{
  "id": 372278,
  "title": "Feasibility of transfer learning",
  "url": "/competitions/rsna-breast-cancer-detection/discussion/372278",
  "author_name": "WangHawsi",
  "post_date": "2022-12-15T08:10:34.767000",
  "votes": 0,
  "comment_count": 2,
  "views": 0,
  "content": "<p>I have a question. The data set of RSNA may be special. In this case, is it not appropriate to adopt the strategy of transfer learning?</p>",
  "messages": [
    {
      "id": 2066083,
      "postDate": "2022-12-15T11:56:07.663Z",
      "content": "<p>I'm no expert, but I think the first layers of transfer learning models usually have information about super generic features like lines, curves etc. This is something you may leverage to most CV problems.</p>",
      "rawMarkdown": "I'm no expert, but I think the first layers of transfer learning models usually have information about super generic features like lines, curves etc. This is something you may leverage to most CV problems.",
      "votes": 1,
      "replies": [
        {
          "id": 2067147,
          "postDate": "2022-12-16T12:56:47.540Z",
          "content": "<p>Thats correct. Depending on how much similarities there are you can freeze more/less layers.</p>",
          "rawMarkdown": "Thats correct. Depending on how much similarities there are you can freeze more/less layers."
        }
      ]
    },
    {
      "id": 2065932,
      "postDate": "2022-12-15T08:10:34.767Z",
      "content": "<p>I have a question. The data set of RSNA may be special. In this case, is it not appropriate to adopt the strategy of transfer learning?</p>",
      "rawMarkdown": "I have a question. The data set of RSNA may be special. In this case, is it not appropriate to adopt the strategy of transfer learning?"
    }
  ],
  "comments": [
    {
      "id": 2066083,
      "author_name": "ManuelKJG",
      "author_url": "",
      "post_date": "2022-12-15T11:56:07.663000",
      "content": "<p>I'm no expert, but I think the first layers of transfer learning models usually have information about super generic features like lines, curves etc. This is something you may leverage to most CV problems.</p>",
      "votes": 1,
      "replies": [
        {
          "id": 2067147,
          "author_name": "Ali Abdin",
          "author_url": "",
          "post_date": "2022-12-16T12:56:47.540000",
          "content": "<p>Thats correct. Depending on how much similarities there are you can freeze more/less layers.</p>",
          "votes": 0,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "2066083": "I'm no expert, but I think the first layers of transfer learning models usually have information about super generic features like lines, curves etc. This is something you may leverage to most CV problems.",
    "2065932": "I have a question. The data set of RSNA may be special. In this case, is it not appropriate to adopt the strategy of transfer learning?"
  }
}