{
  "id": 466921,
  "title": "Labels missing from dense datasets?",
  "url": "/competitions/blood-vessel-segmentation/discussion/466921",
  "author_name": "",
  "post_date": "2024-01-10T12:31:41.999812600Z",
  "votes": null,
  "comment_count": 3,
  "views": 0,
  "content": "<p>Quite often my model sees, what I think are, tiny vessels that are missing from the annotations (in kidney_1_dense, kidney_1_voi or kidney_3). Sometimes only the start of a vessel is annotated and then in the prediction it seems to continue, which makes me believe I'm not mistakenly detecting veins or something. But maybe I'm wrong… there are a few possible explanations:</p>\n<ul>\n<li>The annotations are missing</li>\n<li>I'm detecting veins instead of vessels</li>\n<li>The model is hallucinating</li>\n</ul>\n<p>Anyone have similar experience or an explanation maybe?</p>\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F3086083%2Fcbd8afa091f702edca066ccb01e217ad%2Fkidney_more.png?generation=1704889801023801&amp;alt=media\" alt=\"\"></p>",
  "messages": [
    {
      "id": "2595440",
      "postDate": "01/10/2024 12:31:42",
      "content": "<p>Quite often my model sees, what I think are, tiny vessels that are missing from the annotations (in kidney_1_dense, kidney_1_voi or kidney_3). Sometimes only the start of a vessel is annotated and then in the prediction it seems to continue, which makes me believe I'm not mistakenly detecting veins or something. But maybe I'm wrong… there are a few possible explanations:</p>\n<ul>\n<li>The annotations are missing</li>\n<li>I'm detecting veins instead of vessels</li>\n<li>The model is hallucinating</li>\n</ul>\n<p>Anyone have similar experience or an explanation maybe?</p>\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F3086083%2Fcbd8afa091f702edca066ccb01e217ad%2Fkidney_more.png?generation=1704889801023801&amp;alt=media\" alt=\"\"></p>",
      "rawMarkdown": "Quite often my model sees, what I think are, tiny vessels that are missing from the annotations (in kidney_1_dense, kidney_1_voi or kidney_3). Sometimes only the start of a vessel is annotated and then in the prediction it seems to continue, which makes me believe I'm not mistakenly detecting veins or something. But maybe I'm wrong... there are a few possible explanations:\n\n* The annotations are missing\n* I'm detecting veins instead of vessels\n* The model is hallucinating\n\nAnyone have similar experience or an explanation maybe?\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F3086083%2Fcbd8afa091f702edca066ccb01e217ad%2Fkidney_more.png?generation=1704889801023801&alt=media)",
      "votes": null
    },
    {
      "id": "2595776",
      "postDate": "01/10/2024 16:32:00",
      "content": "<p>i already mention the missing annotation:<br>\n<a href=\"https://www.kaggle.com/competitions/blood-vessel-segmentation/discussion/456118#2583079\" target=\"_blank\">https://www.kaggle.com/competitions/blood-vessel-segmentation/discussion/456118#2583079</a><br>\n<a href=\"https://www.kaggle.com/competitions/blood-vessel-segmentation/discussion/456118#2589018\" target=\"_blank\">https://www.kaggle.com/competitions/blood-vessel-segmentation/discussion/456118#2589018</a></p>",
      "rawMarkdown": "i already mention the missing annotation:\nhttps://www.kaggle.com/competitions/blood-vessel-segmentation/discussion/456118#2583079\nhttps://www.kaggle.com/competitions/blood-vessel-segmentation/discussion/456118#2589018",
      "votes": null
    },
    {
      "id": "2595778",
      "postDate": "01/10/2024 16:33:19",
      "content": "<p>you can also do a 3d floodfill using seed = junction of missing vessel</p>",
      "rawMarkdown": "you can also do a 3d floodfill using seed = junction of missing vessel",
      "votes": null
    },
    {
      "id": "2595791",
      "postDate": "01/10/2024 16:42:54",
      "content": "<p>Ah I see 👍 I feel a bit weird adapting the model to ignore the smallest vessel detections.. just for a better score.. but maybe that's the way forward…</p>",
      "rawMarkdown": "Ah I see 👍 I feel a bit weird adapting the model to ignore the smallest vessel detections.. just for a better score.. but maybe that's the way forward...",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2595776,
      "author_name": "hengck23",
      "author_url": "",
      "post_date": "01/10/2024 16:32:00",
      "content": "<p>i already mention the missing annotation:<br>\n<a href=\"https://www.kaggle.com/competitions/blood-vessel-segmentation/discussion/456118#2583079\" target=\"_blank\">https://www.kaggle.com/competitions/blood-vessel-segmentation/discussion/456118#2583079</a><br>\n<a href=\"https://www.kaggle.com/competitions/blood-vessel-segmentation/discussion/456118#2589018\" target=\"_blank\">https://www.kaggle.com/competitions/blood-vessel-segmentation/discussion/456118#2589018</a></p>",
      "votes": null,
      "replies": [
        {
          "id": 2595778,
          "author_name": "hengck23",
          "author_url": "",
          "post_date": "01/10/2024 16:33:19",
          "content": "<p>you can also do a 3d floodfill using seed = junction of missing vessel</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 2595791,
          "author_name": "limitz",
          "author_url": "",
          "post_date": "01/10/2024 16:42:54",
          "content": "<p>Ah I see 👍 I feel a bit weird adapting the model to ignore the smallest vessel detections.. just for a better score.. but maybe that's the way forward…</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "2595440": "Quite often my model sees, what I think are, tiny vessels that are missing from the annotations (in kidney_1_dense, kidney_1_voi or kidney_3). Sometimes only the start of a vessel is annotated and then in the prediction it seems to continue, which makes me believe I'm not mistakenly detecting veins or something. But maybe I'm wrong... there are a few possible explanations:\n\n* The annotations are missing\n* I'm detecting veins instead of vessels\n* The model is hallucinating\n\nAnyone have similar experience or an explanation maybe?\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F3086083%2Fcbd8afa091f702edca066ccb01e217ad%2Fkidney_more.png?generation=1704889801023801&alt=media)",
    "2595776": "i already mention the missing annotation:\nhttps://www.kaggle.com/competitions/blood-vessel-segmentation/discussion/456118#2583079\nhttps://www.kaggle.com/competitions/blood-vessel-segmentation/discussion/456118#2589018",
    "2595778": "you can also do a 3d floodfill using seed = junction of missing vessel",
    "2595791": "Ah I see 👍 I feel a bit weird adapting the model to ignore the smallest vessel detections.. just for a better score.. but maybe that's the way forward..."
  },
  "source": "meta"
}