{
  "id": 57498,
  "title": "Evaluation: Confidence prediction threshold - False positives",
  "url": "/competitions/cvpr-2018-autonomous-driving/discussion/57498",
  "author_name": "",
  "post_date": "2018-05-24T13:36:53.578226900Z",
  "votes": 1,
  "comment_count": 1,
  "views": 0,
  "content": "<p>I submitted my predictions with by removing predictions with varying confidence_prediction threshold. The threshold went from 0.9 to 0.2. The smaller the threshold I was getting better score. It seems to me that this means that the evaluation score does not penalize much false positives. Should we dump all segmentation masks no matter the prediction confidence?\nHowever, I couldn't submit the predictions with the cut-off threshold 0.1. I got a message that it timed out after 20 minutes.  What is your experience with this? </p>",
  "messages": [
    {
      "id": "333148",
      "postDate": "05/24/2018 13:36:53",
      "content": "<p>I submitted my predictions with by removing predictions with varying confidence_prediction threshold. The threshold went from 0.9 to 0.2. The smaller the threshold I was getting better score. It seems to me that this means that the evaluation score does not penalize much false positives. Should we dump all segmentation masks no matter the prediction confidence?\nHowever, I couldn't submit the predictions with the cut-off threshold 0.1. I got a message that it timed out after 20 minutes.  What is your experience with this? </p>",
      "rawMarkdown": "I submitted my predictions with by removing predictions with varying confidence_prediction threshold. The threshold went from 0.9 to 0.2. The smaller the threshold I was getting better score. It seems to me that this means that the evaluation score does not penalize much false positives. Should we dump all segmentation masks no matter the prediction confidence?\nHowever, I couldn't submit the predictions with the cut-off threshold 0.1. I got a message that it timed out after 20 minutes.  What is your experience with this?",
      "votes": null
    },
    {
      "id": "333457",
      "postDate": "05/25/2018 07:19:58",
      "content": "<p>Our computation is based on interpolated average precision (AP) used in reference [1]. You may check the reference for more details.</p>\n\n<p>[1] Microsoft coco: Common objects in context, <a href=\"http://cocodataset.org/\">http://cocodataset.org/</a></p>",
      "rawMarkdown": "Our computation is based on interpolated average precision (AP) used in reference [1]. You may check the reference for more details.\n\n[1] Microsoft coco: Common objects in context, http://cocodataset.org/",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 333457,
      "author_name": "huangxinyu01",
      "author_url": "",
      "post_date": "05/25/2018 07:19:58",
      "content": "<p>Our computation is based on interpolated average precision (AP) used in reference [1]. You may check the reference for more details.</p>\n\n<p>[1] Microsoft coco: Common objects in context, <a href=\"http://cocodataset.org/\">http://cocodataset.org/</a></p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "333148": "I submitted my predictions with by removing predictions with varying confidence_prediction threshold. The threshold went from 0.9 to 0.2. The smaller the threshold I was getting better score. It seems to me that this means that the evaluation score does not penalize much false positives. Should we dump all segmentation masks no matter the prediction confidence?\nHowever, I couldn't submit the predictions with the cut-off threshold 0.1. I got a message that it timed out after 20 minutes.  What is your experience with this?",
    "333457": "Our computation is based on interpolated average precision (AP) used in reference [1]. You may check the reference for more details.\n\n[1] Microsoft coco: Common objects in context, http://cocodataset.org/"
  },
  "source": "meta"
}