{
  "id": 453308,
  "title": "Tiles/thumbnails",
  "url": "/competitions/UBC-OCEAN/discussion/453308",
  "author_name": "Andreu Arderiu",
  "post_date": "2023-11-05T20:36:03.274000",
  "votes": 1,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Did anyone try the same model using thumbnails and tiles? What is the order of magnitude of accuracy differences you get? I am trying both approaches and In some cases I have better scores using just the thumbnails, so I am wondering I might have something wrong in my code… thanks!</p>",
  "messages": [
    {
      "id": 2513884,
      "postDate": "2023-11-05T20:36:03.273Z",
      "content": "<p>Did anyone try the same model using thumbnails and tiles? What is the order of magnitude of accuracy differences you get? I am trying both approaches and In some cases I have better scores using just the thumbnails, so I am wondering I might have something wrong in my code… thanks!</p>",
      "rawMarkdown": "Did anyone try the same model using thumbnails and tiles? What is the order of magnitude of accuracy differences you get? I am trying both approaches and In some cases I have better scores using just the thumbnails, so I am wondering I might have something wrong in my code... thanks!",
      "votes": 1
    },
    {
      "id": 2514711,
      "postDate": "2023-11-06T13:23:20.230Z",
      "content": "<p>I am using tiles.</p>",
      "rawMarkdown": "I am using tiles."
    },
    {
      "id": 2513969,
      "postDate": "2023-11-05T22:25:35.283Z",
      "content": "<p>I did not try it myself here. Though I can say from the experience from real life - thumbnails contain useful information about positioning of cancer (and other) cells on a WSI, which cannot be obtained if you look at tiles only. So, you may think about additional postprocessing of labels/information obtained from tiles classification. Or maybe using both thumbnails and tiles? 😉</p>",
      "rawMarkdown": "I did not try it myself here. Though I can say from the experience from real life - thumbnails contain useful information about positioning of cancer (and other) cells on a WSI, which cannot be obtained if you look at tiles only. So, you may think about additional postprocessing of labels/information obtained from tiles classification. Or maybe using both thumbnails and tiles? 😉"
    }
  ],
  "comments": [
    {
      "id": 2514711,
      "author_name": "Quan Vu",
      "author_url": "",
      "post_date": "2023-11-06T13:23:20.230000",
      "content": "<p>I am using tiles.</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 2513969,
      "author_name": "Prosperous Silver Alpaca",
      "author_url": "",
      "post_date": "2023-11-05T22:25:35.283000",
      "content": "<p>I did not try it myself here. Though I can say from the experience from real life - thumbnails contain useful information about positioning of cancer (and other) cells on a WSI, which cannot be obtained if you look at tiles only. So, you may think about additional postprocessing of labels/information obtained from tiles classification. Or maybe using both thumbnails and tiles? 😉</p>",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2513884": "Did anyone try the same model using thumbnails and tiles? What is the order of magnitude of accuracy differences you get? I am trying both approaches and In some cases I have better scores using just the thumbnails, so I am wondering I might have something wrong in my code... thanks!",
    "2514711": "I am using tiles.",
    "2513969": "I did not try it myself here. Though I can say from the experience from real life - thumbnails contain useful information about positioning of cancer (and other) cells on a WSI, which cannot be obtained if you look at tiles only. So, you may think about additional postprocessing of labels/information obtained from tiles classification. Or maybe using both thumbnails and tiles? 😉"
  }
}