{
  "id": 225100,
  "title": "Difference in mask before and after image data update",
  "url": "/competitions/hubmap-kidney-segmentation/discussion/225100",
  "author_name": "",
  "post_date": "2021-03-10T20:49:46.800519100Z",
  "votes": 22,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Hello, everyone.<br>\nThe image data has finally been updated, and the reason for the update might be, as you all know</p>\n<ul>\n<li>In some images, the masks (of 095bf7a1f, e79de561c, and 54f2eec69), which should be Ground Truth, was slightly off</li>\n<li>Low annotation quality</li>\n<li>The private data set was open (public)</li>\n</ul>\n<p>I am personally most concerned about the quality of the annotations, because if the private dataset is poorly annotated, then no matter how good a model we create, it will not be validly evaluated in private.</p>\n<p>This time, the host seemed to have spent a lot of effort to update the masks and new images, so I  compared the state of the masks before and after the update (sorry for the small image).<br>\nHere are the two things I noticed.</p>\n<ul>\n<li>I got the impression that the masks that were a little off are being shifted by the host to correct the misalignment.</li>\n<li>Masks that existed in places where glomeruli did not exist seem to have been removed (I did not verify all of them).</li>\n</ul>\n<p>If you notice anything, please share it with us. Thank you.</p>\n<p><code>095bf7a1f.tiff</code><br>\n<img src=\"https://f.easyuploader.app/eu-prd/upload/20210311054615_4a34724f.jpg\" alt=\"\"></p>\n<p><code>e79de561c.tiff</code><br>\n<img src=\"https://f.easyuploader.app/eu-prd/upload/20210311054608_68585464.jpg\" alt=\"\"></p>\n<p><code>54f2eec69.tiff</code><br>\n<img src=\"https://f.easyuploader.app/eu-prd/upload/20210311054608_76463175.jpg\" alt=\"\"></p>\n<p>Happy Kaggling!</p>",
  "messages": [
    {
      "id": "1233991",
      "postDate": "03/10/2021 20:49:46",
      "content": "<p>Hello, everyone.<br>\nThe image data has finally been updated, and the reason for the update might be, as you all know</p>\n<ul>\n<li>In some images, the masks (of 095bf7a1f, e79de561c, and 54f2eec69), which should be Ground Truth, was slightly off</li>\n<li>Low annotation quality</li>\n<li>The private data set was open (public)</li>\n</ul>\n<p>I am personally most concerned about the quality of the annotations, because if the private dataset is poorly annotated, then no matter how good a model we create, it will not be validly evaluated in private.</p>\n<p>This time, the host seemed to have spent a lot of effort to update the masks and new images, so I  compared the state of the masks before and after the update (sorry for the small image).<br>\nHere are the two things I noticed.</p>\n<ul>\n<li>I got the impression that the masks that were a little off are being shifted by the host to correct the misalignment.</li>\n<li>Masks that existed in places where glomeruli did not exist seem to have been removed (I did not verify all of them).</li>\n</ul>\n<p>If you notice anything, please share it with us. Thank you.</p>\n<p><code>095bf7a1f.tiff</code><br>\n<img src=\"https://f.easyuploader.app/eu-prd/upload/20210311054615_4a34724f.jpg\" alt=\"\"></p>\n<p><code>e79de561c.tiff</code><br>\n<img src=\"https://f.easyuploader.app/eu-prd/upload/20210311054608_68585464.jpg\" alt=\"\"></p>\n<p><code>54f2eec69.tiff</code><br>\n<img src=\"https://f.easyuploader.app/eu-prd/upload/20210311054608_76463175.jpg\" alt=\"\"></p>\n<p>Happy Kaggling!</p>",
      "rawMarkdown": "Hello, everyone.\nThe image data has finally been updated, and the reason for the update might be, as you all know\n\n- In some images, the masks (of 095bf7a1f, e79de561c, and 54f2eec69), which should be Ground Truth, was slightly off\n- Low annotation quality\n- The private data set was open (public)\n\nI am personally most concerned about the quality of the annotations, because if the private dataset is poorly annotated, then no matter how good a model we create, it will not be validly evaluated in private.\n\nThis time, the host seemed to have spent a lot of effort to update the masks and new images, so I  compared the state of the masks before and after the update (sorry for the small image).\nHere are the two things I noticed.\n\n- I got the impression that the masks that were a little off are being shifted by the host to correct the misalignment.\n- Masks that existed in places where glomeruli did not exist seem to have been removed (I did not verify all of them).\n\nIf you notice anything, please share it with us. Thank you.\n  \n  \n`095bf7a1f.tiff`\n![](https://f.easyuploader.app/eu-prd/upload/20210311054615_4a34724f.jpg)\n  \n  \n`e79de561c.tiff`\n![](https://f.easyuploader.app/eu-prd/upload/20210311054608_68585464.jpg)\n\n`54f2eec69.tiff`\n![](https://f.easyuploader.app/eu-prd/upload/20210311054608_76463175.jpg)\n\nHappy Kaggling!",
      "votes": null
    },
    {
      "id": "1237678",
      "postDate": "03/14/2021 11:17:14",
      "content": "<p>Thank you for the post <a href=\"https://www.kaggle.com/maxwell110\" target=\"_blank\">@maxwell110</a> , will definitely bring back to you I notice anything. <br>\nHappy kaggling.</p>",
      "rawMarkdown": "Thank you for the post @maxwell110 , will definitely bring back to you I notice anything. \nHappy kaggling.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1237678,
      "author_name": "thrineshduvvuru",
      "author_url": "",
      "post_date": "03/14/2021 11:17:14",
      "content": "<p>Thank you for the post <a href=\"https://www.kaggle.com/maxwell110\" target=\"_blank\">@maxwell110</a> , will definitely bring back to you I notice anything. <br>\nHappy kaggling.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1233991": "Hello, everyone.\nThe image data has finally been updated, and the reason for the update might be, as you all know\n\n- In some images, the masks (of 095bf7a1f, e79de561c, and 54f2eec69), which should be Ground Truth, was slightly off\n- Low annotation quality\n- The private data set was open (public)\n\nI am personally most concerned about the quality of the annotations, because if the private dataset is poorly annotated, then no matter how good a model we create, it will not be validly evaluated in private.\n\nThis time, the host seemed to have spent a lot of effort to update the masks and new images, so I  compared the state of the masks before and after the update (sorry for the small image).\nHere are the two things I noticed.\n\n- I got the impression that the masks that were a little off are being shifted by the host to correct the misalignment.\n- Masks that existed in places where glomeruli did not exist seem to have been removed (I did not verify all of them).\n\nIf you notice anything, please share it with us. Thank you.\n  \n  \n`095bf7a1f.tiff`\n![](https://f.easyuploader.app/eu-prd/upload/20210311054615_4a34724f.jpg)\n  \n  \n`e79de561c.tiff`\n![](https://f.easyuploader.app/eu-prd/upload/20210311054608_68585464.jpg)\n\n`54f2eec69.tiff`\n![](https://f.easyuploader.app/eu-prd/upload/20210311054608_76463175.jpg)\n\nHappy Kaggling!",
    "1237678": "Thank you for the post @maxwell110 , will definitely bring back to you I notice anything. \nHappy kaggling."
  },
  "source": "meta"
}