{
  "id": 68896,
  "title": "bounding box coordinate question",
  "url": "/competitions/rsna-pneumonia-detection-challenge/discussion/68896",
  "author_name": "",
  "post_date": "2018-10-18T09:29:10.186086400Z",
  "votes": 1,
  "comment_count": 4,
  "views": 0,
  "content": "<p>Just wanted to confirm as this does not seem to be stated explicitly anywhere, the coordinates are measured from upper left hand corner? It's not the usual xy plane but one aligned with row / column order in a numpy array?</p>\n\n<p>This means that the min x, min y coordinate will be the upper left hand corner of our bounding box? Y being essentially the row,  x being the column. Height extends downwards and width extends to the right?</p>",
  "messages": [
    {
      "id": "405884",
      "postDate": "10/18/2018 09:29:10",
      "content": "<p>Just wanted to confirm as this does not seem to be stated explicitly anywhere, the coordinates are measured from upper left hand corner? It's not the usual xy plane but one aligned with row / column order in a numpy array?</p>\n\n<p>This means that the min x, min y coordinate will be the upper left hand corner of our bounding box? Y being essentially the row,  x being the column. Height extends downwards and width extends to the right?</p>",
      "rawMarkdown": "Just wanted to confirm as this does not seem to be stated explicitly anywhere, the coordinates are measured from upper left hand corner? It's not the usual xy plane but one aligned with row / column order in a numpy array?\n\nThis means that the min x, min y coordinate will be the upper left hand corner of our bounding box? Y being essentially the row,  x being the column. Height extends downwards and width extends to the right?",
      "votes": null
    },
    {
      "id": "406107",
      "postDate": "10/18/2018 16:40:38",
      "content": "<p>I have the same understanding as you.</p>",
      "rawMarkdown": "I have the same understanding as you.",
      "votes": null
    },
    {
      "id": "406161",
      "postDate": "10/18/2018 18:40:59",
      "content": "<p>From the Data tab:</p>\n\n<blockquote>\n  <p>x_ - the upper-left x coordinate of the bounding box.</p>\n  \n  <p>y_ - the upper-left y coordinate of the bounding box.</p>\n</blockquote>",
      "rawMarkdown": "From the Data tab:\n\n&gt; x_ - the upper-left x coordinate of the bounding box.\n\n&gt; y_ - the upper-left y coordinate of the bounding box.",
      "votes": null
    },
    {
      "id": "408839",
      "postDate": "10/23/2018 14:29:42",
      "content": "<p>Can anyone confirm if the bbox predictions to be submitted (x, y, width, height) should be whole numbers or 1 decimal place accuracy? \nIn the submission example file they used 1 decimal place for x. \nHowever in the training csv all the 4 are whole numbers.\nThanks!</p>",
      "rawMarkdown": "Can anyone confirm if the bbox predictions to be submitted (x, y, width, height) should be whole numbers or 1 decimal place accuracy? \nIn the submission example file they used 1 decimal place for x. \nHowever in the training csv all the 4 are whole numbers.\nThanks!",
      "votes": null
    },
    {
      "id": "408869",
      "postDate": "10/23/2018 14:55:58",
      "content": "<p>I've submitted both integers and floats. Kaggle accepts both. I don't know if they round the floats before evaluating, but changing a coordinate by 1 pixel in either direction seems unlikely to make a meaningful change in your score.</p>",
      "rawMarkdown": "I've submitted both integers and floats. Kaggle accepts both. I don't know if they round the floats before evaluating, but changing a coordinate by 1 pixel in either direction seems unlikely to make a meaningful change in your score.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 406107,
      "author_name": "agentili",
      "author_url": "",
      "post_date": "10/18/2018 16:40:38",
      "content": "<p>I have the same understanding as you.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 406161,
      "author_name": "sgalib",
      "author_url": "",
      "post_date": "10/18/2018 18:40:59",
      "content": "<p>From the Data tab:</p>\n\n<blockquote>\n  <p>x_ - the upper-left x coordinate of the bounding box.</p>\n  \n  <p>y_ - the upper-left y coordinate of the bounding box.</p>\n</blockquote>",
      "votes": null,
      "replies": []
    },
    {
      "id": 408839,
      "author_name": "rthiyagarajan",
      "author_url": "",
      "post_date": "10/23/2018 14:29:42",
      "content": "<p>Can anyone confirm if the bbox predictions to be submitted (x, y, width, height) should be whole numbers or 1 decimal place accuracy? \nIn the submission example file they used 1 decimal place for x. \nHowever in the training csv all the 4 are whole numbers.\nThanks!</p>",
      "votes": null,
      "replies": [
        {
          "id": 408869,
          "author_name": "richardepstein",
          "author_url": "",
          "post_date": "10/23/2018 14:55:58",
          "content": "<p>I've submitted both integers and floats. Kaggle accepts both. I don't know if they round the floats before evaluating, but changing a coordinate by 1 pixel in either direction seems unlikely to make a meaningful change in your score.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "405884": "Just wanted to confirm as this does not seem to be stated explicitly anywhere, the coordinates are measured from upper left hand corner? It's not the usual xy plane but one aligned with row / column order in a numpy array?\n\nThis means that the min x, min y coordinate will be the upper left hand corner of our bounding box? Y being essentially the row,  x being the column. Height extends downwards and width extends to the right?",
    "406107": "I have the same understanding as you.",
    "406161": "From the Data tab:\n\n&gt; x_ - the upper-left x coordinate of the bounding box.\n\n&gt; y_ - the upper-left y coordinate of the bounding box.",
    "408839": "Can anyone confirm if the bbox predictions to be submitted (x, y, width, height) should be whole numbers or 1 decimal place accuracy? \nIn the submission example file they used 1 decimal place for x. \nHowever in the training csv all the 4 are whole numbers.\nThanks!",
    "408869": "I've submitted both integers and floats. Kaggle accepts both. I don't know if they round the floats before evaluating, but changing a coordinate by 1 pixel in either direction seems unlikely to make a meaningful change in your score."
  },
  "source": "meta"
}