{
  "id": 466886,
  "title": "Does SAM-Med3D have the potential to be used in this competition?",
  "url": "/competitions/blood-vessel-segmentation/discussion/466886",
  "author_name": "",
  "post_date": "2024-01-10T09:30:18.180309800Z",
  "votes": null,
  "comment_count": 1,
  "views": 0,
  "content": "<ol>\n<li>SAM-Md3D is a universal open source segmentation model for 3D medical images.</li>\n<li>The methods I currently think of applying it include knowledge distillation and utilizing it to generate pseudo labels for additional data</li>\n<li>Paper link: <a href=\"https://arxiv.org/pdf/2310.15161.pdf\" target=\"_blank\">https://arxiv.org/pdf/2310.15161.pdf</a></li>\n</ol>",
  "messages": [
    {
      "id": "2595179",
      "postDate": "01/10/2024 09:30:18",
      "content": "<ol>\n<li>SAM-Md3D is a universal open source segmentation model for 3D medical images.</li>\n<li>The methods I currently think of applying it include knowledge distillation and utilizing it to generate pseudo labels for additional data</li>\n<li>Paper link: <a href=\"https://arxiv.org/pdf/2310.15161.pdf\" target=\"_blank\">https://arxiv.org/pdf/2310.15161.pdf</a></li>\n</ol>",
      "rawMarkdown": "1. SAM-Md3D is a universal open source segmentation model for 3D medical images.\n2. The methods I currently think of applying it include knowledge distillation and utilizing it to generate pseudo labels for additional data\n3. Paper link: https://arxiv.org/pdf/2310.15161.pdf",
      "votes": null
    },
    {
      "id": "2595378",
      "postDate": "01/10/2024 11:40:37",
      "content": "<p>Interesting read! … but I don't think so, mainly because they are transformer based, so I doubt they can handle the size of this competitions image while retaining the detail needed to get an acceptable score. The small blood vessels make up about 25% of the score, give or take, and these are only a few voxels in diameter, nothing like the data the SAM-Md3D was trained on (I assume).</p>",
      "rawMarkdown": "Interesting read! ... but I don't think so, mainly because they are transformer based, so I doubt they can handle the size of this competitions image while retaining the detail needed to get an acceptable score. The small blood vessels make up about 25% of the score, give or take, and these are only a few voxels in diameter, nothing like the data the SAM-Md3D was trained on (I assume).",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2595378,
      "author_name": "limitz",
      "author_url": "",
      "post_date": "01/10/2024 11:40:37",
      "content": "<p>Interesting read! … but I don't think so, mainly because they are transformer based, so I doubt they can handle the size of this competitions image while retaining the detail needed to get an acceptable score. The small blood vessels make up about 25% of the score, give or take, and these are only a few voxels in diameter, nothing like the data the SAM-Md3D was trained on (I assume).</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2595179": "1. SAM-Md3D is a universal open source segmentation model for 3D medical images.\n2. The methods I currently think of applying it include knowledge distillation and utilizing it to generate pseudo labels for additional data\n3. Paper link: https://arxiv.org/pdf/2310.15161.pdf",
    "2595378": "Interesting read! ... but I don't think so, mainly because they are transformer based, so I doubt they can handle the size of this competitions image while retaining the detail needed to get an acceptable score. The small blood vessels make up about 25% of the score, give or take, and these are only a few voxels in diameter, nothing like the data the SAM-Md3D was trained on (I assume)."
  },
  "source": "meta"
}