{
  "id": 666322,
  "title": "2.5D segmentation solutions",
  "url": "/competitions/vesuvius-challenge-surface-detection/discussion/666322",
  "author_name": "",
  "post_date": "2026-01-06T12:22:06.872919200Z",
  "votes": 7,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Hi,</p>\n<p>Been trying to make 2.5D work given its success in past 3D segmentation competitions. No luck so far… stuck on about 0.47 on the public LB. Anyone got any sucess  with such setup? </p>",
  "messages": [
    {
      "id": "3387121",
      "postDate": "01/06/2026 12:22:06",
      "content": "<p>Hi,</p>\n<p>Been trying to make 2.5D work given its success in past 3D segmentation competitions. No luck so far… stuck on about 0.47 on the public LB. Anyone got any sucess  with such setup? </p>",
      "rawMarkdown": "Hi,\n\nBeen trying to make 2.5D work given its success in past 3D segmentation competitions. No luck so far... stuck on about 0.47 on the public LB. Anyone got any sucess  with such setup?",
      "votes": null
    },
    {
      "id": "3393705",
      "postDate": "01/19/2026 16:19:07",
      "content": "<p>Do you use a pretrained model or train from scratch 🧐</p>",
      "rawMarkdown": "Do you use a pretrained model or train from scratch 🧐",
      "votes": null
    },
    {
      "id": "3393716",
      "postDate": "01/19/2026 16:30:20",
      "content": "<p>Yep. Convnext tiny. Got to 0.525 LB meanwhile. Still far from the top. </p>",
      "rawMarkdown": "Yep. Convnext tiny. Got to 0.525 LB meanwhile. Still far from the top.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3393705,
      "author_name": "nguyncdngs",
      "author_url": "",
      "post_date": "01/19/2026 16:19:07",
      "content": "<p>Do you use a pretrained model or train from scratch 🧐</p>",
      "votes": null,
      "replies": [
        {
          "id": 3393716,
          "author_name": "roysegalz",
          "author_url": "",
          "post_date": "01/19/2026 16:30:20",
          "content": "<p>Yep. Convnext tiny. Got to 0.525 LB meanwhile. Still far from the top. </p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "3387121": "Hi,\n\nBeen trying to make 2.5D work given its success in past 3D segmentation competitions. No luck so far... stuck on about 0.47 on the public LB. Anyone got any sucess  with such setup?",
    "3393705": "Do you use a pretrained model or train from scratch 🧐",
    "3393716": "Yep. Convnext tiny. Got to 0.525 LB meanwhile. Still far from the top."
  },
  "source": "meta"
}