{
  "id": 285825,
  "title": "Question about metric at test time",
  "url": "/competitions/sartorius-cell-instance-segmentation/discussion/285825",
  "author_name": "",
  "post_date": "2021-11-06T12:55:18.797656100Z",
  "votes": 7,
  "comment_count": 2,
  "views": 0,
  "content": "<p>In evaluation page, it's stated that</p>\n<blockquote>\n  <p>It also checks that no two predicted masks for the same image are overlapping.</p>\n</blockquote>\n<p>Does it means ground truth labels for test set are processed so that they don't have overlapping area? Or, do they have overlapping like annotations for train set?</p>\n<p>If the former is correct, It's preferable if the method to eliminate overlapping is stated in description area as it tells us what Sartorius team want us to do. Or, it also works if you provide non-overlapping version of train annotations.</p>\n<p>Thanks,</p>",
  "messages": [
    {
      "id": "1573264",
      "postDate": "11/06/2021 12:55:18",
      "content": "<p>In evaluation page, it's stated that</p>\n<blockquote>\n  <p>It also checks that no two predicted masks for the same image are overlapping.</p>\n</blockquote>\n<p>Does it means ground truth labels for test set are processed so that they don't have overlapping area? Or, do they have overlapping like annotations for train set?</p>\n<p>If the former is correct, It's preferable if the method to eliminate overlapping is stated in description area as it tells us what Sartorius team want us to do. Or, it also works if you provide non-overlapping version of train annotations.</p>\n<p>Thanks,</p>",
      "rawMarkdown": "In evaluation page, it's stated that\n> It also checks that no two predicted masks for the same image are overlapping.\n\nDoes it means ground truth labels for test set are processed so that they don't have overlapping area? Or, do they have overlapping like annotations for train set?\n\nIf the former is correct, It's preferable if the method to eliminate overlapping is stated in description area as it tells us what Sartorius team want us to do. Or, it also works if you provide non-overlapping version of train annotations.\n\nThanks,",
      "votes": null
    },
    {
      "id": "1573435",
      "postDate": "11/06/2021 16:01:51",
      "content": "<p>Please check host explanation <a href=\"https://www.kaggle.com/c/sartorius-cell-instance-segmentation/discussion/280250#1571081\" target=\"_blank\">here</a>.</p>",
      "rawMarkdown": "Please check host explanation [here](https://www.kaggle.com/c/sartorius-cell-instance-segmentation/discussion/280250#1571081).",
      "votes": null
    },
    {
      "id": "1573878",
      "postDate": "11/07/2021 02:13:58",
      "content": "<p>Thanks, I haven't checked this thread! <br>\nIt should be pinned at the top of discussions cause it's so important information🤔</p>",
      "rawMarkdown": "Thanks, I haven't checked this thread! \nIt should be pinned at the top of discussions cause it's so important information🤔",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1573435,
      "author_name": "rednikotin",
      "author_url": "",
      "post_date": "11/06/2021 16:01:51",
      "content": "<p>Please check host explanation <a href=\"https://www.kaggle.com/c/sartorius-cell-instance-segmentation/discussion/280250#1571081\" target=\"_blank\">here</a>.</p>",
      "votes": null,
      "replies": [
        {
          "id": 1573878,
          "author_name": "bamps53",
          "author_url": "",
          "post_date": "11/07/2021 02:13:58",
          "content": "<p>Thanks, I haven't checked this thread! <br>\nIt should be pinned at the top of discussions cause it's so important information🤔</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1573264": "In evaluation page, it's stated that\n> It also checks that no two predicted masks for the same image are overlapping.\n\nDoes it means ground truth labels for test set are processed so that they don't have overlapping area? Or, do they have overlapping like annotations for train set?\n\nIf the former is correct, It's preferable if the method to eliminate overlapping is stated in description area as it tells us what Sartorius team want us to do. Or, it also works if you provide non-overlapping version of train annotations.\n\nThanks,",
    "1573435": "Please check host explanation [here](https://www.kaggle.com/c/sartorius-cell-instance-segmentation/discussion/280250#1571081).",
    "1573878": "Thanks, I haven't checked this thread! \nIt should be pinned at the top of discussions cause it's so important information🤔"
  },
  "source": "meta"
}