{
  "id": 250178,
  "title": "Notebook for standardizing imagesets",
  "url": "/competitions/siim-covid19-detection/discussion/250178",
  "author_name": "David Roberts",
  "post_date": "2021-07-01T13:03:45.914000",
  "votes": 1,
  "comment_count": 0,
  "views": 0,
  "content": "<p>It occurred to me that the variances in image bit depths, auto-leveling and post processing filters make the image <em>colors</em> vary wildly in this dataset. </p>\n<p>I figured a little standardization might go a long way.</p>\n<p>So, I made a notebook that applies basic cropping, histogram equalization and unsharp masking to the entire dataset .. which does results in a slight increase in my crappy NN model's accuracy :)</p>\n<p><a href=\"https://www.kaggle.com/davidbroberts/standardizing-cxr-datasets\" target=\"_blank\">https://www.kaggle.com/davidbroberts/standardizing-cxr-datasets</a></p>\n<p>(<em>I only applied it to a few images for demonstration purposes in this notebook.</em>)</p>",
  "messages": [
    {
      "id": 1372135,
      "postDate": "2021-07-01T13:03:45.913Z",
      "content": "<p>It occurred to me that the variances in image bit depths, auto-leveling and post processing filters make the image <em>colors</em> vary wildly in this dataset. </p>\n<p>I figured a little standardization might go a long way.</p>\n<p>So, I made a notebook that applies basic cropping, histogram equalization and unsharp masking to the entire dataset .. which does results in a slight increase in my crappy NN model's accuracy :)</p>\n<p><a href=\"https://www.kaggle.com/davidbroberts/standardizing-cxr-datasets\" target=\"_blank\">https://www.kaggle.com/davidbroberts/standardizing-cxr-datasets</a></p>\n<p>(<em>I only applied it to a few images for demonstration purposes in this notebook.</em>)</p>",
      "rawMarkdown": "It occurred to me that the variances in image bit depths, auto-leveling and post processing filters make the image *colors* vary wildly in this dataset. \n\nI figured a little standardization might go a long way.\n\nSo, I made a notebook that applies basic cropping, histogram equalization and unsharp masking to the entire dataset .. which does results in a slight increase in my crappy NN model's accuracy :)\n\nhttps://www.kaggle.com/davidbroberts/standardizing-cxr-datasets\n\n(*I only applied it to a few images for demonstration purposes in this notebook.*)",
      "votes": 1
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "1372135": "It occurred to me that the variances in image bit depths, auto-leveling and post processing filters make the image *colors* vary wildly in this dataset. \n\nI figured a little standardization might go a long way.\n\nSo, I made a notebook that applies basic cropping, histogram equalization and unsharp masking to the entire dataset .. which does results in a slight increase in my crappy NN model's accuracy :)\n\nhttps://www.kaggle.com/davidbroberts/standardizing-cxr-datasets\n\n(*I only applied it to a few images for demonstration purposes in this notebook.*)"
  }
}