dd\uc131\ud558\uac70\ub098
\n \uc8fc\uc5b4\uc9c4 \ud14d\uc2a4\ud2b8\uc5d0 \ub9de\ub294 \uc774\ubbf8\uc9c0\ub97c \uc120\ud0dd\ud558\ub294 \ub4f1\uc758 \uc791\uc5c5\uc744 \uc218\ud589\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.\n <\/p>\n

\n CLIP\uc758 \ud575\uc2ec \uc544\uc774\ub514\uc5b4\ub294 \u201c\ub300\uc870 \ud559\uc2b5\u201d\uc785\ub2c8\ub2e4. \uc774\ub294 \ube44\uc2b7\ud55c \ub0b4\uc6a9\uc758 \ud14d\uc2a4\ud2b8\uc640 \uc774\ubbf8\uc9c0 \uc30d\uc740 \uc11c\ub85c \uac00\uae4c\uc6b4 \ubca1\ud130 \uacf5\uac04\uc5d0 \uc704\uce58\ud558\uace0,
\n \ub0b4\uc6a9\uc774 \ub2e4\ub978 \uc30d\uc740 \uba40\ub9ac \ub5a8\uc5b4\uc9c0\ub3c4\ub85d \ud559\uc2b5\ud558\ub294 \ubc29\uc2dd\uc785\ub2c8\ub2e4. \uc774\ub97c \ud1b5\ud574 CLIP\uc740 \ube44\uc9c0\ub3c4 \ud559\uc2b5\uc73c\ub85c\ub3c4 \ub9e4\uc6b0 \ub69c\ub837\ud55c
\n \uc131\ub2a5\uc744 \ubc1c\ud718\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.\n <\/p>\n

3. \ud5c8\uae45\ud398\uc774\uc2a4 Transformers \ub77c\uc774\ube0c\ub7ec\ub9ac<\/h2>\n

\n \ud5c8\uae45\ud398\uc774\uc2a4\uc758 Transformers \ub77c\uc774\ube0c\ub7ec\ub9ac\ub294 \uc790\uc5f0\uc5b4 \ucc98\ub9ac(NLP)\uc640 \uad00\ub828\ub41c \uc5ec\ub7ec \uac00\uc9c0 \ubaa8\ub378\uc744 \uac04\ud3b8\ud558\uac8c \uc0ac\uc6a9\ud560 \uc218 \uc788\ub3c4\ub85d
\n \ub3c4\uc640\uc8fc\ub294 \ub77c\uc774\ube0c\ub7ec\ub9ac\uc785\ub2c8\ub2e4. \uc774 \ub77c\uc774\ube0c\ub7ec\ub9ac\ub97c \ud1b5\ud574 \ub2e4\uc591\ud55c \uc0ac\uc804\ud559\uc2b5 \ubaa8\ub378\uc744 \ubd88\ub7ec\uc624\uace0, \ud1a0\ud070\ud654 \ubc0f \ub370\uc774\ud130 \uc804\ucc98\ub9ac
\n \ub4f1\uc758 \uc791\uc5c5\uc744 \uc190\uc27d\uac8c \uc218\ud589\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4. CLIP \ubaa8\ub378 \uc5ed\uc2dc \uc774 \ub77c\uc774\ube0c\ub7ec\ub9ac\uc5d0\uc11c \uc9c0\uc6d0\ub429\ub2c8\ub2e4.\n <\/p>\n

4. \ud658\uacbd \uc124\uc815<\/h2>\n

\n CLIP \ubaa8\ub378\uc744 \uc0ac\uc6a9\ud558\uae30 \uc704\ud574\uc11c\ub294 \uba3c\uc800 \ud544\uc694\ud55c \ub77c\uc774\ube0c\ub7ec\ub9ac\ub97c \uc124\uce58\ud574\uc57c \ud569\ub2c8\ub2e4. Python \ud658\uacbd\uc5d0\uc11c
\n \uc544\ub798\uc758 \uba85\ub839\uc5b4\ub97c \ud1b5\ud574 Transformers \ub77c\uc774\ube0c\ub7ec\ub9ac\uc640 \uad00\ub828\ub41c \ud328\ud0a4\uc9c0\ub97c \uc124\uce58\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.\n <\/p>\n

pip install transformers torch torchvision<\/code><\/pre>\n
5. CLIP \ubaa8\ub378 \ubd88\ub7ec\uc624\uae30<\/h2>\n
\n        \uc774\uc81c \ubcf8\uaca9\uc801\uc73c\ub85c CLIP \ubaa8\ub378\uc744 \ubd88\ub7ec\uc624\ub294 \ubc29\ubc95\uc5d0 \ub300\ud574 \uc124\uba85\ud558\uaca0\uc2b5\ub2c8\ub2e4. \ub77c\uc774\ube0c\ub7ec\ub9ac\uc5d0\uc11c \uc0ac\uc804\ud559\uc2b5\ub41c CLIP \ubaa8\ub378\uc744
\n        \uc27d\uac8c \uc0ac\uc6a9\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4. \uc544\ub798\uc758 Python \ucf54\ub4dc\ub97c \ud1b5\ud574 CLIP \ubaa8\ub378\uacfc \ud1a0\ud06c\ub098\uc774\uc800\ub97c \ubd88\ub7ec\uc624\ub294 \uc608\uc81c\ub97c
\n        \uc0b4\ud3b4\ubcf4\uaca0\uc2b5\ub2c8\ub2e4.\n    <\/p>\n
from transformers import CLIPProcessor, CLIPModel\n\n# CLIP \ubaa8\ub378\uacfc \ud504\ub85c\uc138\uc11c \ubd88\ub7ec\uc624\uae30\nmodel = CLIPModel.from_pretrained('openai\/clip-vit-base-patch16')\nprocessor = CLIPProcessor.from_pretrained('openai\/clip-vit-base-patch16')<\/code><\/pre>\n
5.1. \ubaa8\ub378\uacfc \ud504\ub85c\uc138\uc11c \uc124\uba85<\/h3>\n
\n        \uc704 \ucf54\ub4dc\uc5d0\uc11c `from_pretrained` \uba54\uc18c\ub4dc\ub97c \uc0ac\uc6a9\ud558\uc5ec \uc0ac\uc804\ud559\uc2b5\ub41c CLIP \ubaa8\ub378\uacfc \ud504\ub85c\uc138\uc11c\ub97c \ubd88\ub7ec\uc635\ub2c8\ub2e4. \uc5ec\uae30\uc11c
\n        \ud504\ub85c\uc138\uc11c\ub294 \uc785\ub825\ub41c \ud14d\uc2a4\ud2b8\uc640 \uc774\ubbf8\uc9c0\ub97c \ucc98\ub9ac\ud558\uc5ec \ubaa8\ub378\uc774 \uc774\ud574\ud560 \uc218 \uc788\ub294 \ud615\uc2dd\uc73c\ub85c \ubcc0\ud658\ud558\ub294 \uc5ed\ud560\uc744 \ud569\ub2c8\ub2e4.
\n        \uc989, \uc774\ubbf8\uc9c0\ub97c \ud150\uc11c \ud615\ud0dc\ub85c \ubcc0\ud658\ud558\uace0, \ud14d\uc2a4\ud2b8\ub97c \ud1a0\ud070\ud654\ud558\uc5ec \ubaa8\ub378\uc774 \uc785\ub825\uc73c\ub85c \ubc1b\uc544\ub4e4\uc77c \uc218 \uc788\ub3c4\ub85d \ud569\ub2c8\ub2e4.\n    <\/p>\n
6. \uc774\ubbf8\uc9c0\uc640 \ud14d\uc2a4\ud2b8\uc758 \uc785\ub825<\/h2>\n
\n        CLIP \ubaa8\ub378\uc740 \uc774\ubbf8\uc9c0\uc640 \ud14d\uc2a4\ud2b8 \ubaa8\ub450\ub97c \uc785\ub825\uc73c\ub85c \ubc1b\uc744 \uc218 \uc788\uc2b5\ub2c8\ub2e4. \uc544\ub798\uc758 \ucf54\ub4dc\uc5d0\uc11c\ub294 \uc784\uc758\uc758 \uc774\ubbf8\uc9c0\ub97c
\n        \ub2e4\uc6b4\ub85c\ub4dc\ud558\uace0, \ud574\ub2f9 \uc774\ubbf8\uc9c0\uc5d0 \ub300\ud55c \ud14d\uc2a4\ud2b8\uc640 \ud568\uaed8 \ubaa8\ub378\uc5d0 \uc785\ub825\ud558\ub294 \uacfc\uc815\uc744 \ubcf4\uc5ec\ub4dc\ub9bd\ub2c8\ub2e4.\n    <\/p>\n
import requests\nfrom PIL import Image\n\n# \uc774\ubbf8\uc9c0 \ub2e4\uc6b4\ub85c\ub4dc\nurl = \"https:\/\/example.com\/sample.jpg\"\nimage = Image.open(requests.get(url, stream=True).raw)\n\n# \ud14d\uc2a4\ud2b8 \uc785\ub825\ntext = \"A sample image description\"<\/code><\/pre>\n
6.1. \uc774\ubbf8\uc9c0 \ud30c\uc77c \uc900\ube44\ud558\uae30<\/h3>\n
\n        \uc704 \ucf54\ub4dc\uc5d0\uc11c\ub294 requests \ub77c\uc774\ube0c\ub7ec\ub9ac\ub97c \uc0ac\uc6a9\ud558\uc5ec \uc774\ubbf8\uc9c0 \ud30c\uc77c\uc744 \ub2e4\uc6b4\ub85c\ub4dc\ud569\ub2c8\ub2e4. \uc774\ud6c4 Pillow \ub77c\uc774\ube0c\ub7ec\ub9ac\ub97c
\n        \uc0ac\uc6a9\ud558\uc5ec \uc774\ubbf8\uc9c0\ub97c \uc5f4\uc5b4\uc90d\ub2c8\ub2e4. \uc2e4\uc81c\ub85c \uc0ac\uc6a9\ud560 \uc774\ubbf8\uc9c0\ub97c URL\ub85c \uc9c0\uc815\ud558\uc5ec \ub2e4\uc6b4\ub85c\ub4dc\ud560 \uc218 \uc788\uc73c\uba70, \ub85c\uceec\uc5d0 \uc800\uc7a5\ub41c
\n        \uc774\ubbf8\uc9c0 \ud30c\uc77c\uc744 \uc0ac\uc6a9\ud574\ub3c4 \ub429\ub2c8\ub2e4.\n    <\/p>\n
7. CLIP \ubaa8\ub378\uc758 \ucd94\ub860<\/h2>\n
\n        \uc774\uc81c \ubaa8\ub378\uc5d0 \uc774\ubbf8\uc9c0\uc640 \ud14d\uc2a4\ud2b8\ub97c \uc785\ub825\ud558\uc5ec \ucd94\ub860\uc744 \uc9c4\ud589\ud574\ubcf4\uaca0\uc2b5\ub2c8\ub2e4. \ub2e4\uc74c \ucf54\ub4dc\ub97c \ud1b5\ud574 \ubaa8\ub378\uc758
\n        \ucd9c\ub825\uc744 \ud655\uc778\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.\n    <\/p>\n
# \uc785\ub825 \ub370\uc774\ud130 \uc804\ucc98\ub9ac\ninputs = processor(text=text, images=image, return_tensors=\"pt\", padding=True)\n\n# \ubaa8\ub378 \ucd94\ub860\noutputs = model(**inputs)\n\n# \uc720\uc0ac\ub3c4 \uc810\uc218 \ucd94\ucd9c\nlogits_per_image = outputs.logits_per_image\nprobs = logits_per_image.softmax(dim=1)\n\nprint(f\"\uc608\uce21 \ud655\ub960: {probs}\")<\/code><\/pre>\n
7.1. \ubaa8\ub378 \ucd94\ub860 \uacfc\uc815 \uc124\uba85<\/h3>\n
\n        \uc785\ub825\ub41c \ud14d\uc2a4\ud2b8\uc640 \uc774\ubbf8\uc9c0\uc5d0 \ub300\ud574\uc11c `processor`\ub97c \uc0ac\uc6a9\ud558\uc5ec \uc804\ucc98\ub9ac\ub97c \uc9c4\ud589\ud55c \ud6c4, \ud574\ub2f9 \uc815\ubcf4\ub97c model\uc5d0 \ub123\uc5b4 \ucd94\ub860\ud569\ub2c8\ub2e4.
\n        \uadf8 \uacb0\uacfc\ub85c \ubc18\ud658\ub41c logits\ub97c softmax \ud568\uc218\ub97c \ud1b5\ud574 \ud655\ub960 \ubd84\ud3ec\ub85c \ubcc0\ud658\ud558\uc5ec \ucd5c\uc885\uc801\uc778 \uc608\uce21 \ud655\ub960\uc744 \ub3c4\ucd9c\ud569\ub2c8\ub2e4.\n    <\/p>\n
8. \uc608\uc81c: CLIP \ubaa8\ub378 \ud65c\uc6a9\ud558\uae30<\/h2>\n
\n        \uc544\ub798\uc758 \uc804\uccb4 \ucf54\ub4dc\ub97c \ud1b5\ud574 CLIP \ubaa8\ub378\uc744 \uc2e4\uc81c\ub85c \uc5b4\ub5bb\uac8c \ud65c\uc6a9\ud560 \uc218 \uc788\ub294\uc9c0 \ubcf4\uc5ec\uc8fc\uaca0\uc2b5\ub2c8\ub2e4.
\n        \uc774 \ucf54\ub4dc\ub294 \uc8fc\uc5b4\uc9c4 \ud14d\uc2a4\ud2b8\uc5d0 \ub300\ud55c \uc774\ubbf8\uc9c0\uc758 \uc720\uc0ac\ub3c4\ub97c \ud3c9\uac00\ud569\ub2c8\ub2e4.\n    <\/p>\n
import requests\nimport torch\nfrom PIL import Image\nfrom transformers import CLIPProcessor, CLIPModel\n\n# \ubaa8\ub378\uacfc \ud504\ub85c\uc138\uc11c \ubd88\ub7ec\uc624\uae30\nmodel = CLIPModel.from_pretrained('openai\/clip-vit-base-patch16')\nprocessor = CLIPProcessor.from_pretrained('openai\/clip-vit-base-patch16')\n\n# \uc774\ubbf8\uc9c0 \ub2e4\uc6b4\ub85c\ub4dc\nurl = \"https:\/\/example.com\/sample.jpg\"\nimage = Image.open(requests.get(url, stream=True).raw)\n\n# \ud14d\uc2a4\ud2b8 \uc785\ub825\ntext = \"A sample image description\"\n\n# \uc785\ub825 \ub370\uc774\ud130 \uc804\ucc98\ub9ac\ninputs = processor(text=text, images=image, return_tensors=\"pt\", padding=True)\n\n# \ubaa8\ub378 \ucd94\ub860\nwith torch.no_grad():\n    outputs = model(**inputs)\n\n# \uc720\uc0ac\ub3c4 \uc810\uc218 \ucd94\ucd9c\nlogits_per_image = outputs.logits_per_image\nprobs = logits_per_image.softmax(dim=1)\n\nprint(f\"\uc608\uce21 \ud655\ub960: {probs}\")<\/code><\/pre>\n
9. \uacb0\ub860<\/h2>\n
\n        \uc774\ubc88 \uac15\uc88c\uc5d0\uc11c\ub294 \ud5c8\uae45\ud398\uc774\uc2a4\uc758 Transformers \ub77c\uc774\ube0c\ub7ec\ub9ac\ub97c \ud65c\uc6a9\ud558\uc5ec CLIP \uae30\ubc18\uc758 \uc0ac\uc804\ud559\uc2b5 \ubaa8\ub378\uc744 \ubd88\ub7ec\uc624\uace0,
\n        \uc774\ubbf8\uc9c0\uc640 \ud14d\uc2a4\ud2b8 \uc30d\uc744 \uc785\ub825\ud558\uc5ec \uc720\uc0ac\ub3c4\ub97c \ud3c9\uac00\ud558\ub294 \ubc29\ubc95\uc5d0 \ub300\ud574 \uc124\uba85\ud588\uc2b5\ub2c8\ub2e4. CLIP \ubaa8\ub378\uc740 \ub2e4\uc591\ud55c \uc751\uc6a9\uc774
\n        \uac00\ub2a5\ud558\uba70, \ub354\uc6b1 \ubc1c\uc804\ub41c AI \uc2dc\uc2a4\ud15c\uc758 \uac1c\ubc1c\uc5d0 \uae30\uc5ec\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4. \uc55e\uc73c\ub85c\ub3c4 \ub2e4\uc591\ud55c \ub525\ub7ec\ub2dd \uae30\uc220\uc744
\n        \ud65c\uc6a9\ud558\uc5ec \uc778\uacf5\uc9c0\ub2a5\uc758 \uac00\ub2a5\uc131\uc744 \ub113\ud600 \ub098\uac00\uc2dc\uae30 \ubc14\ub78d\ub2c8\ub2e4.\n    <\/p>\n
10. \ucc38\uace0 \uc790\ub8cc<\/h2>\n