{
  "id": 68944,
  "title": "Question about original and small images",
  "url": "/competitions/human-protein-atlas-image-classification/discussion/68944",
  "author_name": "",
  "post_date": "2018-10-18T17:41:35.194274Z",
  "votes": 3,
  "comment_count": 4,
  "views": 0,
  "content": "<p>Hi all, </p>\n\n<p>I have only a few questions regarding the original and small data. I haven't downloaded the big data set - and I have read several discussion posts and not got an answer (but maybe I have over read something, then please excuse me). Here are my questions: \n1) Are the images in the small data set just resized from the big data set or are they corps? \n2) If they are crops - are they center crops or randomly chosen crops?\n3) Are all images of the same scale, i.e. if a nucleus is 50 pixel wide on one image and another one is 100 pixels wide - does that mean that the second is in reality also twice as wide or is that conclusion not correct, because the images come from different different devices with different magnifications?\nHope that makes sense, as English isn't my native language - and thanks for the answers, \nbest regards, \nWolfgang </p>",
  "messages": [
    {
      "id": "406125",
      "postDate": "10/18/2018 17:41:35",
      "content": "<p>Hi all, </p>\n\n<p>I have only a few questions regarding the original and small data. I haven't downloaded the big data set - and I have read several discussion posts and not got an answer (but maybe I have over read something, then please excuse me). Here are my questions: \n1) Are the images in the small data set just resized from the big data set or are they corps? \n2) If they are crops - are they center crops or randomly chosen crops?\n3) Are all images of the same scale, i.e. if a nucleus is 50 pixel wide on one image and another one is 100 pixels wide - does that mean that the second is in reality also twice as wide or is that conclusion not correct, because the images come from different different devices with different magnifications?\nHope that makes sense, as English isn't my native language - and thanks for the answers, \nbest regards, \nWolfgang </p>",
      "rawMarkdown": "Hi all, \n\nI have only a few questions regarding the original and small data. I haven't downloaded the big data set - and I have read several discussion posts and not got an answer (but maybe I have over read something, then please excuse me). Here are my questions: \n1) Are the images in the small data set just resized from the big data set or are they corps? \n2) If they are crops - are they center crops or randomly chosen crops?\n3) Are all images of the same scale, i.e. if a nucleus is 50 pixel wide on one image and another one is 100 pixels wide - does that mean that the second is in reality also twice as wide or is that conclusion not correct, because the images come from different different devices with different magnifications?\nHope that makes sense, as English isn't my native language - and thanks for the answers, \nbest regards, \nWolfgang",
      "votes": null
    },
    {
      "id": "406409",
      "postDate": "10/19/2018 07:42:38",
      "content": "<p>Hello!</p>\n\n<p>For your first and second question: The PNGs are resized versions of the original TIFFs. They are not crops.</p>\n\n<p>For your third: The images are taken using the same kind of device with the same magnification so they should scale as you suggest. This assumption is not true for all image attributes as some settings are different between experiments (in particular laser and detector settings).</p>\n\n<p>Edit: I forgot about the different image sizes of the original TIFFs. I apologize! The images that are originally 3072x3072 will have a different pixel size than the ones that are 2048x2048.</p>",
      "rawMarkdown": "Hello!\n\nFor your first and second question: The PNGs are resized versions of the original TIFFs. They are not crops.\n\nFor your third: The images are taken using the same kind of device with the same magnification so they should scale as you suggest. This assumption is not true for all image attributes as some settings are different between experiments (in particular laser and detector settings).\n\nEdit: I forgot about the different image sizes of the original TIFFs. I apologize! The images that are originally 3072x3072 will have a different pixel size than the ones that are 2048x2048.",
      "votes": null
    },
    {
      "id": "406428",
      "postDate": "10/19/2018 08:22:26",
      "content": "<p>Hello back - and thank you very much!!!\nBest regards, \nWolfgang </p>",
      "rawMarkdown": "Hello back - and thank you very much!!!\nBest regards, \nWolfgang",
      "votes": null
    },
    {
      "id": "407950",
      "postDate": "10/22/2018 03:38:22",
      "content": "<blockquote>\n  <p>This assumption is not true for all image attributes as some settings are different between experiments (in particular laser and detector settings)</p>\n</blockquote>\n\n<ul>\n<li>Sorry for bothering you, but I am still confused. </li>\n<li>What does \"This\nassumption\" refer to? - the assumption of using the same device with\nthe same magnification? </li>\n<li>What does \"image attributes\" refer to? -\nother channels besides the green ones?</li>\n</ul>\n\n<p>How I interpret this</p>\n\n<blockquote>\n  <p>Alternatively, if you wish to work with full size original images (a mix of 2048x2048 and 3072x3072 TIFF files)</p>\n</blockquote>\n\n<p>Is that the original images could be 2048x2048 or 3072x3072. But the resized ones are the same. There was no cropping involved in any step of creating the dataset.</p>",
      "rawMarkdown": "&gt; This assumption is not true for all image attributes as some settings are different between experiments (in particular laser and detector settings)\n\n - Sorry for bothering you, but I am still confused. \n - What does \"This\n   assumption\" refer to? - the assumption of using the same device with\n   the same magnification? \n - What does \"image attributes\" refer to? -\n   other channels besides the green ones?\n\nHow I interpret this\n\n &gt; Alternatively, if you wish to work with full size original images (a mix of 2048x2048 and 3072x3072 TIFF files)\n\nIs that the original images could be 2048x2048 or 3072x3072. But the resized ones are the same. There was no cropping involved in any step of creating the dataset.",
      "votes": null
    },
    {
      "id": "408012",
      "postDate": "10/22/2018 06:56:08",
      "content": "<p>With image attributes I meant numerical features of an image such as the pixel values. For example, since the laser/detector settings may differ from experiment to experiment you can not assume that a pixel value of X in one experiment and 2X in another means that the second experiment has exactly twice the amount of flourescence. </p>\n\n<p>Neither 2048x2048 or 3072x3072 were cropped beforehand. This means they will have slightly different pixel sizes in the resized version. I unfortunately forgot about the different sized originals in my first answer, which I apologize for.</p>",
      "rawMarkdown": "With image attributes I meant numerical features of an image such as the pixel values. For example, since the laser/detector settings may differ from experiment to experiment you can not assume that a pixel value of X in one experiment and 2X in another means that the second experiment has exactly twice the amount of flourescence. \n\nNeither 2048x2048 or 3072x3072 were cropped beforehand. This means they will have slightly different pixel sizes in the resized version. I unfortunately forgot about the different sized originals in my first answer, which I apologize for.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 406409,
      "author_name": "cwinsnes",
      "author_url": "",
      "post_date": "10/19/2018 07:42:38",
      "content": "<p>Hello!</p>\n\n<p>For your first and second question: The PNGs are resized versions of the original TIFFs. They are not crops.</p>\n\n<p>For your third: The images are taken using the same kind of device with the same magnification so they should scale as you suggest. This assumption is not true for all image attributes as some settings are different between experiments (in particular laser and detector settings).</p>\n\n<p>Edit: I forgot about the different image sizes of the original TIFFs. I apologize! The images that are originally 3072x3072 will have a different pixel size than the ones that are 2048x2048.</p>",
      "votes": null,
      "replies": [
        {
          "id": 406428,
          "author_name": "wreuter",
          "author_url": "",
          "post_date": "10/19/2018 08:22:26",
          "content": "<p>Hello back - and thank you very much!!!\nBest regards, \nWolfgang </p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 407950,
          "author_name": "kokecacao",
          "author_url": "",
          "post_date": "10/22/2018 03:38:22",
          "content": "<blockquote>\n  <p>This assumption is not true for all image attributes as some settings are different between experiments (in particular laser and detector settings)</p>\n</blockquote>\n\n<ul>\n<li>Sorry for bothering you, but I am still confused. </li>\n<li>What does \"This\nassumption\" refer to? - the assumption of using the same device with\nthe same magnification? </li>\n<li>What does \"image attributes\" refer to? -\nother channels besides the green ones?</li>\n</ul>\n\n<p>How I interpret this</p>\n\n<blockquote>\n  <p>Alternatively, if you wish to work with full size original images (a mix of 2048x2048 and 3072x3072 TIFF files)</p>\n</blockquote>\n\n<p>Is that the original images could be 2048x2048 or 3072x3072. But the resized ones are the same. There was no cropping involved in any step of creating the dataset.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 408012,
          "author_name": "cwinsnes",
          "author_url": "",
          "post_date": "10/22/2018 06:56:08",
          "content": "<p>With image attributes I meant numerical features of an image such as the pixel values. For example, since the laser/detector settings may differ from experiment to experiment you can not assume that a pixel value of X in one experiment and 2X in another means that the second experiment has exactly twice the amount of flourescence. </p>\n\n<p>Neither 2048x2048 or 3072x3072 were cropped beforehand. This means they will have slightly different pixel sizes in the resized version. I unfortunately forgot about the different sized originals in my first answer, which I apologize for.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "406125": "Hi all, \n\nI have only a few questions regarding the original and small data. I haven't downloaded the big data set - and I have read several discussion posts and not got an answer (but maybe I have over read something, then please excuse me). Here are my questions: \n1) Are the images in the small data set just resized from the big data set or are they corps? \n2) If they are crops - are they center crops or randomly chosen crops?\n3) Are all images of the same scale, i.e. if a nucleus is 50 pixel wide on one image and another one is 100 pixels wide - does that mean that the second is in reality also twice as wide or is that conclusion not correct, because the images come from different different devices with different magnifications?\nHope that makes sense, as English isn't my native language - and thanks for the answers, \nbest regards, \nWolfgang",
    "406409": "Hello!\n\nFor your first and second question: The PNGs are resized versions of the original TIFFs. They are not crops.\n\nFor your third: The images are taken using the same kind of device with the same magnification so they should scale as you suggest. This assumption is not true for all image attributes as some settings are different between experiments (in particular laser and detector settings).\n\nEdit: I forgot about the different image sizes of the original TIFFs. I apologize! The images that are originally 3072x3072 will have a different pixel size than the ones that are 2048x2048.",
    "406428": "Hello back - and thank you very much!!!\nBest regards, \nWolfgang",
    "407950": "&gt; This assumption is not true for all image attributes as some settings are different between experiments (in particular laser and detector settings)\n\n - Sorry for bothering you, but I am still confused. \n - What does \"This\n   assumption\" refer to? - the assumption of using the same device with\n   the same magnification? \n - What does \"image attributes\" refer to? -\n   other channels besides the green ones?\n\nHow I interpret this\n\n &gt; Alternatively, if you wish to work with full size original images (a mix of 2048x2048 and 3072x3072 TIFF files)\n\nIs that the original images could be 2048x2048 or 3072x3072. But the resized ones are the same. There was no cropping involved in any step of creating the dataset.",
    "408012": "With image attributes I meant numerical features of an image such as the pixel values. For example, since the laser/detector settings may differ from experiment to experiment you can not assume that a pixel value of X in one experiment and 2X in another means that the second experiment has exactly twice the amount of flourescence. \n\nNeither 2048x2048 or 3072x3072 were cropped beforehand. This means they will have slightly different pixel sizes in the resized version. I unfortunately forgot about the different sized originals in my first answer, which I apologize for."
  },
  "source": "meta"
}