{
  "id": 94524,
  "title": "What is the format of PredictionString?",
  "url": "/competitions/open-images-2019-visual-relationship/discussion/94524",
  "author_name": "",
  "post_date": "2019-06-05T02:47:42.560233200Z",
  "votes": 1,
  "comment_count": 2,
  "views": 0,
  "content": "<p>I don't know what is the exact format of PredictionString in submission sample file. I think it contain location and class label of 2 objects and a string denoting the relationship between them (for example: \"on\"). Any better explanation?</p>",
  "messages": [
    {
      "id": "543997",
      "postDate": "06/05/2019 02:47:42",
      "content": "<p>I don't know what is the exact format of PredictionString in submission sample file. I think it contain location and class label of 2 objects and a string denoting the relationship between them (for example: \"on\"). Any better explanation?</p>",
      "rawMarkdown": "I don't know what is the exact format of PredictionString in submission sample file. I think it contain location and class label of 2 objects and a string denoting the relationship between them (for example: \"on\"). Any better explanation?",
      "votes": null
    },
    {
      "id": "544018",
      "postDate": "06/05/2019 03:25:27",
      "content": "<p><code>ImageId, {Confidence Label1 XMin1 YMin1 XMax1 YMax1 Label2 XMin2 YMin2 XMax2 YMax2 RelationLabel}...</code></p>\n\n<p>The first number of the 12-tuple just seems to be used for the Recall@50 metric to sort your relationship triples by confidence and see if any of the top 50 predictions (sorted by confidence) are true. Then two 5-tuples of [object, x1, y1, x2, y2] for the 2 objects, finally the relationship label for the two objects.</p>",
      "rawMarkdown": "`ImageId, {Confidence Label1 XMin1 YMin1 XMax1 YMax1 Label2 XMin2 YMin2 XMax2 YMax2 RelationLabel}...`\n\nThe first number of the 12-tuple just seems to be used for the Recall@50 metric to sort your relationship triples by confidence and see if any of the top 50 predictions (sorted by confidence) are true. Then two 5-tuples of [object, x1, y1, x2, y2] for the 2 objects, finally the relationship label for the two objects.",
      "votes": null
    },
    {
      "id": "553173",
      "postDate": "06/15/2019 09:02:15",
      "content": "<p>Label1: The label of the first bounding box. The label may be an object or a person.\nXMin1 &amp; XMax1: Minimum and maximum value of the x-coordinate of the bounding box 1.\nYMin1 &amp; YMax1: Minimum and maximum value of the y-coordinate of the bounding box 1.\nLabel2: The label of the second bounding box. The label may be an object or a person.\nXMin2 &amp; XMax2: Minimum and maximum value of the x-coordinate of the bounding box 2.\nYMin2 &amp; YMax2: Minimum and maximum value of the y-coordinate of the bounding box 2.\nThe confidence score is the probability of that label being that object.\nRelationLabel: It shows the interactions among the objects.\nThis can be extracted from the prediction string by the code below:</p>\n\n<p>values_what = df_sample[df_sample['ImageId']=='b4c3b52a8723d431']['PredictionString'].values\nvalues = str(values_what)[2:-2].split(' ')</p>\n\n<p>Then acess the values label wise.</p>",
      "rawMarkdown": "Label1: The label of the first bounding box. The label may be an object or a person.\nXMin1 &amp; XMax1: Minimum and maximum value of the x-coordinate of the bounding box 1.\nYMin1 &amp; YMax1: Minimum and maximum value of the y-coordinate of the bounding box 1.\nLabel2: The label of the second bounding box. The label may be an object or a person.\nXMin2 &amp; XMax2: Minimum and maximum value of the x-coordinate of the bounding box 2.\nYMin2 &amp; YMax2: Minimum and maximum value of the y-coordinate of the bounding box 2.\nThe confidence score is the probability of that label being that object.\nRelationLabel: It shows the interactions among the objects.\nThis can be extracted from the prediction string by the code below:\n\nvalues_what = df_sample[df_sample['ImageId']=='b4c3b52a8723d431']['PredictionString'].values\nvalues = str(values_what)[2:-2].split(' ')\n\nThen acess the values label wise.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 544018,
      "author_name": "bharrisau",
      "author_url": "",
      "post_date": "06/05/2019 03:25:27",
      "content": "<p><code>ImageId, {Confidence Label1 XMin1 YMin1 XMax1 YMax1 Label2 XMin2 YMin2 XMax2 YMax2 RelationLabel}...</code></p>\n\n<p>The first number of the 12-tuple just seems to be used for the Recall@50 metric to sort your relationship triples by confidence and see if any of the top 50 predictions (sorted by confidence) are true. Then two 5-tuples of [object, x1, y1, x2, y2] for the 2 objects, finally the relationship label for the two objects.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 553173,
      "author_name": "swainsubrat",
      "author_url": "",
      "post_date": "06/15/2019 09:02:15",
      "content": "<p>Label1: The label of the first bounding box. The label may be an object or a person.\nXMin1 &amp; XMax1: Minimum and maximum value of the x-coordinate of the bounding box 1.\nYMin1 &amp; YMax1: Minimum and maximum value of the y-coordinate of the bounding box 1.\nLabel2: The label of the second bounding box. The label may be an object or a person.\nXMin2 &amp; XMax2: Minimum and maximum value of the x-coordinate of the bounding box 2.\nYMin2 &amp; YMax2: Minimum and maximum value of the y-coordinate of the bounding box 2.\nThe confidence score is the probability of that label being that object.\nRelationLabel: It shows the interactions among the objects.\nThis can be extracted from the prediction string by the code below:</p>\n\n<p>values_what = df_sample[df_sample['ImageId']=='b4c3b52a8723d431']['PredictionString'].values\nvalues = str(values_what)[2:-2].split(' ')</p>\n\n<p>Then acess the values label wise.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "543997": "I don't know what is the exact format of PredictionString in submission sample file. I think it contain location and class label of 2 objects and a string denoting the relationship between them (for example: \"on\"). Any better explanation?",
    "544018": "`ImageId, {Confidence Label1 XMin1 YMin1 XMax1 YMax1 Label2 XMin2 YMin2 XMax2 YMax2 RelationLabel}...`\n\nThe first number of the 12-tuple just seems to be used for the Recall@50 metric to sort your relationship triples by confidence and see if any of the top 50 predictions (sorted by confidence) are true. Then two 5-tuples of [object, x1, y1, x2, y2] for the 2 objects, finally the relationship label for the two objects.",
    "553173": "Label1: The label of the first bounding box. The label may be an object or a person.\nXMin1 &amp; XMax1: Minimum and maximum value of the x-coordinate of the bounding box 1.\nYMin1 &amp; YMax1: Minimum and maximum value of the y-coordinate of the bounding box 1.\nLabel2: The label of the second bounding box. The label may be an object or a person.\nXMin2 &amp; XMax2: Minimum and maximum value of the x-coordinate of the bounding box 2.\nYMin2 &amp; YMax2: Minimum and maximum value of the y-coordinate of the bounding box 2.\nThe confidence score is the probability of that label being that object.\nRelationLabel: It shows the interactions among the objects.\nThis can be extracted from the prediction string by the code below:\n\nvalues_what = df_sample[df_sample['ImageId']=='b4c3b52a8723d431']['PredictionString'].values\nvalues = str(values_what)[2:-2].split(' ')\n\nThen acess the values label wise."
  },
  "source": "meta"
}