{
  "id": 247049,
  "title": "mAP calculation on \"none\" class",
  "url": "/competitions/siim-covid19-detection/discussion/247049",
  "author_name": "Yousef Rabi",
  "post_date": "2021-06-18T00:58:56.735000",
  "votes": 4,
  "comment_count": 0,
  "views": 0,
  "content": "<p>As far as I understand, the image level predictions have two classes, none and opacity.</p>\n<p>My question is, why does the score change if we only change the score of the none class given that there is only one none prediction in the string? Doesn't mAP rank the predictions per class based on score and calculate precision? Why would the score for this single none prediction make a difference?</p>",
  "messages": [
    {
      "id": 1354886,
      "postDate": "2021-06-18T00:58:56.737Z",
      "content": "<p>As far as I understand, the image level predictions have two classes, none and opacity.</p>\n<p>My question is, why does the score change if we only change the score of the none class given that there is only one none prediction in the string? Doesn't mAP rank the predictions per class based on score and calculate precision? Why would the score for this single none prediction make a difference?</p>",
      "rawMarkdown": "As far as I understand, the image level predictions have two classes, none and opacity.\n\nMy question is, why does the score change if we only change the score of the none class given that there is only one none prediction in the string? Doesn't mAP rank the predictions per class based on score and calculate precision? Why would the score for this single none prediction make a difference?",
      "votes": 4
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "1354886": "As far as I understand, the image level predictions have two classes, none and opacity.\n\nMy question is, why does the score change if we only change the score of the none class given that there is only one none prediction in the string? Doesn't mAP rank the predictions per class based on score and calculate precision? Why would the score for this single none prediction make a difference?"
  }
}