{
  "id": 163628,
  "title": "Any difference between this dataset and GLDv2 ?",
  "url": "/competitions/landmark-retrieval-2020/discussion/163628",
  "author_name": "Noé Pion",
  "post_date": "2020-07-02T19:45:08.596000",
  "votes": 4,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Hi,</p>\n\n<p>Thanks for organizing this great competition !</p>\n\n<p>Are there any differences between this year's dataset (train/test/index) and the one released this year ?</p>\n\n<p>Also, last year's test set contained a lot of images in the test set which were big outliers, such as drawings or photos inside a building when trying to retrieve the outside. Those test images represented a big part of the final test retrieval images. Will such differences be present in the final test set or is it a more typical landmarks retrieval dataset ?</p>\n\n<p>Cheers,\nNoé</p>",
  "messages": [
    {
      "id": 912935,
      "postDate": "2020-07-02T20:35:34.457Z",
      "content": "<p>Hi Noe,</p>\n\n<p>The <em>private</em> test and index image sets (which are not visible to participants) are completely new. They are the ones used for all scoring. The <em>public</em> test/index/train image sets (which are visible to participants) are subsets of last year's data.</p>\n\n<p>The private test set contains only landmark images (no distractors such as drawings are present).</p>\n\n<p>One small comment regarding last year's data: there were indeed distractor images in the test set, but note that they were actually not used in scoring for the 2019 retrieval challenge (they were only used for the 2019 recognition challenge). So, the 2020 and the 2019 retrieval competitions are similar in that distractor test images are not used for scoring.</p>\n\n<p>Thanks,</p>\n\n<p>Andre</p>",
      "rawMarkdown": "Hi Noe,\n\nThe *private* test and index image sets (which are not visible to participants) are completely new. They are the ones used for all scoring. The *public* test/index/train image sets (which are visible to participants) are subsets of last year's data.\n\nThe private test set contains only landmark images (no distractors such as drawings are present).\n\nOne small comment regarding last year's data: there were indeed distractor images in the test set, but note that they were actually not used in scoring for the 2019 retrieval challenge (they were only used for the 2019 recognition challenge). So, the 2020 and the 2019 retrieval competitions are similar in that distractor test images are not used for scoring.\n\nThanks,\n\nAndre",
      "votes": 7
    },
    {
      "id": 912892,
      "postDate": "2020-07-02T19:45:08.597Z",
      "content": "<p>Hi,</p>\n\n<p>Thanks for organizing this great competition !</p>\n\n<p>Are there any differences between this year's dataset (train/test/index) and the one released this year ?</p>\n\n<p>Also, last year's test set contained a lot of images in the test set which were big outliers, such as drawings or photos inside a building when trying to retrieve the outside. Those test images represented a big part of the final test retrieval images. Will such differences be present in the final test set or is it a more typical landmarks retrieval dataset ?</p>\n\n<p>Cheers,\nNoé</p>",
      "rawMarkdown": "Hi,\n\nThanks for organizing this great competition !\n\nAre there any differences between this year's dataset (train/test/index) and the one released this year ?\n\nAlso, last year's test set contained a lot of images in the test set which were big outliers, such as drawings or photos inside a building when trying to retrieve the outside. Those test images represented a big part of the final test retrieval images. Will such differences be present in the final test set or is it a more typical landmarks retrieval dataset ?\n\nCheers,\nNoé",
      "votes": 4
    },
    {
      "id": 1361802,
      "postDate": "2021-06-23T04:22:35.107Z",
      "rawMarkdown": "",
      "isDeleted": true
    }
  ],
  "comments": [
    {
      "id": 912935,
      "author_name": "Andre Araujo",
      "author_url": "",
      "post_date": "2020-07-02T20:35:34.457000",
      "content": "<p>Hi Noe,</p>\n\n<p>The <em>private</em> test and index image sets (which are not visible to participants) are completely new. They are the ones used for all scoring. The <em>public</em> test/index/train image sets (which are visible to participants) are subsets of last year's data.</p>\n\n<p>The private test set contains only landmark images (no distractors such as drawings are present).</p>\n\n<p>One small comment regarding last year's data: there were indeed distractor images in the test set, but note that they were actually not used in scoring for the 2019 retrieval challenge (they were only used for the 2019 recognition challenge). So, the 2020 and the 2019 retrieval competitions are similar in that distractor test images are not used for scoring.</p>\n\n<p>Thanks,</p>\n\n<p>Andre</p>",
      "votes": 7,
      "replies": []
    },
    {
      "id": 1361802,
      "author_name": "",
      "author_url": "",
      "post_date": "2021-06-23T04:22:35.107000",
      "content": "",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "912935": "Hi Noe,\n\nThe *private* test and index image sets (which are not visible to participants) are completely new. They are the ones used for all scoring. The *public* test/index/train image sets (which are visible to participants) are subsets of last year's data.\n\nThe private test set contains only landmark images (no distractors such as drawings are present).\n\nOne small comment regarding last year's data: there were indeed distractor images in the test set, but note that they were actually not used in scoring for the 2019 retrieval challenge (they were only used for the 2019 recognition challenge). So, the 2020 and the 2019 retrieval competitions are similar in that distractor test images are not used for scoring.\n\nThanks,\n\nAndre",
    "912892": "Hi,\n\nThanks for organizing this great competition !\n\nAre there any differences between this year's dataset (train/test/index) and the one released this year ?\n\nAlso, last year's test set contained a lot of images in the test set which were big outliers, such as drawings or photos inside a building when trying to retrieve the outside. Those test images represented a big part of the final test retrieval images. Will such differences be present in the final test set or is it a more typical landmarks retrieval dataset ?\n\nCheers,\nNoé",
    "1361802": ""
  }
}