{
  "id": 54618,
  "title": "Car vs Car Group and impact on score",
  "url": "/competitions/cvpr-2018-autonomous-driving/discussion/54618",
  "author_name": "",
  "post_date": "2018-04-15T23:35:48.699194300Z",
  "votes": 1,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Regarding: \"In this competition, we evaluate seven different instance-level annotations, which are car, motorcycle, bicycle, pedestrian, truck, bus, and tricycle. The corresponding groups, such as car group and bicycle group, are annotated when boundaries cannot be distinguished by labelers. These groups are not evaluated currently.\"</p>\n\n<p>What if my model is able to distinguish individual cars within a car group? It seems that will decrease the intersection over union metrics for cars and thus hurt the overall score.  Am I correct?  I think we should not be penalized for this.</p>",
  "messages": [
    {
      "id": "314606",
      "postDate": "04/15/2018 23:35:48",
      "content": "<p>Regarding: \"In this competition, we evaluate seven different instance-level annotations, which are car, motorcycle, bicycle, pedestrian, truck, bus, and tricycle. The corresponding groups, such as car group and bicycle group, are annotated when boundaries cannot be distinguished by labelers. These groups are not evaluated currently.\"</p>\n\n<p>What if my model is able to distinguish individual cars within a car group? It seems that will decrease the intersection over union metrics for cars and thus hurt the overall score.  Am I correct?  I think we should not be penalized for this.</p>",
      "rawMarkdown": "Regarding: \"In this competition, we evaluate seven different instance-level annotations, which are car, motorcycle, bicycle, pedestrian, truck, bus, and tricycle. The corresponding groups, such as car group and bicycle group, are annotated when boundaries cannot be distinguished by labelers. These groups are not evaluated currently.\"\n\nWhat if my model is able to distinguish individual cars within a car group? It seems that will decrease the intersection over union metrics for cars and thus hurt the overall score.  Am I correct?  I think we should not be penalized for this.",
      "votes": null
    },
    {
      "id": "314710",
      "postDate": "04/16/2018 06:42:49",
      "content": "<p>It will not hurt the overall score. Any predicted cars that have overlapping areas with car groups will be ignored.</p>",
      "rawMarkdown": "It will not hurt the overall score. Any predicted cars that have overlapping areas with car groups will be ignored.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 314710,
      "author_name": "huangxinyu01",
      "author_url": "",
      "post_date": "04/16/2018 06:42:49",
      "content": "<p>It will not hurt the overall score. Any predicted cars that have overlapping areas with car groups will be ignored.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "314606": "Regarding: \"In this competition, we evaluate seven different instance-level annotations, which are car, motorcycle, bicycle, pedestrian, truck, bus, and tricycle. The corresponding groups, such as car group and bicycle group, are annotated when boundaries cannot be distinguished by labelers. These groups are not evaluated currently.\"\n\nWhat if my model is able to distinguish individual cars within a car group? It seems that will decrease the intersection over union metrics for cars and thus hurt the overall score.  Am I correct?  I think we should not be penalized for this.",
    "314710": "It will not hurt the overall score. Any predicted cars that have overlapping areas with car groups will be ignored."
  },
  "source": "meta"
}