{
  "id": 209351,
  "title": "Is the magnification the same for all images?",
  "url": "/competitions/hubmap-kidney-segmentation/discussion/209351",
  "author_name": "",
  "post_date": "2021-01-07T10:17:18.868082500Z",
  "votes": 7,
  "comment_count": 7,
  "views": 0,
  "content": "<p>Is the magnification the same for all images? Would a glomerulus in image A be roughly the same number of pixels to image B (within say an order of magnitude)?</p>",
  "messages": [
    {
      "id": "1142332",
      "postDate": "01/07/2021 10:17:18",
      "content": "<p>Is the magnification the same for all images? Would a glomerulus in image A be roughly the same number of pixels to image B (within say an order of magnitude)?</p>",
      "rawMarkdown": "Is the magnification the same for all images? Would a glomerulus in image A be roughly the same number of pixels to image B (within say an order of magnitude)?",
      "votes": null
    },
    {
      "id": "1142507",
      "postDate": "01/07/2021 12:36:32",
      "content": "<p>I decided to stop being lazy and find out myself 😛</p>\n<table>\n<thead>\n<tr>\n<th>ID</th>\n<th>Num Masks</th>\n<th>Mean Pixels</th>\n<th>Median Pixels</th>\n<th>Min Pixels</th>\n<th>Max Pixels</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td>2f6ecfcdf</td>\n<td>162</td>\n<td>51758.6</td>\n<td>52142.0</td>\n<td>817</td>\n<td>119614</td>\n</tr>\n<tr>\n<td>aaa6a05cc</td>\n<td>95</td>\n<td>45435.5</td>\n<td>44226.0</td>\n<td>16972</td>\n<td>122733</td>\n</tr>\n<tr>\n<td>cb2d976f4</td>\n<td>316</td>\n<td>80323.6</td>\n<td>78424.5</td>\n<td>1061</td>\n<td>266687</td>\n</tr>\n<tr>\n<td>0486052bb</td>\n<td>125</td>\n<td>76997.8</td>\n<td>79203.0</td>\n<td>422</td>\n<td>179377</td>\n</tr>\n<tr>\n<td>e79de561c</td>\n<td>198</td>\n<td>85347.6</td>\n<td>83063.5</td>\n<td>23480</td>\n<td>240237</td>\n</tr>\n<tr>\n<td>095bf7a1f</td>\n<td>342</td>\n<td>103998.6</td>\n<td>105365.5</td>\n<td>14291</td>\n<td>277603</td>\n</tr>\n<tr>\n<td>54f2eec69</td>\n<td>140</td>\n<td>104292.8</td>\n<td>103451.5</td>\n<td>24516</td>\n<td>316734</td>\n</tr>\n<tr>\n<td>1e2425f28</td>\n<td>180</td>\n<td>106582.6</td>\n<td>108962.5</td>\n<td>11271</td>\n<td>220343</td>\n</tr>\n</tbody>\n</table>\n<p>Based on the mask sizes maybe you could group the images as follows:</p>\n<ol>\n<li>2f6ecfcdf  &amp; aaa6a05cc </li>\n<li>cb2d976f4, 0486052bb  &amp;  e79de561c </li>\n<li>095bf7a1f, 54f2eec69 &amp; 1e2425f28 </li>\n</ol>\n<p>Might have implications for CV?</p>",
      "rawMarkdown": "I decided to stop being lazy and find out myself 😛\n| ID | Num Masks | Mean Pixels | Median Pixels | Min Pixels | Max Pixels | \n| --- | --- | --- | --- | --- | --- |\n| 2f6ecfcdf |162 | 51758.6 | 52142.0 | 817 | 119614 |\n| aaa6a05cc | 95 | 45435.5 | 44226.0 | 16972 | 122733 |\n| cb2d976f4 | 316 | 80323.6 | 78424.5 | 1061 | 266687 |\n| 0486052bb | 125 | 76997.8 | 79203.0 | 422 | 179377 |\n| e79de561c | 198 | 85347.6 | 83063.5 | 23480 | 240237 |\n| 095bf7a1f | 342 | 103998.6 | 105365.5 | 14291 | 277603 |\n| 54f2eec69 | 140 | 104292.8 | 103451.5 | 24516 | 316734 |\n| 1e2425f28 | 180 | 106582.6 | 108962.5 | 11271 | 220343 |\n\nBased on the mask sizes maybe you could group the images as follows:\n1. 2f6ecfcdf  & aaa6a05cc \n2. cb2d976f4, 0486052bb  &  e79de561c \n3. 095bf7a1f, 54f2eec69 & 1e2425f28 \n\nMight have implications for CV?",
      "votes": null
    },
    {
      "id": "1142583",
      "postDate": "01/07/2021 13:35:42",
      "content": "<p>your finding is imopressive,thx! Maybe image scale rate can be set by your findings</p>",
      "rawMarkdown": "your finding is imopressive,thx! Maybe image scale rate can be set by your findings",
      "votes": null
    },
    {
      "id": "1144083",
      "postDate": "01/08/2021 08:17:20",
      "content": "<p>Hi <a href=\"https://www.kaggle.com/anjum48\" target=\"_blank\">@anjum48</a>,<br>\nIf I understand it correctly, the purpose of doing this is to normalize the image mask accordingly. Is that correct?</p>",
      "rawMarkdown": "Hi @anjum48,\nIf I understand it correctly, the purpose of doing this is to normalize the image mask accordingly. Is that correct?",
      "votes": null
    },
    {
      "id": "1144132",
      "postDate": "01/08/2021 08:52:53",
      "content": "<p>Yes, you could rescale the images if you thought they were significantly different sizes, although you wouldn't know how much to scale the test images.</p>\n<p>I was mainly curious to see if the mask sizes are consistent between images. If they were significantly different, it might affect how you would go about postprocessing masks</p>",
      "rawMarkdown": "Yes, you could rescale the images if you thought they were significantly different sizes, although you wouldn't know how much to scale the test images.\n\nI was mainly curious to see if the mask sizes are consistent between images. If they were significantly different, it might affect how you would go about postprocessing masks",
      "votes": null
    },
    {
      "id": "1147888",
      "postDate": "01/10/2021 19:35:04",
      "content": "<p>Hi <a href=\"https://www.kaggle.com/anjum48\" target=\"_blank\">@anjum48</a> without knowing the exact details of how the images were made it could be difficult to determine whether or not there are any differences in the images concerning scaling.</p>\n<p>A 'safer' method would/could be is to apply some limited scaling as part of your data augmentation pipeline..In that way you also train the model on slightly different scales. Goodluck!</p>",
      "rawMarkdown": "Hi @anjum48 without knowing the exact details of how the images were made it could be difficult to determine whether or not there are any differences in the images concerning scaling.\n\nA 'safer' method would/could be is to apply some limited scaling as part of your data augmentation pipeline..In that way you also train the model on slightly different scales. Goodluck!",
      "votes": null
    },
    {
      "id": "1152248",
      "postDate": "01/14/2021 01:20:39",
      "content": "<p>I analyzed the images with a senior pathologist, and some sclerotic glomeruli (the ones that almost don't have \"white parts\") were not labeled. Probably this is another factor that can underperform the models.</p>",
      "rawMarkdown": "I analyzed the images with a senior pathologist, and some sclerotic glomeruli (the ones that almost don't have \"white parts\") were not labeled. Probably this is another factor that can underperform the models.",
      "votes": null
    },
    {
      "id": "1180041",
      "postDate": "02/01/2021 03:58:52",
      "content": "<p><a href=\"https://www.kaggle.com/llh1818\" target=\"_blank\">@llh1818</a> </p>",
      "rawMarkdown": "llh1818",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1142507,
      "author_name": "anjum48",
      "author_url": "",
      "post_date": "01/07/2021 12:36:32",
      "content": "<p>I decided to stop being lazy and find out myself 😛</p>\n<table>\n<thead>\n<tr>\n<th>ID</th>\n<th>Num Masks</th>\n<th>Mean Pixels</th>\n<th>Median Pixels</th>\n<th>Min Pixels</th>\n<th>Max Pixels</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td>2f6ecfcdf</td>\n<td>162</td>\n<td>51758.6</td>\n<td>52142.0</td>\n<td>817</td>\n<td>119614</td>\n</tr>\n<tr>\n<td>aaa6a05cc</td>\n<td>95</td>\n<td>45435.5</td>\n<td>44226.0</td>\n<td>16972</td>\n<td>122733</td>\n</tr>\n<tr>\n<td>cb2d976f4</td>\n<td>316</td>\n<td>80323.6</td>\n<td>78424.5</td>\n<td>1061</td>\n<td>266687</td>\n</tr>\n<tr>\n<td>0486052bb</td>\n<td>125</td>\n<td>76997.8</td>\n<td>79203.0</td>\n<td>422</td>\n<td>179377</td>\n</tr>\n<tr>\n<td>e79de561c</td>\n<td>198</td>\n<td>85347.6</td>\n<td>83063.5</td>\n<td>23480</td>\n<td>240237</td>\n</tr>\n<tr>\n<td>095bf7a1f</td>\n<td>342</td>\n<td>103998.6</td>\n<td>105365.5</td>\n<td>14291</td>\n<td>277603</td>\n</tr>\n<tr>\n<td>54f2eec69</td>\n<td>140</td>\n<td>104292.8</td>\n<td>103451.5</td>\n<td>24516</td>\n<td>316734</td>\n</tr>\n<tr>\n<td>1e2425f28</td>\n<td>180</td>\n<td>106582.6</td>\n<td>108962.5</td>\n<td>11271</td>\n<td>220343</td>\n</tr>\n</tbody>\n</table>\n<p>Based on the mask sizes maybe you could group the images as follows:</p>\n<ol>\n<li>2f6ecfcdf  &amp; aaa6a05cc </li>\n<li>cb2d976f4, 0486052bb  &amp;  e79de561c </li>\n<li>095bf7a1f, 54f2eec69 &amp; 1e2425f28 </li>\n</ol>\n<p>Might have implications for CV?</p>",
      "votes": null,
      "replies": [
        {
          "id": 1142583,
          "author_name": "jiahecao",
          "author_url": "",
          "post_date": "01/07/2021 13:35:42",
          "content": "<p>your finding is imopressive,thx! Maybe image scale rate can be set by your findings</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1144083,
          "author_name": "prvnkmr",
          "author_url": "",
          "post_date": "01/08/2021 08:17:20",
          "content": "<p>Hi <a href=\"https://www.kaggle.com/anjum48\" target=\"_blank\">@anjum48</a>,<br>\nIf I understand it correctly, the purpose of doing this is to normalize the image mask accordingly. Is that correct?</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1144132,
          "author_name": "anjum48",
          "author_url": "",
          "post_date": "01/08/2021 08:52:53",
          "content": "<p>Yes, you could rescale the images if you thought they were significantly different sizes, although you wouldn't know how much to scale the test images.</p>\n<p>I was mainly curious to see if the mask sizes are consistent between images. If they were significantly different, it might affect how you would go about postprocessing masks</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1147888,
          "author_name": "rsmits",
          "author_url": "",
          "post_date": "01/10/2021 19:35:04",
          "content": "<p>Hi <a href=\"https://www.kaggle.com/anjum48\" target=\"_blank\">@anjum48</a> without knowing the exact details of how the images were made it could be difficult to determine whether or not there are any differences in the images concerning scaling.</p>\n<p>A 'safer' method would/could be is to apply some limited scaling as part of your data augmentation pipeline..In that way you also train the model on slightly different scales. Goodluck!</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1152248,
          "author_name": "joaorodriguez",
          "author_url": "",
          "post_date": "01/14/2021 01:20:39",
          "content": "<p>I analyzed the images with a senior pathologist, and some sclerotic glomeruli (the ones that almost don't have \"white parts\") were not labeled. Probably this is another factor that can underperform the models.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1180041,
      "author_name": "abiaozju",
      "author_url": "",
      "post_date": "02/01/2021 03:58:52",
      "content": "<p><a href=\"https://www.kaggle.com/llh1818\" target=\"_blank\">@llh1818</a> </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1142332": "Is the magnification the same for all images? Would a glomerulus in image A be roughly the same number of pixels to image B (within say an order of magnitude)?",
    "1142507": "I decided to stop being lazy and find out myself 😛\n| ID | Num Masks | Mean Pixels | Median Pixels | Min Pixels | Max Pixels | \n| --- | --- | --- | --- | --- | --- |\n| 2f6ecfcdf |162 | 51758.6 | 52142.0 | 817 | 119614 |\n| aaa6a05cc | 95 | 45435.5 | 44226.0 | 16972 | 122733 |\n| cb2d976f4 | 316 | 80323.6 | 78424.5 | 1061 | 266687 |\n| 0486052bb | 125 | 76997.8 | 79203.0 | 422 | 179377 |\n| e79de561c | 198 | 85347.6 | 83063.5 | 23480 | 240237 |\n| 095bf7a1f | 342 | 103998.6 | 105365.5 | 14291 | 277603 |\n| 54f2eec69 | 140 | 104292.8 | 103451.5 | 24516 | 316734 |\n| 1e2425f28 | 180 | 106582.6 | 108962.5 | 11271 | 220343 |\n\nBased on the mask sizes maybe you could group the images as follows:\n1. 2f6ecfcdf  & aaa6a05cc \n2. cb2d976f4, 0486052bb  &  e79de561c \n3. 095bf7a1f, 54f2eec69 & 1e2425f28 \n\nMight have implications for CV?",
    "1142583": "your finding is imopressive,thx! Maybe image scale rate can be set by your findings",
    "1144083": "Hi @anjum48,\nIf I understand it correctly, the purpose of doing this is to normalize the image mask accordingly. Is that correct?",
    "1144132": "Yes, you could rescale the images if you thought they were significantly different sizes, although you wouldn't know how much to scale the test images.\n\nI was mainly curious to see if the mask sizes are consistent between images. If they were significantly different, it might affect how you would go about postprocessing masks",
    "1147888": "Hi @anjum48 without knowing the exact details of how the images were made it could be difficult to determine whether or not there are any differences in the images concerning scaling.\n\nA 'safer' method would/could be is to apply some limited scaling as part of your data augmentation pipeline..In that way you also train the model on slightly different scales. Goodluck!",
    "1152248": "I analyzed the images with a senior pathologist, and some sclerotic glomeruli (the ones that almost don't have \"white parts\") were not labeled. Probably this is another factor that can underperform the models.",
    "1180041": "llh1818"
  },
  "source": "meta"
}