{
  "id": 232577,
  "title": "Model prediction error? Problem with annotations?",
  "url": "/competitions/hubmap-kidney-segmentation/discussion/232577",
  "author_name": "",
  "post_date": "2021-04-14T12:26:15.431359200Z",
  "votes": 5,
  "comment_count": 2,
  "views": 0,
  "content": "<p>After visually analyzing the prediction results of <strong>aaa6a05cc</strong>, the following problems were discovered （<strong>The red represents groundtruth, and green represents the prediction</strong>. ）:</p>\n<p><img src=\"https://i.loli.net/2021/04/14/9wKWmfGTSOR6EAd.png\" alt=\"image _2_.png\"><br>\nIs it really a glomerulus in the upper left corner?<br>\n<img src=\"https://i.loli.net/2021/04/14/VZDpTudKhsCi2w6.png\" alt=\"image _1_.png\"><br>\nFor the semantic segmentation of pixel-level classification, the pixels close to the background should be correctly classified as glomeruli. Should more consideration be given to the overall contour information (for example, some shape priors are introduced as post-processing)?<br>\n<img src=\"https://i.loli.net/2021/04/14/37WjAToiK4kOpc6.png\" alt=\"image.png\"><br>\nIs the upper right corner missing the annotations? Need a professional pathologist to explain. ( :</p>",
  "messages": [
    {
      "id": "1273504",
      "postDate": "04/14/2021 12:26:15",
      "content": "<p>After visually analyzing the prediction results of <strong>aaa6a05cc</strong>, the following problems were discovered （<strong>The red represents groundtruth, and green represents the prediction</strong>. ）:</p>\n<p><img src=\"https://i.loli.net/2021/04/14/9wKWmfGTSOR6EAd.png\" alt=\"image _2_.png\"><br>\nIs it really a glomerulus in the upper left corner?<br>\n<img src=\"https://i.loli.net/2021/04/14/VZDpTudKhsCi2w6.png\" alt=\"image _1_.png\"><br>\nFor the semantic segmentation of pixel-level classification, the pixels close to the background should be correctly classified as glomeruli. Should more consideration be given to the overall contour information (for example, some shape priors are introduced as post-processing)?<br>\n<img src=\"https://i.loli.net/2021/04/14/37WjAToiK4kOpc6.png\" alt=\"image.png\"><br>\nIs the upper right corner missing the annotations? Need a professional pathologist to explain. ( :</p>",
      "rawMarkdown": "After visually analyzing the prediction results of **aaa6a05cc**, the following problems were discovered （**The red represents groundtruth, and green represents the prediction**. ）:\n\n![image _2_.png](https://i.loli.net/2021/04/14/9wKWmfGTSOR6EAd.png)\nIs it really a glomerulus in the upper left corner?\n![image _1_.png](https://i.loli.net/2021/04/14/VZDpTudKhsCi2w6.png)\nFor the semantic segmentation of pixel-level classification, the pixels close to the background should be correctly classified as glomeruli. Should more consideration be given to the overall contour information (for example, some shape priors are introduced as post-processing)?\n![image.png](https://i.loli.net/2021/04/14/37WjAToiK4kOpc6.png)\nIs the upper right corner missing the annotations? Need a professional pathologist to explain. ( :",
      "votes": null
    },
    {
      "id": "1273970",
      "postDate": "04/14/2021 20:21:23",
      "content": "<p>Many users have found those odd behavior on the ground truth after the last update.  Sad, but probably, you would do better  removing  them from your model , or hand annotate if you can ( I still want to know a good way to do it).</p>",
      "rawMarkdown": "Many users have found those odd behavior on the ground truth after the last update.  Sad, but probably, you would do better  removing  them from your model , or hand annotate if you can ( I still want to know a good way to do it).",
      "votes": null
    },
    {
      "id": "1275278",
      "postDate": "04/16/2021 06:32:08",
      "content": "<p>jiu hen mi</p>",
      "rawMarkdown": "jiu hen mi",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1273970,
      "author_name": "rpsantosakaggle",
      "author_url": "",
      "post_date": "04/14/2021 20:21:23",
      "content": "<p>Many users have found those odd behavior on the ground truth after the last update.  Sad, but probably, you would do better  removing  them from your model , or hand annotate if you can ( I still want to know a good way to do it).</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1275278,
      "author_name": "southsakura",
      "author_url": "",
      "post_date": "04/16/2021 06:32:08",
      "content": "<p>jiu hen mi</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1273504": "After visually analyzing the prediction results of **aaa6a05cc**, the following problems were discovered （**The red represents groundtruth, and green represents the prediction**. ）:\n\n![image _2_.png](https://i.loli.net/2021/04/14/9wKWmfGTSOR6EAd.png)\nIs it really a glomerulus in the upper left corner?\n![image _1_.png](https://i.loli.net/2021/04/14/VZDpTudKhsCi2w6.png)\nFor the semantic segmentation of pixel-level classification, the pixels close to the background should be correctly classified as glomeruli. Should more consideration be given to the overall contour information (for example, some shape priors are introduced as post-processing)?\n![image.png](https://i.loli.net/2021/04/14/37WjAToiK4kOpc6.png)\nIs the upper right corner missing the annotations? Need a professional pathologist to explain. ( :",
    "1273970": "Many users have found those odd behavior on the ground truth after the last update.  Sad, but probably, you would do better  removing  them from your model , or hand annotate if you can ( I still want to know a good way to do it).",
    "1275278": "jiu hen mi"
  },
  "source": "meta"
}