{
  "id": 461556,
  "title": "BE AWARE of the misleading info from the competition description on TMA thumbnail",
  "url": "/competitions/UBC-OCEAN/discussion/461556",
  "author_name": "Yoobin",
  "post_date": "2023-12-15T04:36:25.624000",
  "votes": 14,
  "comment_count": 1,
  "views": 0,
  "content": "<p>😵‍💫 I was trying to handle WSI and TMA in different ways and I found during test, no matter how I handle the TMA images the results are the same -&gt; I thought maybe all images are WSI in the public test data. </p>\n<p>😲 Then I realized that it is possible to assumption from the organizer's description \"image-id in thumbnail folder is all WSI\" is wrong. So I did some testing:</p>\n<p><strong>Experiment 1</strong> <br>\nGiving all image-id in thumbnail EC label and giving the rest wrong label results in 0.16 - the same if you gives all images EC label -&gt; either the thumbnail images ==all images or there is no TMA in the data.</p>\n<p><strong>Experiment 2</strong> <br>\nGiving all image-id as EC if the max(width, height) of the original image &gt;5500px -&gt; score is 0.07 -&gt; this means only 7/16 images in the data have image size &gt;5500px and these likely are WSI and we then have 9/16 TMA in the data.</p>\n<p><strong>Conclusion</strong> <br>\nThere are TMAs in the test data and they are also in thumbnail folder! The info in the competition could be misleading. </p>\n<p>But of course this is a conclusion from my testing. Happy to discuss other possibilities!</p>",
  "messages": [
    {
      "id": 2562065,
      "postDate": "2023-12-15T04:36:25.623Z",
      "content": "<p>😵‍💫 I was trying to handle WSI and TMA in different ways and I found during test, no matter how I handle the TMA images the results are the same -&gt; I thought maybe all images are WSI in the public test data. </p>\n<p>😲 Then I realized that it is possible to assumption from the organizer's description \"image-id in thumbnail folder is all WSI\" is wrong. So I did some testing:</p>\n<p><strong>Experiment 1</strong> <br>\nGiving all image-id in thumbnail EC label and giving the rest wrong label results in 0.16 - the same if you gives all images EC label -&gt; either the thumbnail images ==all images or there is no TMA in the data.</p>\n<p><strong>Experiment 2</strong> <br>\nGiving all image-id as EC if the max(width, height) of the original image &gt;5500px -&gt; score is 0.07 -&gt; this means only 7/16 images in the data have image size &gt;5500px and these likely are WSI and we then have 9/16 TMA in the data.</p>\n<p><strong>Conclusion</strong> <br>\nThere are TMAs in the test data and they are also in thumbnail folder! The info in the competition could be misleading. </p>\n<p>But of course this is a conclusion from my testing. Happy to discuss other possibilities!</p>",
      "rawMarkdown": "😵‍💫 I was trying to handle WSI and TMA in different ways and I found during test, no matter how I handle the TMA images the results are the same -> I thought maybe all images are WSI in the public test data. \n\n😲 Then I realized that it is possible to assumption from the organizer's description \"image-id in thumbnail folder is all WSI\" is wrong. So I did some testing:\n\n**Experiment 1** \nGiving all image-id in thumbnail EC label and giving the rest wrong label results in 0.16 - the same if you gives all images EC label -> either the thumbnail images ==all images or there is no TMA in the data.\n\n**Experiment 2** \nGiving all image-id as EC if the max(width, height) of the original image >5500px -> score is 0.07 -> this means only 7/16 images in the data have image size >5500px and these likely are WSI and we then have 9/16 TMA in the data.\n\n **Conclusion** \nThere are TMAs in the test data and they are also in thumbnail folder! The info in the competition could be misleading. \n\nBut of course this is a conclusion from my testing. Happy to discuss other possibilities!\n",
      "votes": 13
    },
    {
      "id": 2562659,
      "postDate": "2023-12-15T15:36:52.960Z",
      "content": "<p>Thank you for sharing your great insights! <br>\nI was planning to classify wsi and tma image by whether they are included in the thumbnail folder, but I realized that might have been crucial mistake… </p>",
      "rawMarkdown": "Thank you for sharing your great insights! \nI was planning to classify wsi and tma image by whether they are included in the thumbnail folder, but I realized that might have been crucial mistake... ",
      "votes": 3
    }
  ],
  "comments": [
    {
      "id": 2562659,
      "author_name": "Lee Jun Hyuk",
      "author_url": "",
      "post_date": "2023-12-15T15:36:52.960000",
      "content": "<p>Thank you for sharing your great insights! <br>\nI was planning to classify wsi and tma image by whether they are included in the thumbnail folder, but I realized that might have been crucial mistake… </p>",
      "votes": 3,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2562065": "😵‍💫 I was trying to handle WSI and TMA in different ways and I found during test, no matter how I handle the TMA images the results are the same -> I thought maybe all images are WSI in the public test data. \n\n😲 Then I realized that it is possible to assumption from the organizer's description \"image-id in thumbnail folder is all WSI\" is wrong. So I did some testing:\n\n**Experiment 1** \nGiving all image-id in thumbnail EC label and giving the rest wrong label results in 0.16 - the same if you gives all images EC label -> either the thumbnail images ==all images or there is no TMA in the data.\n\n**Experiment 2** \nGiving all image-id as EC if the max(width, height) of the original image >5500px -> score is 0.07 -> this means only 7/16 images in the data have image size >5500px and these likely are WSI and we then have 9/16 TMA in the data.\n\n **Conclusion** \nThere are TMAs in the test data and they are also in thumbnail folder! The info in the competition could be misleading. \n\nBut of course this is a conclusion from my testing. Happy to discuss other possibilities!\n",
    "2562659": "Thank you for sharing your great insights! \nI was planning to classify wsi and tma image by whether they are included in the thumbnail folder, but I realized that might have been crucial mistake... "
  }
}