{
  "id": 433885,
  "title": "Head or base?",
  "url": "/competitions/rsna-2023-abdominal-trauma-detection/discussion/433885",
  "author_name": "Gyula Maloveczky4",
  "post_date": "2023-08-23T09:02:18.187000",
  "votes": 1,
  "comment_count": 2,
  "views": 0,
  "content": "<p>I am almost completely new to computer vision. I have recently learned that many times people use base models that are already carefully developed and only create the head model themselves. Is this an advisable strategy on this competition? If so, what is a good base model to use for this dataset?</p>",
  "messages": [
    {
      "id": 2405387,
      "postDate": "2023-08-23T20:59:22.300Z",
      "content": "<p>Use of base models very common in kaggle competitions, with lots of different versions of heads.  Many of the kaggle data sets are a little too small to train a large model from scratch.</p>\n<p>Unet models seem very common for CT and x-ray competitions.  Not sure I have seen many base 3D models which seems like a decent approach for the weak labels in this competition.  There are couple different base models used in the shared notebooks.</p>",
      "rawMarkdown": "Use of base models very common in kaggle competitions, with lots of different versions of heads.  Many of the kaggle data sets are a little too small to train a large model from scratch.\n\nUnet models seem very common for CT and x-ray competitions.  Not sure I have seen many base 3D models which seems like a decent approach for the weak labels in this competition.  There are couple different base models used in the shared notebooks.",
      "votes": 1,
      "replies": [
        {
          "id": 2405900,
          "postDate": "2023-08-24T06:37:35.183Z",
          "rawMarkdown": "",
          "isDeleted": true
        }
      ]
    },
    {
      "id": 2404441,
      "postDate": "2023-08-23T09:02:18.187Z",
      "content": "<p>I am almost completely new to computer vision. I have recently learned that many times people use base models that are already carefully developed and only create the head model themselves. Is this an advisable strategy on this competition? If so, what is a good base model to use for this dataset?</p>",
      "rawMarkdown": "I am almost completely new to computer vision. I have recently learned that many times people use base models that are already carefully developed and only create the head model themselves. Is this an advisable strategy on this competition? If so, what is a good base model to use for this dataset?",
      "votes": 1
    }
  ],
  "comments": [
    {
      "id": 2405387,
      "author_name": "PC Jimmmy",
      "author_url": "",
      "post_date": "2023-08-23T20:59:22.300000",
      "content": "<p>Use of base models very common in kaggle competitions, with lots of different versions of heads.  Many of the kaggle data sets are a little too small to train a large model from scratch.</p>\n<p>Unet models seem very common for CT and x-ray competitions.  Not sure I have seen many base 3D models which seems like a decent approach for the weak labels in this competition.  There are couple different base models used in the shared notebooks.</p>",
      "votes": 1,
      "replies": [
        {
          "id": 2405900,
          "author_name": "",
          "author_url": "",
          "post_date": "2023-08-24T06:37:35.183000",
          "content": "",
          "votes": 0,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "2405387": "Use of base models very common in kaggle competitions, with lots of different versions of heads.  Many of the kaggle data sets are a little too small to train a large model from scratch.\n\nUnet models seem very common for CT and x-ray competitions.  Not sure I have seen many base 3D models which seems like a decent approach for the weak labels in this competition.  There are couple different base models used in the shared notebooks.",
    "2404441": "I am almost completely new to computer vision. I have recently learned that many times people use base models that are already carefully developed and only create the head model themselves. Is this an advisable strategy on this competition? If so, what is a good base model to use for this dataset?"
  }
}