{
  "id": 100601,
  "title": "Comparison of different image prepossessing ",
  "url": "/competitions/aptos2019-blindness-detection/discussion/100601",
  "author_name": "",
  "post_date": "2019-07-19T13:56:38.625887700Z",
  "votes": null,
  "comment_count": 1,
  "views": 0,
  "content": "<p>I have tried different image prepossessing methods and it seems the results are variant, I have noticed that some public kernels using different image prepossessing as well. I will keep updating the performance here. \nHere  show the\n<a href=\"https://www.kaggle.com/c/aptos2019-blindness-detection/discussion/100280#latest-579501\">\nComparison of different models</a></p>",
  "messages": [
    {
      "id": "579980",
      "postDate": "07/19/2019 13:56:38",
      "content": "<p>I have tried different image prepossessing methods and it seems the results are variant, I have noticed that some public kernels using different image prepossessing as well. I will keep updating the performance here. \nHere  show the\n<a href=\"https://www.kaggle.com/c/aptos2019-blindness-detection/discussion/100280#latest-579501\">\nComparison of different models</a></p>",
      "rawMarkdown": "I have tried different image prepossessing methods and it seems the results are variant, I have noticed that some public kernels using different image prepossessing as well. I will keep updating the performance here. \nHere  show the\n[\nComparison of different models](https://www.kaggle.com/c/aptos2019-blindness-detection/discussion/100280#latest-579501)",
      "votes": null
    },
    {
      "id": "579985",
      "postDate": "07/19/2019 14:01:23",
      "content": "<ol>\n<li>Pytorch LB 0.757: <code>image = cv2.addWeighted ( image,4, cv2.GaussianBlur( image , (0,0) , 30) ,-4 ,128)</code> <code>transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])])</code> </li>\n<li>Keras  LB 0.706: <code>return ImageDataGenerator(\n    fill_mode='constant',\n    cval=0.,  # value used for fill_mode = \"constant\"\n    horizontal_flip=True, \n    vertical_flip=True, \n    rotation_range=15,\n)</code></li>\n</ol>\n\n<p>For the second one, if I use the rescale, the kappa score is always 0  and I don't know why. </p>",
      "rawMarkdown": "1. Pytorch LB 0.757: `image = cv2.addWeighted ( image,4, cv2.GaussianBlur( image , (0,0) , 30) ,-4 ,128)` `transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])])` \n2. Keras  LB 0.706: `    return ImageDataGenerator(\n        fill_mode='constant',\n        cval=0.,  # value used for fill_mode = \"constant\"\n        horizontal_flip=True, \n        vertical_flip=True, \n        rotation_range=15,\n    )`\n\nFor the second one, if I use the rescale, the kappa score is always 0  and I don't know why.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 579985,
      "author_name": "runninglion",
      "author_url": "",
      "post_date": "07/19/2019 14:01:23",
      "content": "<ol>\n<li>Pytorch LB 0.757: <code>image = cv2.addWeighted ( image,4, cv2.GaussianBlur( image , (0,0) , 30) ,-4 ,128)</code> <code>transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])])</code> </li>\n<li>Keras  LB 0.706: <code>return ImageDataGenerator(\n    fill_mode='constant',\n    cval=0.,  # value used for fill_mode = \"constant\"\n    horizontal_flip=True, \n    vertical_flip=True, \n    rotation_range=15,\n)</code></li>\n</ol>\n\n<p>For the second one, if I use the rescale, the kappa score is always 0  and I don't know why. </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "579980": "I have tried different image prepossessing methods and it seems the results are variant, I have noticed that some public kernels using different image prepossessing as well. I will keep updating the performance here. \nHere  show the\n[\nComparison of different models](https://www.kaggle.com/c/aptos2019-blindness-detection/discussion/100280#latest-579501)",
    "579985": "1. Pytorch LB 0.757: `image = cv2.addWeighted ( image,4, cv2.GaussianBlur( image , (0,0) , 30) ,-4 ,128)` `transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])])` \n2. Keras  LB 0.706: `    return ImageDataGenerator(\n        fill_mode='constant',\n        cval=0.,  # value used for fill_mode = \"constant\"\n        horizontal_flip=True, \n        vertical_flip=True, \n        rotation_range=15,\n    )`\n\nFor the second one, if I use the rescale, the kappa score is always 0  and I don't know why."
  },
  "source": "meta"
}