{
  "id": 418009,
  "title": "De-duplicated annotations - dataset",
  "url": "/competitions/hubmap-hacking-the-human-vasculature/discussion/418009",
  "author_name": "",
  "post_date": "2023-06-18T10:16:43.839021800Z",
  "votes": 13,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Just for those who haven't seen it, but about 6% of the labels are duplicates. <br>\nThis is likely not a huge issue, but might skew your validation score a bit and could introduce a little bit of bias in your training. </p>\n<p>I've created a de-duplicated version of the annotations here, in exactly the same format as the ones supplied by the organizers: <a href=\"https://www.kaggle.com/datasets/fnands/de-duplicated-annotations-for-hubmap-hhv\" target=\"_blank\">https://www.kaggle.com/datasets/fnands/de-duplicated-annotations-for-hubmap-hhv</a></p>\n<p>Of course, this begs the question if the testing annotations are also duplicated, but not much we can do about that. </p>",
  "messages": [
    {
      "id": "2307609",
      "postDate": "06/18/2023 10:16:43",
      "content": "<p>Just for those who haven't seen it, but about 6% of the labels are duplicates. <br>\nThis is likely not a huge issue, but might skew your validation score a bit and could introduce a little bit of bias in your training. </p>\n<p>I've created a de-duplicated version of the annotations here, in exactly the same format as the ones supplied by the organizers: <a href=\"https://www.kaggle.com/datasets/fnands/de-duplicated-annotations-for-hubmap-hhv\" target=\"_blank\">https://www.kaggle.com/datasets/fnands/de-duplicated-annotations-for-hubmap-hhv</a></p>\n<p>Of course, this begs the question if the testing annotations are also duplicated, but not much we can do about that. </p>",
      "rawMarkdown": "Just for those who haven't seen it, but about 6% of the labels are duplicates. \nThis is likely not a huge issue, but might skew your validation score a bit and could introduce a little bit of bias in your training. \n\nI've created a de-duplicated version of the annotations here, in exactly the same format as the ones supplied by the organizers: https://www.kaggle.com/datasets/fnands/de-duplicated-annotations-for-hubmap-hhv\n\n\nOf course, this begs the question if the testing annotations are also duplicated, but not much we can do about that.",
      "votes": null
    },
    {
      "id": "2317132",
      "postDate": "06/25/2023 13:17:18",
      "content": "<p>Thanks for sharing the dataset, I'll definitely be using it. </p>",
      "rawMarkdown": "Thanks for sharing the dataset, I'll definitely be using it.",
      "votes": null
    },
    {
      "id": "2351955",
      "postDate": "07/20/2023 14:12:22",
      "content": "<p>Are you sure the duplicates are 6%? Raw annotation count is 17518 and filtered annotation count is 16696 for me. I eliminated 822 annotations which is 4.7%.</p>",
      "rawMarkdown": "Are you sure the duplicates are 6%? Raw annotation count is 17518 and filtered annotation count is 16696 for me. I eliminated 822 annotations which is 4.7%.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2317132,
      "author_name": "samuelcortinhas",
      "author_url": "",
      "post_date": "06/25/2023 13:17:18",
      "content": "<p>Thanks for sharing the dataset, I'll definitely be using it. </p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 2351955,
      "author_name": "gunesevitan",
      "author_url": "",
      "post_date": "07/20/2023 14:12:22",
      "content": "<p>Are you sure the duplicates are 6%? Raw annotation count is 17518 and filtered annotation count is 16696 for me. I eliminated 822 annotations which is 4.7%.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2307609": "Just for those who haven't seen it, but about 6% of the labels are duplicates. \nThis is likely not a huge issue, but might skew your validation score a bit and could introduce a little bit of bias in your training. \n\nI've created a de-duplicated version of the annotations here, in exactly the same format as the ones supplied by the organizers: https://www.kaggle.com/datasets/fnands/de-duplicated-annotations-for-hubmap-hhv\n\n\nOf course, this begs the question if the testing annotations are also duplicated, but not much we can do about that.",
    "2317132": "Thanks for sharing the dataset, I'll definitely be using it.",
    "2351955": "Are you sure the duplicates are 6%? Raw annotation count is 17518 and filtered annotation count is 16696 for me. I eliminated 822 annotations which is 4.7%."
  },
  "source": "meta"
}