{
  "id": 333549,
  "title": "how to identify lung alveolus?",
  "url": "/competitions/hubmap-organ-segmentation/discussion/333549",
  "author_name": "",
  "post_date": "2022-06-27T07:19:40.346632400Z",
  "votes": 8,
  "comment_count": 7,
  "views": 0,
  "content": "<p>any good resources on how to identify lung alveolus?</p>\n<p>After reading through the web and watching youtube videos, it is difficult for me to locate lung alveolus.<br>\nanyone has any tips or advices?</p>",
  "messages": [
    {
      "id": "1834778",
      "postDate": "06/27/2022 07:19:40",
      "content": "<p>any good resources on how to identify lung alveolus?</p>\n<p>After reading through the web and watching youtube videos, it is difficult for me to locate lung alveolus.<br>\nanyone has any tips or advices?</p>",
      "rawMarkdown": "any good resources on how to identify lung alveolus?\n\nAfter reading through the web and watching youtube videos, it is difficult for me to locate lung alveolus.\nanyone has any tips or advices?",
      "votes": null
    },
    {
      "id": "1834800",
      "postDate": "06/27/2022 07:48:50",
      "content": "<p>The Human Protein Atlas Dictionary may be a good resource for this<br>\n<a href=\"https://www.proteinatlas.org/learn/dictionary\" target=\"_blank\">https://www.proteinatlas.org/learn/dictionary</a></p>",
      "rawMarkdown": "The Human Protein Atlas Dictionary may be a good resource for this\nhttps://www.proteinatlas.org/learn/dictionary",
      "votes": null
    },
    {
      "id": "1837557",
      "postDate": "06/29/2022 17:02:22",
      "content": "<p>one may note that if you set empty  rle='' for lung in submission csv, your lb score drop by only about 0.02.<br>\nthis means that we are not detecting the alveolus for the lung at all</p>\n<p>(this is the same for validation set as well)</p>",
      "rawMarkdown": "one may note that if you set empty  rle='' for lung in submission csv, your lb score drop by only about 0.02.\nthis means that we are not detecting the alveolus for the lung at all\n\n(this is the same for validation set as well)",
      "votes": null
    },
    {
      "id": "1837572",
      "postDate": "06/29/2022 17:22:47",
      "content": "<p>one reason is the high inconsistency in label</p>\n<p><img src=\"https://i.ibb.co/64dN6Zh/Selection-063.png\" alt=\"https://i.ibb.co/64dN6Zh/Selection-063.png\"></p>",
      "rawMarkdown": "one reason is the high inconsistency in label\n\n![https://i.ibb.co/64dN6Zh/Selection-063.png](https://i.ibb.co/64dN6Zh/Selection-063.png)",
      "votes": null
    },
    {
      "id": "1851800",
      "postDate": "07/11/2022 14:47:01",
      "content": "<p>There is no way to learn annotators' inconsistencies so I wouldn't bother dealing with them. Most logical way would be predicting them and hoping that the hubmap is more consistent. </p>",
      "rawMarkdown": "There is no way to learn annotators' inconsistencies so I wouldn't bother dealing with them. Most logical way would be predicting them and hoping that the hubmap is more consistent.",
      "votes": null
    },
    {
      "id": "1916621",
      "postDate": "08/28/2022 03:22:11",
      "content": "<p>Dear <a href=\"https://www.kaggle.com/hengck23\" target=\"_blank\">@hengck23</a>, I deleted inconsistent images from the  training set. My lung score is 0.03 in hubmap, for other organs score is normal. How did you improve the alveolar detection?</p>",
      "rawMarkdown": "Dear @hengck23, I deleted inconsistent images from the  training set. My lung score is 0.03 in hubmap, for other organs score is normal. How did you improve the alveolar detection?",
      "votes": null
    },
    {
      "id": "1916636",
      "postDate": "08/28/2022 03:45:47",
      "content": "<p><a href=\"https://www.kaggle.com/ynhuhu\" target=\"_blank\">@ynhuhu</a> <br>\nyou can forget about lung for the time being.<br>\njust set  threshold=0.10 for lung improve other organs first.<br>\nthink of lung after you get LB in the range 0.80</p>\n<p>for LB 0.80~0.82:<br>\nlocal CV for lung 0.23~0.30<br>\nlb-hubmap 0.08~0.10</p>\n<p>tip: set threshold=0.10 for lung</p>",
      "rawMarkdown": "ynhuhu \nyou can forget about lung for the time being.\njust set  threshold=0.10 for lung improve other organs first.\nthink of lung after you get LB in the range 0.80\n\n\nfor LB 0.80~0.82:\nlocal CV for lung 0.23~0.30\nlb-hubmap 0.08~0.10\n\ntip: set threshold=0.10 for lung",
      "votes": null
    },
    {
      "id": "1916637",
      "postDate": "08/28/2022 03:47:19",
      "content": "<blockquote>\n  <blockquote>\n    <p>I deleted inconsistent images</p>\n  </blockquote>\n</blockquote>\n<p>the correct way is not to delete or add label by hand.</p>\n<p>it is to use machine learning to relabel in such a way to improve LB score.<br>\n(i.e. you pesudo label the lung. but you need a good model first.)</p>",
      "rawMarkdown": ">>I deleted inconsistent images\n\nthe correct way is not to delete or add label by hand.\n\nit is to use machine learning to relabel in such a way to improve LB score.\n(i.e. you pesudo label the lung. but you need a good model first.)",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1834800,
      "author_name": "emmalumpan",
      "author_url": "",
      "post_date": "06/27/2022 07:48:50",
      "content": "<p>The Human Protein Atlas Dictionary may be a good resource for this<br>\n<a href=\"https://www.proteinatlas.org/learn/dictionary\" target=\"_blank\">https://www.proteinatlas.org/learn/dictionary</a></p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1837557,
      "author_name": "hengck23",
      "author_url": "",
      "post_date": "06/29/2022 17:02:22",
      "content": "<p>one may note that if you set empty  rle='' for lung in submission csv, your lb score drop by only about 0.02.<br>\nthis means that we are not detecting the alveolus for the lung at all</p>\n<p>(this is the same for validation set as well)</p>",
      "votes": null,
      "replies": [
        {
          "id": 1837572,
          "author_name": "hengck23",
          "author_url": "",
          "post_date": "06/29/2022 17:22:47",
          "content": "<p>one reason is the high inconsistency in label</p>\n<p><img src=\"https://i.ibb.co/64dN6Zh/Selection-063.png\" alt=\"https://i.ibb.co/64dN6Zh/Selection-063.png\"></p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1851800,
          "author_name": "gunesevitan",
          "author_url": "",
          "post_date": "07/11/2022 14:47:01",
          "content": "<p>There is no way to learn annotators' inconsistencies so I wouldn't bother dealing with them. Most logical way would be predicting them and hoping that the hubmap is more consistent. </p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1916621,
          "author_name": "ynhuhu",
          "author_url": "",
          "post_date": "08/28/2022 03:22:11",
          "content": "<p>Dear <a href=\"https://www.kaggle.com/hengck23\" target=\"_blank\">@hengck23</a>, I deleted inconsistent images from the  training set. My lung score is 0.03 in hubmap, for other organs score is normal. How did you improve the alveolar detection?</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1916636,
          "author_name": "hengck23",
          "author_url": "",
          "post_date": "08/28/2022 03:45:47",
          "content": "<p><a href=\"https://www.kaggle.com/ynhuhu\" target=\"_blank\">@ynhuhu</a> <br>\nyou can forget about lung for the time being.<br>\njust set  threshold=0.10 for lung improve other organs first.<br>\nthink of lung after you get LB in the range 0.80</p>\n<p>for LB 0.80~0.82:<br>\nlocal CV for lung 0.23~0.30<br>\nlb-hubmap 0.08~0.10</p>\n<p>tip: set threshold=0.10 for lung</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1916637,
          "author_name": "hengck23",
          "author_url": "",
          "post_date": "08/28/2022 03:47:19",
          "content": "<blockquote>\n  <blockquote>\n    <p>I deleted inconsistent images</p>\n  </blockquote>\n</blockquote>\n<p>the correct way is not to delete or add label by hand.</p>\n<p>it is to use machine learning to relabel in such a way to improve LB score.<br>\n(i.e. you pesudo label the lung. but you need a good model first.)</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1834778": "any good resources on how to identify lung alveolus?\n\nAfter reading through the web and watching youtube videos, it is difficult for me to locate lung alveolus.\nanyone has any tips or advices?",
    "1834800": "The Human Protein Atlas Dictionary may be a good resource for this\nhttps://www.proteinatlas.org/learn/dictionary",
    "1837557": "one may note that if you set empty  rle='' for lung in submission csv, your lb score drop by only about 0.02.\nthis means that we are not detecting the alveolus for the lung at all\n\n(this is the same for validation set as well)",
    "1837572": "one reason is the high inconsistency in label\n\n![https://i.ibb.co/64dN6Zh/Selection-063.png](https://i.ibb.co/64dN6Zh/Selection-063.png)",
    "1851800": "There is no way to learn annotators' inconsistencies so I wouldn't bother dealing with them. Most logical way would be predicting them and hoping that the hubmap is more consistent.",
    "1916621": "Dear @hengck23, I deleted inconsistent images from the  training set. My lung score is 0.03 in hubmap, for other organs score is normal. How did you improve the alveolar detection?",
    "1916636": "ynhuhu \nyou can forget about lung for the time being.\njust set  threshold=0.10 for lung improve other organs first.\nthink of lung after you get LB in the range 0.80\n\n\nfor LB 0.80~0.82:\nlocal CV for lung 0.23~0.30\nlb-hubmap 0.08~0.10\n\ntip: set threshold=0.10 for lung",
    "1916637": ">>I deleted inconsistent images\n\nthe correct way is not to delete or add label by hand.\n\nit is to use machine learning to relabel in such a way to improve LB score.\n(i.e. you pesudo label the lung. but you need a good model first.)"
  },
  "source": "meta"
}