{
  "id": 343553,
  "title": "Would MIL really work for this data?",
  "url": "/competitions/mayo-clinic-strip-ai/discussion/343553",
  "author_name": "hjunlee941",
  "post_date": "2022-08-11T18:45:24.188000",
  "votes": 2,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Hi folks,</p>\n<p>I've been trying to approach this challenge using MIL (Multi Instance Learning), thinking that a lot of space of original images are background while detail patterns of cells matter for classification. I've fed as many instances (e.g., patches) as possible to maximize view of each tissue, however, this approach somehow does not work very well so far. Has anyone tried to build MIL pipeline to solve this problem? </p>",
  "messages": [
    {
      "id": 1894858,
      "postDate": "2022-08-11T18:45:24.190Z",
      "content": "<p>Hi folks,</p>\n<p>I've been trying to approach this challenge using MIL (Multi Instance Learning), thinking that a lot of space of original images are background while detail patterns of cells matter for classification. I've fed as many instances (e.g., patches) as possible to maximize view of each tissue, however, this approach somehow does not work very well so far. Has anyone tried to build MIL pipeline to solve this problem? </p>",
      "rawMarkdown": "Hi folks,\n\nI've been trying to approach this challenge using MIL (Multi Instance Learning), thinking that a lot of space of original images are background while detail patterns of cells matter for classification. I've fed as many instances (e.g., patches) as possible to maximize view of each tissue, however, this approach somehow does not work very well so far. Has anyone tried to build MIL pipeline to solve this problem? ",
      "votes": 2
    },
    {
      "id": 1895679,
      "postDate": "2022-08-12T10:08:46.883Z",
      "rawMarkdown": "",
      "isDeleted": true
    }
  ],
  "comments": [
    {
      "id": 1895679,
      "author_name": "",
      "author_url": "",
      "post_date": "2022-08-12T10:08:46.883000",
      "content": "",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1894858": "Hi folks,\n\nI've been trying to approach this challenge using MIL (Multi Instance Learning), thinking that a lot of space of original images are background while detail patterns of cells matter for classification. I've fed as many instances (e.g., patches) as possible to maximize view of each tissue, however, this approach somehow does not work very well so far. Has anyone tried to build MIL pipeline to solve this problem? ",
    "1895679": ""
  }
}