{
  "id": 219099,
  "title": "What do height and width field in dataset mean?",
  "url": "/competitions/vinbigdata-chest-xray-abnormalities-detection/discussion/219099",
  "author_name": "skj",
  "post_date": "2021-02-13T10:05:56.876000",
  "votes": 1,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Official description of the dataset only described other 8 data fields.<br>\nDo they mean original image size?<br>\nThen where do the 1024<em>1024,512</em>512 AND 256*256 version of the dataset come from ?</p>",
  "messages": [
    {
      "id": 1198787,
      "postDate": "2021-02-13T10:05:56.877Z",
      "content": "<p>Official description of the dataset only described other 8 data fields.<br>\nDo they mean original image size?<br>\nThen where do the 1024<em>1024,512</em>512 AND 256*256 version of the dataset come from ?</p>",
      "rawMarkdown": "Official description of the dataset only described other 8 data fields.\nDo they mean original image size?\nThen where do the 1024*1024,512*512 AND 256*256 version of the dataset come from ?",
      "votes": 1
    },
    {
      "id": 1199023,
      "postDate": "2021-02-13T13:47:50.380Z",
      "content": "<p>Thanks for the clear and excellent answer!</p>",
      "rawMarkdown": "Thanks for the clear and excellent answer!"
    },
    {
      "id": 1199001,
      "postDate": "2021-02-13T13:36:00.537Z",
      "content": "<p>These are indeed the size of the image in the .dicom file (it's an array with a single grayscale \"color\"-channel, and the indicated height &amp; width). Resized versions were created by other Kaggle users (rather than the competition hosts). Thus, do consider whether these may or may not have some issues (e.g. with how normalization of colors/brightness are treated, with stretching/compressing some dimensions of the image to fit into some standard square shape etc. - although these are likely not too much of an issue, because most of the images are of similar aspect ratio, so they mostly all get squashed to the same extent). There's various notebooks you can look at in the code tab that do this processing and you can decide whether you like the particular approach.</p>",
      "rawMarkdown": "These are indeed the size of the image in the .dicom file (it's an array with a single grayscale \"color\"-channel, and the indicated height & width). Resized versions were created by other Kaggle users (rather than the competition hosts). Thus, do consider whether these may or may not have some issues (e.g. with how normalization of colors/brightness are treated, with stretching/compressing some dimensions of the image to fit into some standard square shape etc. - although these are likely not too much of an issue, because most of the images are of similar aspect ratio, so they mostly all get squashed to the same extent). There's various notebooks you can look at in the code tab that do this processing and you can decide whether you like the particular approach."
    }
  ],
  "comments": [
    {
      "id": 1199023,
      "author_name": "skj",
      "author_url": "",
      "post_date": "2021-02-13T13:47:50.380000",
      "content": "<p>Thanks for the clear and excellent answer!</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 1199001,
      "author_name": "Björn",
      "author_url": "",
      "post_date": "2021-02-13T13:36:00.537000",
      "content": "<p>These are indeed the size of the image in the .dicom file (it's an array with a single grayscale \"color\"-channel, and the indicated height &amp; width). Resized versions were created by other Kaggle users (rather than the competition hosts). Thus, do consider whether these may or may not have some issues (e.g. with how normalization of colors/brightness are treated, with stretching/compressing some dimensions of the image to fit into some standard square shape etc. - although these are likely not too much of an issue, because most of the images are of similar aspect ratio, so they mostly all get squashed to the same extent). There's various notebooks you can look at in the code tab that do this processing and you can decide whether you like the particular approach.</p>",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1198787": "Official description of the dataset only described other 8 data fields.\nDo they mean original image size?\nThen where do the 1024*1024,512*512 AND 256*256 version of the dataset come from ?",
    "1199023": "Thanks for the clear and excellent answer!",
    "1199001": "These are indeed the size of the image in the .dicom file (it's an array with a single grayscale \"color\"-channel, and the indicated height & width). Resized versions were created by other Kaggle users (rather than the competition hosts). Thus, do consider whether these may or may not have some issues (e.g. with how normalization of colors/brightness are treated, with stretching/compressing some dimensions of the image to fit into some standard square shape etc. - although these are likely not too much of an issue, because most of the images are of similar aspect ratio, so they mostly all get squashed to the same extent). There's various notebooks you can look at in the code tab that do this processing and you can decide whether you like the particular approach."
  }
}