{
  "id": 271077,
  "title": "Non-landmarks / Distractors",
  "url": "/competitions/landmark-retrieval-2021/discussion/271077",
  "author_name": "",
  "post_date": "2021-09-08T12:02:58.871381500Z",
  "votes": null,
  "comment_count": 1,
  "views": 0,
  "content": "<p>What role do out-of-domain images (non-landmarks/ distractors) have in this competition? In the recognition competition, there are out-of-domain images in the test set for which an empty string must be predicted. What about this competition? Are there such images in the test/ index folders?</p>",
  "messages": [
    {
      "id": "1506626",
      "postDate": "09/08/2021 12:02:58",
      "content": "<p>What role do out-of-domain images (non-landmarks/ distractors) have in this competition? In the recognition competition, there are out-of-domain images in the test set for which an empty string must be predicted. What about this competition? Are there such images in the test/ index folders?</p>",
      "rawMarkdown": "What role do out-of-domain images (non-landmarks/ distractors) have in this competition? In the recognition competition, there are out-of-domain images in the test set for which an empty string must be predicted. What about this competition? Are there such images in the test/ index folders?",
      "votes": null
    },
    {
      "id": "1507254",
      "postDate": "09/09/2021 03:00:00",
      "content": "<p>I was also thinking on the same lines. IMHO if we look at the <a href=\"https://www.kaggle.com/c/landmark-retrieval-2021/overview/evaluation\" target=\"_blank\">evaluation metric</a> for this competition, precision at rank k plays an important role. Thus, the higher number of non-landmarks/distractors in the prediction the lower will be precision at rank k and lower will be the overall score. The prediction string should ideally have less of non-landmarks as much as possible. <br>\nWould love to hear others thoughts on this.</p>",
      "rawMarkdown": "I was also thinking on the same lines. IMHO if we look at the [evaluation metric](https://www.kaggle.com/c/landmark-retrieval-2021/overview/evaluation) for this competition, precision at rank k plays an important role. Thus, the higher number of non-landmarks/distractors in the prediction the lower will be precision at rank k and lower will be the overall score. The prediction string should ideally have less of non-landmarks as much as possible. \nWould love to hear others thoughts on this.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1507254,
      "author_name": "sandy1112",
      "author_url": "",
      "post_date": "09/09/2021 03:00:00",
      "content": "<p>I was also thinking on the same lines. IMHO if we look at the <a href=\"https://www.kaggle.com/c/landmark-retrieval-2021/overview/evaluation\" target=\"_blank\">evaluation metric</a> for this competition, precision at rank k plays an important role. Thus, the higher number of non-landmarks/distractors in the prediction the lower will be precision at rank k and lower will be the overall score. The prediction string should ideally have less of non-landmarks as much as possible. <br>\nWould love to hear others thoughts on this.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1506626": "What role do out-of-domain images (non-landmarks/ distractors) have in this competition? In the recognition competition, there are out-of-domain images in the test set for which an empty string must be predicted. What about this competition? Are there such images in the test/ index folders?",
    "1507254": "I was also thinking on the same lines. IMHO if we look at the [evaluation metric](https://www.kaggle.com/c/landmark-retrieval-2021/overview/evaluation) for this competition, precision at rank k plays an important role. Thus, the higher number of non-landmarks/distractors in the prediction the lower will be precision at rank k and lower will be the overall score. The prediction string should ideally have less of non-landmarks as much as possible. \nWould love to hear others thoughts on this."
  },
  "source": "meta"
}