{
  "id": 70220,
  "title": "Non-Consistent Organelle Labeling",
  "url": "/competitions/human-protein-atlas-image-classification/discussion/70220",
  "author_name": "",
  "post_date": "2018-11-01T03:31:05.700091200Z",
  "votes": 2,
  "comment_count": 4,
  "views": 0,
  "content": "<p>Many of labeled images do not have consistent labeling regarding to the located organelles by the protein of interest.</p>\n\n<p>Example:</p>\n\n<ol>\n<li><p>Label: Nucleoplam, ID: 01095156-bbb7-11e8-b2ba-ac1f6b6435d0</p></li>\n<li><p>Label: Nucleoli, ID: 0115f496-bbb8-11e8-b2ba-ac1f6b6435d0</p></li>\n</ol>\n\n<p>Both images have clear evidence of having \"Nucleoplam\" located by the protein of interest. However 2 is only labeled by Nucleoli.</p>\n\n<p>We can find many of these cases in the labeled data.</p>\n\n<p>Is this expected?\nWill the test data contains the same non-consistent organelle labeling?</p>",
  "messages": [
    {
      "id": "413499",
      "postDate": "11/01/2018 03:31:05",
      "content": "<p>Many of labeled images do not have consistent labeling regarding to the located organelles by the protein of interest.</p>\n\n<p>Example:</p>\n\n<ol>\n<li><p>Label: Nucleoplam, ID: 01095156-bbb7-11e8-b2ba-ac1f6b6435d0</p></li>\n<li><p>Label: Nucleoli, ID: 0115f496-bbb8-11e8-b2ba-ac1f6b6435d0</p></li>\n</ol>\n\n<p>Both images have clear evidence of having \"Nucleoplam\" located by the protein of interest. However 2 is only labeled by Nucleoli.</p>\n\n<p>We can find many of these cases in the labeled data.</p>\n\n<p>Is this expected?\nWill the test data contains the same non-consistent organelle labeling?</p>",
      "rawMarkdown": "Many of labeled images do not have consistent labeling regarding to the located organelles by the protein of interest.\n\nExample:\n\n 1.  Label: Nucleoplam, ID: 01095156-bbb7-11e8-b2ba-ac1f6b6435d0\n\n 2. Label: Nucleoli, ID: 0115f496-bbb8-11e8-b2ba-ac1f6b6435d0\n\nBoth images have clear evidence of having \"Nucleoplam\" located by the protein of interest. However 2 is only labeled by Nucleoli.\n\nWe can find many of these cases in the labeled data.\n\nIs this expected?\nWill the test data contains the same non-consistent organelle labeling?",
      "votes": null
    },
    {
      "id": "413544",
      "postDate": "11/01/2018 05:26:27",
      "content": "<p>Will try to run it.</p>",
      "rawMarkdown": "Will try to run it.",
      "votes": null
    },
    {
      "id": "413627",
      "postDate": "11/01/2018 08:57:10",
      "content": "<p>I believe so. There was another discussion topic pointing out, that the labels have been probably assigned by volunteers in crowdsourcing project (if I understood it correctly). We can a priori expect, there can be some data quality issues. Like in almost any real-life data science applications.</p>",
      "rawMarkdown": "I believe so. There was another discussion topic pointing out, that the labels have been probably assigned by volunteers in crowdsourcing project (if I understood it correctly). We can a priori expect, there can be some data quality issues. Like in almost any real-life data science applications.",
      "votes": null
    },
    {
      "id": "413757",
      "postDate": "11/01/2018 13:36:39",
      "content": "<p>i think the crowd paper did mention the amount of \"label noise\"</p>",
      "rawMarkdown": "i think the crowd paper did mention the amount of \"label noise\"",
      "votes": null
    },
    {
      "id": "413962",
      "postDate": "11/01/2018 21:50:58",
      "content": "<p>One thing that is interesting to note, the training labels appear to be ordered but not sorted numerically.</p>",
      "rawMarkdown": "One thing that is interesting to note, the training labels appear to be ordered but not sorted numerically.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 413544,
      "author_name": "shubhamtatvamasi",
      "author_url": "",
      "post_date": "11/01/2018 05:26:27",
      "content": "<p>Will try to run it.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 413627,
      "author_name": "rejpalcz",
      "author_url": "",
      "post_date": "11/01/2018 08:57:10",
      "content": "<p>I believe so. There was another discussion topic pointing out, that the labels have been probably assigned by volunteers in crowdsourcing project (if I understood it correctly). We can a priori expect, there can be some data quality issues. Like in almost any real-life data science applications.</p>",
      "votes": null,
      "replies": [
        {
          "id": 413757,
          "author_name": "hengck23",
          "author_url": "",
          "post_date": "11/01/2018 13:36:39",
          "content": "<p>i think the crowd paper did mention the amount of \"label noise\"</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 413962,
          "author_name": "ldm314",
          "author_url": "",
          "post_date": "11/01/2018 21:50:58",
          "content": "<p>One thing that is interesting to note, the training labels appear to be ordered but not sorted numerically.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "413499": "Many of labeled images do not have consistent labeling regarding to the located organelles by the protein of interest.\n\nExample:\n\n 1.  Label: Nucleoplam, ID: 01095156-bbb7-11e8-b2ba-ac1f6b6435d0\n\n 2. Label: Nucleoli, ID: 0115f496-bbb8-11e8-b2ba-ac1f6b6435d0\n\nBoth images have clear evidence of having \"Nucleoplam\" located by the protein of interest. However 2 is only labeled by Nucleoli.\n\nWe can find many of these cases in the labeled data.\n\nIs this expected?\nWill the test data contains the same non-consistent organelle labeling?",
    "413544": "Will try to run it.",
    "413627": "I believe so. There was another discussion topic pointing out, that the labels have been probably assigned by volunteers in crowdsourcing project (if I understood it correctly). We can a priori expect, there can be some data quality issues. Like in almost any real-life data science applications.",
    "413757": "i think the crowd paper did mention the amount of \"label noise\"",
    "413962": "One thing that is interesting to note, the training labels appear to be ordered but not sorted numerically."
  },
  "source": "meta"
}