{
  "id": 240441,
  "title": "Trying to understand the competition better",
  "url": "/competitions/siim-covid19-detection/discussion/240441",
  "author_name": "",
  "post_date": "2021-05-19T20:11:24.159081900Z",
  "votes": 15,
  "comment_count": 2,
  "views": 0,
  "content": "<p>For me, the \"exact\" task in this competition wasn't very obvious. So took an age-old approach to understand it better - pen, paper, and lofi. </p>\n<p>This is what I understood. It might be obvious for many but for those still struggling to figure out what exactly is this competition about here's my two cents. </p>\n<pre><code>This competition is unique because of how the data is presented and thus the problem statement. The DICOM formatted radiograms of chest scans are available in this directory structure `study/series/images`. What's `study` and `image`?\n\nThere are 6334 unique chest scans or `images` while 6054 unique `study` \"directories\". This means that in some study directory there are more than one images. \n\nIn this competition, the task is to provide image level predictions - `none` vs `opacity` as well as study level predictions - `negative`, `typical`, `atypical`, `indeterminate`. Thus we will build two separate classifiers - one for study level prediction and another for image level prediction. Then when bounding boxes or localization?\n\nFor images with `opacity` label i.e, image-level label, we need to train an object detector for localization. \n</code></pre>",
  "messages": [
    {
      "id": "1315468",
      "postDate": "05/19/2021 20:11:24",
      "content": "<p>For me, the \"exact\" task in this competition wasn't very obvious. So took an age-old approach to understand it better - pen, paper, and lofi. </p>\n<p>This is what I understood. It might be obvious for many but for those still struggling to figure out what exactly is this competition about here's my two cents. </p>\n<pre><code>This competition is unique because of how the data is presented and thus the problem statement. The DICOM formatted radiograms of chest scans are available in this directory structure `study/series/images`. What's `study` and `image`?\n\nThere are 6334 unique chest scans or `images` while 6054 unique `study` \"directories\". This means that in some study directory there are more than one images. \n\nIn this competition, the task is to provide image level predictions - `none` vs `opacity` as well as study level predictions - `negative`, `typical`, `atypical`, `indeterminate`. Thus we will build two separate classifiers - one for study level prediction and another for image level prediction. Then when bounding boxes or localization?\n\nFor images with `opacity` label i.e, image-level label, we need to train an object detector for localization. \n</code></pre>",
      "rawMarkdown": "For me, the \"exact\" task in this competition wasn't very obvious. So took an age-old approach to understand it better - pen, paper, and lofi. \n\nThis is what I understood. It might be obvious for many but for those still struggling to figure out what exactly is this competition about here's my two cents. \n\n```\nThis competition is unique because of how the data is presented and thus the problem statement. The DICOM formatted radiograms of chest scans are available in this directory structure `study/series/images`. What's `study` and `image`?\n \nThere are 6334 unique chest scans or `images` while 6054 unique `study` \"directories\". This means that in some study directory there are more than one images. \n\nIn this competition, the task is to provide image level predictions - `none` vs `opacity` as well as study level predictions - `negative`, `typical`, `atypical`, `indeterminate`. Thus we will build two separate classifiers - one for study level prediction and another for image level prediction. Then when bounding boxes or localization?\n\nFor images with `opacity` label i.e, image-level label, we need to train an object detector for localization. \n```",
      "votes": null
    },
    {
      "id": "1316192",
      "postDate": "05/20/2021 11:10:41",
      "content": "<p>Hey <a href=\"https://www.kaggle.com/ayuraj\" target=\"_blank\">@ayuraj</a>, I explained this in more detail in this <a href=\"https://www.kaggle.com/c/siim-covid19-detection/discussion/240329#1314661\" target=\"_blank\">thread</a>. Let me know if it makes sense to you.</p>",
      "rawMarkdown": "Hey @ayuraj, I explained this in more detail in this [thread](https://www.kaggle.com/c/siim-covid19-detection/discussion/240329#1314661). Let me know if it makes sense to you.",
      "votes": null
    },
    {
      "id": "1317414",
      "postDate": "05/21/2021 11:52:00",
      "content": "<p>Yups it makes sense. Thanks for sharing. I might have missed your post and went ahead creating a new one. :P</p>",
      "rawMarkdown": "Yups it makes sense. Thanks for sharing. I might have missed your post and went ahead creating a new one. :P",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1316192,
      "author_name": "hassiahk",
      "author_url": "",
      "post_date": "05/20/2021 11:10:41",
      "content": "<p>Hey <a href=\"https://www.kaggle.com/ayuraj\" target=\"_blank\">@ayuraj</a>, I explained this in more detail in this <a href=\"https://www.kaggle.com/c/siim-covid19-detection/discussion/240329#1314661\" target=\"_blank\">thread</a>. Let me know if it makes sense to you.</p>",
      "votes": null,
      "replies": [
        {
          "id": 1317414,
          "author_name": "ayuraj",
          "author_url": "",
          "post_date": "05/21/2021 11:52:00",
          "content": "<p>Yups it makes sense. Thanks for sharing. I might have missed your post and went ahead creating a new one. :P</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1315468": "For me, the \"exact\" task in this competition wasn't very obvious. So took an age-old approach to understand it better - pen, paper, and lofi. \n\nThis is what I understood. It might be obvious for many but for those still struggling to figure out what exactly is this competition about here's my two cents. \n\n```\nThis competition is unique because of how the data is presented and thus the problem statement. The DICOM formatted radiograms of chest scans are available in this directory structure `study/series/images`. What's `study` and `image`?\n \nThere are 6334 unique chest scans or `images` while 6054 unique `study` \"directories\". This means that in some study directory there are more than one images. \n\nIn this competition, the task is to provide image level predictions - `none` vs `opacity` as well as study level predictions - `negative`, `typical`, `atypical`, `indeterminate`. Thus we will build two separate classifiers - one for study level prediction and another for image level prediction. Then when bounding boxes or localization?\n\nFor images with `opacity` label i.e, image-level label, we need to train an object detector for localization. \n```",
    "1316192": "Hey @ayuraj, I explained this in more detail in this [thread](https://www.kaggle.com/c/siim-covid19-detection/discussion/240329#1314661). Let me know if it makes sense to you.",
    "1317414": "Yups it makes sense. Thanks for sharing. I might have missed your post and went ahead creating a new one. :P"
  },
  "source": "meta"
}