{
  "id": 61978,
  "title": "Can somebody explain me the string result?",
  "url": "/competitions/google-ai-open-images-object-detection-track/discussion/61978",
  "author_name": "",
  "post_date": "2018-07-25T19:47:20.343757100Z",
  "votes": null,
  "comment_count": 3,
  "views": 0,
  "content": "<p>I read the description of contest but i dont know that this string results means . I think that they represent the position and the type of object but i dont know how to get the information from it.\nFor example what this mean :/m/01g317 0.8299584984779358 0.1 0.1 0.9 0.9 /m/01g317 0.8990703821182251 0.1 0.1 0.9 0.9 </p>",
  "messages": [
    {
      "id": "362157",
      "postDate": "07/25/2018 19:47:20",
      "content": "<p>I read the description of contest but i dont know that this string results means . I think that they represent the position and the type of object but i dont know how to get the information from it.\nFor example what this mean :/m/01g317 0.8299584984779358 0.1 0.1 0.9 0.9 /m/01g317 0.8990703821182251 0.1 0.1 0.9 0.9 </p>",
      "rawMarkdown": "I read the description of contest but i dont know that this string results means . I think that they represent the position and the type of object but i dont know how to get the information from it.\nFor example what this mean :/m/01g317 0.8299584984779358 0.1 0.1 0.9 0.9 /m/01g317 0.8990703821182251 0.1 0.1 0.9 0.9",
      "votes": null
    },
    {
      "id": "362221",
      "postDate": "07/26/2018 00:11:55",
      "content": "<p>the \"/m/01g317\" is the class id (you can find the json with the 546 classes on the open images site)\nthe 0.8299584984779358 is the confidence score of the detection\nthe four number after that are the bounding box of the object in relative format (0 - left or top, 1 right or bottom). In this case you probably ran one of the baseline kernels that use classifier. the classifiers gives you classification for the photo, not list of bounding objects so the author of the kernel just framed it generally as most of the photo ( (0.1, 0.1) almost top left corner, (0.9, 0.9) almost bottom right corner)</p>\n\n<p>hope this helps.</p>",
      "rawMarkdown": "the \"/m/01g317\" is the class id (you can find the json with the 546 classes on the open images site)\nthe 0.8299584984779358 is the confidence score of the detection\nthe four number after that are the bounding box of the object in relative format (0 - left or top, 1 right or bottom). In this case you probably ran one of the baseline kernels that use classifier. the classifiers gives you classification for the photo, not list of bounding objects so the author of the kernel just framed it generally as most of the photo ( (0.1, 0.1) almost top left corner, (0.9, 0.9) almost bottom right corner)\n\nhope this helps.",
      "votes": null
    },
    {
      "id": "362910",
      "postDate": "07/27/2018 11:38:10",
      "content": "<p>Is the order of the box coordinates left, top, right, bottom for the submission?\nThe groundtruth data is in a different order: XMin, XMax, YMind, YMax (left, right, top, bottom), but the sample_submission.csv examples have values (0.46 0.08 0.93 0.5) that would make an invalid box (left &gt; right) if that were the submission order. I couldn't find this documented anywhere.</p>",
      "rawMarkdown": "Is the order of the box coordinates left, top, right, bottom for the submission?\nThe groundtruth data is in a different order: XMin, XMax, YMind, YMax (left, right, top, bottom), but the sample_submission.csv examples have values (0.46 0.08 0.93 0.5) that would make an invalid box (left &gt; right) if that were the submission order. I couldn't find this documented anywhere.",
      "votes": null
    },
    {
      "id": "363077",
      "postDate": "07/27/2018 19:52:16",
      "content": "<p>The beauty of standards is that you have so many to choose from...\nYes, every platform have different format. Yes, even within the platform files have different formats. </p>",
      "rawMarkdown": "The beauty of standards is that you have so many to choose from...\nYes, every platform have different format. Yes, even within the platform files have different formats.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 362221,
      "author_name": "moshel",
      "author_url": "",
      "post_date": "07/26/2018 00:11:55",
      "content": "<p>the \"/m/01g317\" is the class id (you can find the json with the 546 classes on the open images site)\nthe 0.8299584984779358 is the confidence score of the detection\nthe four number after that are the bounding box of the object in relative format (0 - left or top, 1 right or bottom). In this case you probably ran one of the baseline kernels that use classifier. the classifiers gives you classification for the photo, not list of bounding objects so the author of the kernel just framed it generally as most of the photo ( (0.1, 0.1) almost top left corner, (0.9, 0.9) almost bottom right corner)</p>\n\n<p>hope this helps.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 362910,
      "author_name": "dannuffer",
      "author_url": "",
      "post_date": "07/27/2018 11:38:10",
      "content": "<p>Is the order of the box coordinates left, top, right, bottom for the submission?\nThe groundtruth data is in a different order: XMin, XMax, YMind, YMax (left, right, top, bottom), but the sample_submission.csv examples have values (0.46 0.08 0.93 0.5) that would make an invalid box (left &gt; right) if that were the submission order. I couldn't find this documented anywhere.</p>",
      "votes": null,
      "replies": [
        {
          "id": 363077,
          "author_name": "moshel",
          "author_url": "",
          "post_date": "07/27/2018 19:52:16",
          "content": "<p>The beauty of standards is that you have so many to choose from...\nYes, every platform have different format. Yes, even within the platform files have different formats. </p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "362157": "I read the description of contest but i dont know that this string results means . I think that they represent the position and the type of object but i dont know how to get the information from it.\nFor example what this mean :/m/01g317 0.8299584984779358 0.1 0.1 0.9 0.9 /m/01g317 0.8990703821182251 0.1 0.1 0.9 0.9",
    "362221": "the \"/m/01g317\" is the class id (you can find the json with the 546 classes on the open images site)\nthe 0.8299584984779358 is the confidence score of the detection\nthe four number after that are the bounding box of the object in relative format (0 - left or top, 1 right or bottom). In this case you probably ran one of the baseline kernels that use classifier. the classifiers gives you classification for the photo, not list of bounding objects so the author of the kernel just framed it generally as most of the photo ( (0.1, 0.1) almost top left corner, (0.9, 0.9) almost bottom right corner)\n\nhope this helps.",
    "362910": "Is the order of the box coordinates left, top, right, bottom for the submission?\nThe groundtruth data is in a different order: XMin, XMax, YMind, YMax (left, right, top, bottom), but the sample_submission.csv examples have values (0.46 0.08 0.93 0.5) that would make an invalid box (left &gt; right) if that were the submission order. I couldn't find this documented anywhere.",
    "363077": "The beauty of standards is that you have so many to choose from...\nYes, every platform have different format. Yes, even within the platform files have different formats."
  },
  "source": "meta"
}