{"id":29572,"date":"2024-10-28T01:55:31","date_gmt":"2024-10-28T01:55:31","guid":{"rendered":"http:\/\/atmokpo.com\/w\/?p=29572"},"modified":"2024-11-26T06:52:13","modified_gmt":"2024-11-26T06:52:13","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-%ec%b6%94%eb%a1%a0","status":"publish","type":"post","link":"https:\/\/atmokpo.com\/w\/29572\/","title":{"rendered":"\ud5c8\uae45\ud398\uc774\uc2a4 \ud2b8\ub80c\uc2a4\ud3ec\uba38 \ud65c\uc6a9\uac15\uc88c, CLIP \ucd94\ub860"},"content":{"rendered":"

<\/p>\n

\ucd5c\uadfc \uc778\uacf5\uc9c0\ub2a5 \ubd84\uc57c\uc5d0\uc11c CLIP (Contrastive Language-Image Pretraining) \ubaa8\ub378\uc774 \uc8fc\ubaa9\ubc1b\uace0 \uc788\uc2b5\ub2c8\ub2e4. CLIP \ubaa8\ub378\uc740 \uc790\uc5f0\uc5b4\uc640 \uc774\ubbf8\uc9c0 \uac04\uc758 \uad00\uacc4\ub97c \ud559\uc2b5\ud558\uc5ec \ub2e4\uc591\ud55c \uc751\uc6a9 \ud504\ub85c\uadf8\ub7a8\uc5d0\uc11c \ud65c\uc6a9\ub420 \uc218 \uc788\ub3c4\ub85d \ud569\ub2c8\ub2e4. \ubcf8 \uac15\uc88c\uc5d0\uc11c\ub294 CLIP \ubaa8\ub378\uc744 \uc0ac\uc6a9\ud558\uc5ec \uc774\ubbf8\uc9c0\uc640 \ud14d\uc2a4\ud2b8 \uac04\uc758 \uc720\uc0ac\ub3c4\ub97c \ucd94\ub860\ud558\ub294 \ubc29\ubc95\uc5d0 \ub300\ud574 \uc790\uc138\ud788 \uc0b4\ud3b4\ubcf4\uaca0\uc2b5\ub2c8\ub2e4.<\/p>\n

CLIP \ubaa8\ub378\uc774\ub780?<\/h2>\n

CLIP \ubaa8\ub378\uc740 OpenAI\uc5d0\uc11c \uac1c\ubc1c\ud55c \ubaa8\ub378\ub85c, \ub300\ub7c9\uc758 \uc6f9\uc5d0\uc11c \uc218\uc9d1\ub41c \ud14d\uc2a4\ud2b8-\uc774\ubbf8\uc9c0 \uc30d \ub370\uc774\ud130\ub97c \ud1b5\ud574 \ud559\uc2b5\ub418\uc5c8\uc2b5\ub2c8\ub2e4. \uc774 \ubaa8\ub378\uc740 \uc774\ubbf8\uc9c0\uc640 \ud14d\uc2a4\ud2b8\ub97c \uac01\uac01\uc758 \uc784\ubca0\ub529 \uacf5\uac04\uc73c\ub85c \ub9e4\ud551\ud558\uace0, \uc774 \ub450 \uc784\ubca0\ub529 \uac04\uc758 \uac70\ub9ac(\uc720\uc0ac\ub3c4)\ub97c \uacc4\uc0b0\ud558\uc5ec \ud2b9\uc815 \uc774\ubbf8\uc9c0\ub97c \uc124\uba85\ud558\ub294 \ud14d\uc2a4\ud2b8\ub97c \ucc3e\uac70\ub098, \ubc18\ub300\ub85c \uc774\ubbf8\uc9c0\ub97c \uae30\ubc18\uc73c\ub85c \uac00\uc7a5 \uad00\ub828\uc774 \uae4a\uc740 \ud14d\uc2a4\ud2b8\ub97c \ucc3e\ub294 \ub370 \uc0ac\uc6a9\ub429\ub2c8\ub2e4.<\/p>\n

CLIP\uc758 \uc791\ub3d9 \uc6d0\ub9ac<\/h2>\n

CLIP\ub294 \ub2e4\uc74c\uacfc \uac19\uc740 \ub450 \uac00\uc9c0 \uc8fc\uc694 \uad6c\uc131 \uc694\uc18c\ub85c \uc774\ub8e8\uc5b4\uc838 \uc788\uc2b5\ub2c8\ub2e4:<\/p>\n

    \n
  1. \uc774\ubbf8\uc9c0 \uc778\ucf54\ub354:<\/strong> \uc8fc\uc5b4\uc9c4 \uc774\ubbf8\uc9c0\ub97c \uc785\ub825\ubc1b\uc544 \uc774\ubbf8\uc9c0\ub97c \ud45c\ud604\ud558\ub294 \ubca1\ud130\ub85c \ubcc0\ud658\ud569\ub2c8\ub2e4.<\/li>\n
  2. \ud14d\uc2a4\ud2b8 \uc778\ucf54\ub354:<\/strong> \uc785\ub825\ub41c \ud14d\uc2a4\ud2b8\ub97c \ubc1b\uc544\ub4e4\uc774\uba70, \ud14d\uc2a4\ud2b8\ub97c \ud45c\ud604\ud558\ub294 \ubca1\ud130\ub97c \uc0dd\uc131\ud569\ub2c8\ub2e4.<\/li>\n<\/ol>\n

    \uc774 \ub450 \uc778\ucf54\ub354\ub294 \uac01\uae30 \ub2e4\ub978 \ubc29\uc2dd\uc73c\ub85c \uc791\ub3d9\ud558\uc9c0\ub9cc, \ucd5c\uc885\uc801\uc73c\ub85c \ub3d9\uc77c\ud55c \ucc28\uc6d0\uc758 \ubca1\ud130 \uacf5\uac04\uc5d0 \ub9e4\ud551\ub429\ub2c8\ub2e4. \uadf8 \ud6c4, \ub450 \ubca1\ud130 \uac04\uc758 \ucf54\uc0ac\uc778 \uc720\uc0ac\uc131\uc744 \ud1b5\ud574 \uc774\ubbf8\uc9c0\uc640 \ud14d\uc2a4\ud2b8\uc758 \uad00\ub828\uc131\uc744 \ud3c9\uac00\ud569\ub2c8\ub2e4.<\/p>\n

    CLIP \ubaa8\ub378 \ud65c\uc6a9 \uc608\uc81c<\/h2>\n

    \uc774\uc81c CLIP \ubaa8\ub378\uc744 \uc0ac\uc6a9\ud574\ubcf4\uaca0\uc2b5\ub2c8\ub2e4. \uc544\ub798\ub294 Hugging Face\uc758 Transformers \ub77c\uc774\ube0c\ub7ec\ub9ac\ub97c \ud65c\uc6a9\ud558\uc5ec CLIP \ubaa8\ub378\uc744 \ub2e4\uc6b4\ub85c\ub4dc\ud558\uace0 \uc0ac\uc6a9\ud558\ub294 \ubc29\ubc95\uc5d0 \ub300\ud55c \uc608\uc81c \ucf54\ub4dc\uc785\ub2c8\ub2e4.<\/p>\n

    import torch\nfrom transformers import CLIPProcessor, CLIPModel\nimport PIL\n\n# CLIP \ubaa8\ub378\uacfc \ud504\ub85c\uc138\uc11c \ub85c\ub4dc\nmodel = CLIPModel.from_pretrained(\"openai\/clip-vit-base-patch16\")\nprocessor = CLIPProcessor.from_pretrained(\"openai\/clip-vit-base-patch16\")\n\n# \uc774\ubbf8\uc9c0\uc640 \ud14d\uc2a4\ud2b8 \uc815\uc758\nimage_path = 'path_to_your_image.jpg'  # \ucc98\ub9ac\ud560 \uc774\ubbf8\uc9c0\uc758 \uacbd\ub85c\ntext = \"A description of the image\"  # \uc774\ubbf8\uc9c0\uc5d0 \ub300\ud55c \uc124\uba85\n\n# \uc774\ubbf8\uc9c0 \uc5f4\uae30\nimage = PIL.Image.open(image_path)\n\n# \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# \ub85c\uc9d3\uacfc \uc720\uc0ac\ub3c4 \uacc4\uc0b0\nlogits_per_image = outputs.logits_per_image  # \uc774\ubbf8\uc9c0\ub97c \uc124\uba85\ud558\ub294 \ud14d\uc2a4\ud2b8\uc758 \ub85c\uc9d3\nprobs = logits_per_image.softmax(dim=1)      # Softmax\ub97c \ud1b5\ud574 \ud655\ub960\ub85c \ubcc0\ud658\n\nprint(f\"The probability that the image matches the text '{text}' is: {probs[0][0].item()}\")\n<\/code><\/pre>\n
    \ucf54\ub4dc \uc124\uba85<\/h3>\n