{
  "id": 93868,
  "title": "Image Sizes and Detection Boxes",
  "url": "/competitions/iwildcam-2019-fgvc6/discussion/93868",
  "author_name": "Josh Shapiro",
  "post_date": "2019-05-30T17:44:14.136000",
  "votes": 1,
  "comment_count": 3,
  "views": 0,
  "content": "<p>Is there a way to get the image height and widths of the original (full size) test and train images without downloading the full large dataset?</p>\n\n<p>I am trying to use the provided bounding boxes (<a href=\"https://github.com/visipedia/iwildcam_comp\">https://github.com/visipedia/iwildcam_comp</a> ), but they appear to be generated using the original, full size images, since they have pixel values like: <code>[2023, 1261, 2371, 1725]</code>. </p>\n\n<p>I would like to apply them to the smaller images, but need to know the width/height of the original images so I can scale down the bounding box. Is this data available anywhere other than by extracting it from the large images? train/test.csv list width as 1024, and the annotation files do not contain dimensions for the IDFG images.</p>",
  "messages": [
    {
      "id": 540080,
      "postDate": "2019-05-31T02:10:49.587Z",
      "content": "<p>You can get the original image sizes for the training set from the original annotations: <a href=\"https://lilablobssc.blob.core.windows.net/iwildcam2019/iWildCam_2019_Annotations.tar.gz\">https://lilablobssc.blob.core.windows.net/iwildcam2019/iWildCam_2019_Annotations.tar.gz</a></p>\n\n<p>Unfortunately that doesn't include the test set original sizes...</p>",
      "rawMarkdown": "You can get the original image sizes for the training set from the original annotations: https://lilablobssc.blob.core.windows.net/iwildcam2019/iWildCam_2019_Annotations.tar.gz\n\nUnfortunately that doesn't include the test set original sizes...",
      "votes": 1,
      "replies": [
        {
          "id": 540364,
          "postDate": "2019-05-31T12:06:16.543Z",
          "content": "<p>Thanks for the reply Dave. I ended up just re-running the detector on the smaller test images.</p>",
          "rawMarkdown": "Thanks for the reply Dave. I ended up just re-running the detector on the smaller test images."
        }
      ]
    },
    {
      "id": 539915,
      "postDate": "2019-05-30T17:44:14.137Z",
      "content": "<p>Is there a way to get the image height and widths of the original (full size) test and train images without downloading the full large dataset?</p>\n\n<p>I am trying to use the provided bounding boxes (<a href=\"https://github.com/visipedia/iwildcam_comp\">https://github.com/visipedia/iwildcam_comp</a> ), but they appear to be generated using the original, full size images, since they have pixel values like: <code>[2023, 1261, 2371, 1725]</code>. </p>\n\n<p>I would like to apply them to the smaller images, but need to know the width/height of the original images so I can scale down the bounding box. Is this data available anywhere other than by extracting it from the large images? train/test.csv list width as 1024, and the annotation files do not contain dimensions for the IDFG images.</p>",
      "rawMarkdown": "Is there a way to get the image height and widths of the original (full size) test and train images without downloading the full large dataset?\n\nI am trying to use the provided bounding boxes (https://github.com/visipedia/iwildcam_comp ), but they appear to be generated using the original, full size images, since they have pixel values like: `[2023, 1261, 2371, 1725]`. \n\nI would like to apply them to the smaller images, but need to know the width/height of the original images so I can scale down the bounding box. Is this data available anywhere other than by extracting it from the large images? train/test.csv list width as 1024, and the annotation files do not contain dimensions for the IDFG images.",
      "votes": 1
    },
    {
      "id": 540361,
      "postDate": "2019-05-31T12:05:44.823Z",
      "rawMarkdown": "",
      "isDeleted": true
    }
  ],
  "comments": [
    {
      "id": 540080,
      "author_name": "Dave Coates",
      "author_url": "",
      "post_date": "2019-05-31T02:10:49.587000",
      "content": "<p>You can get the original image sizes for the training set from the original annotations: <a href=\"https://lilablobssc.blob.core.windows.net/iwildcam2019/iWildCam_2019_Annotations.tar.gz\">https://lilablobssc.blob.core.windows.net/iwildcam2019/iWildCam_2019_Annotations.tar.gz</a></p>\n\n<p>Unfortunately that doesn't include the test set original sizes...</p>",
      "votes": 1,
      "replies": [
        {
          "id": 540364,
          "author_name": "Josh Shapiro",
          "author_url": "",
          "post_date": "2019-05-31T12:06:16.543000",
          "content": "<p>Thanks for the reply Dave. I ended up just re-running the detector on the smaller test images.</p>",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 540361,
      "author_name": "",
      "author_url": "",
      "post_date": "2019-05-31T12:05:44.823000",
      "content": "",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "540080": "You can get the original image sizes for the training set from the original annotations: https://lilablobssc.blob.core.windows.net/iwildcam2019/iWildCam_2019_Annotations.tar.gz\n\nUnfortunately that doesn't include the test set original sizes...",
    "539915": "Is there a way to get the image height and widths of the original (full size) test and train images without downloading the full large dataset?\n\nI am trying to use the provided bounding boxes (https://github.com/visipedia/iwildcam_comp ), but they appear to be generated using the original, full size images, since they have pixel values like: `[2023, 1261, 2371, 1725]`. \n\nI would like to apply them to the smaller images, but need to know the width/height of the original images so I can scale down the bounding box. Is this data available anywhere other than by extracting it from the large images? train/test.csv list width as 1024, and the annotation files do not contain dimensions for the IDFG images.",
    "540361": ""
  }
}