{
  "id": 270020,
  "title": "Does the test set contain landmarks which are outside of the 81313 ones? ",
  "url": "/competitions/landmark-recognition-2021/discussion/270020",
  "author_name": "",
  "post_date": "2021-09-03T06:59:22.976651800Z",
  "votes": 4,
  "comment_count": 5,
  "views": 0,
  "content": "<p>From the data description in the paper:</p>\n<blockquote>\n  <p>eval dataset<br>\n   This contains images of the landmarks that are present in the<br>\n  above (recognition challenge) index dataset and nonlandmark (distractor) images.</p>\n  <p>index<br>\n  Index dataset (recognition challenge): 100,000 images sampled from the GLDv2 training dataset.</p>\n</blockquote>\n<p>Since GLDv2 has more than 81313 landmarks, does this mean the test set will contain landmarks outside of the 81313 ones? </p>",
  "messages": [
    {
      "id": "1501389",
      "postDate": "09/03/2021 06:59:22",
      "content": "<p>From the data description in the paper:</p>\n<blockquote>\n  <p>eval dataset<br>\n   This contains images of the landmarks that are present in the<br>\n  above (recognition challenge) index dataset and nonlandmark (distractor) images.</p>\n  <p>index<br>\n  Index dataset (recognition challenge): 100,000 images sampled from the GLDv2 training dataset.</p>\n</blockquote>\n<p>Since GLDv2 has more than 81313 landmarks, does this mean the test set will contain landmarks outside of the 81313 ones? </p>",
      "rawMarkdown": "From the data description in the paper:\n\n> eval dataset\n This contains images of the landmarks that are present in the\nabove (recognition challenge) index dataset and nonlandmark (distractor) images.\n\n>index\nIndex dataset (recognition challenge): 100,000 images sampled from the GLDv2 training dataset.\n\nSince GLDv2 has more than 81313 landmarks, does this mean the test set will contain landmarks outside of the 81313 ones?",
      "votes": null
    },
    {
      "id": "1501830",
      "postDate": "09/03/2021 15:17:18",
      "content": "<p>I have checked that they are outside the public training ones</p>",
      "rawMarkdown": "I have checked that they are outside the public training ones",
      "votes": null
    },
    {
      "id": "1501905",
      "postDate": "09/03/2021 16:23:12",
      "content": "<p>Thanks for the clarification. If this is the case, I suppose it is better to 'finetune' models in the original dataset(GLDv2) in order to catch those landmarks. </p>",
      "rawMarkdown": "Thanks for the clarification. If this is the case, I suppose it is better to 'finetune' models in the original dataset(GLDv2) in order to catch those landmarks.",
      "votes": null
    },
    {
      "id": "1501927",
      "postDate": "09/03/2021 16:42:03",
      "content": "<p>Take care of non-landmarks </p>",
      "rawMarkdown": "Take care of non-landmarks",
      "votes": null
    },
    {
      "id": "1515711",
      "postDate": "09/17/2021 12:04:41",
      "content": "<p>Hi, can you make inference on 512x512 images with GPU on kaggle? I'm getting OOM error while extracting embeddings from the index images? can anyone help? how to resolve this problem?</p>",
      "rawMarkdown": "Hi, can you make inference on 512x512 images with GPU on kaggle? I'm getting OOM error while extracting embeddings from the index images? can anyone help? how to resolve this problem?",
      "votes": null
    },
    {
      "id": "1516291",
      "postDate": "09/18/2021 06:51:56",
      "content": "<p>Oh, is that true? Based on the description of this competition below, I thought the private training set are made from the public training set.</p>\n<blockquote>\n  <p>To facilitate recognition-by-retrieval approaches, the private training set contains only a 100k subset of the total public training set.</p>\n</blockquote>",
      "rawMarkdown": "Oh, is that true? Based on the description of this competition below, I thought the private training set are made from the public training set.\n\n> To facilitate recognition-by-retrieval approaches, the private training set contains only a 100k subset of the total public training set.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1501830,
      "author_name": "enric1296",
      "author_url": "",
      "post_date": "09/03/2021 15:17:18",
      "content": "<p>I have checked that they are outside the public training ones</p>",
      "votes": null,
      "replies": [
        {
          "id": 1501905,
          "author_name": "phoenixl",
          "author_url": "",
          "post_date": "09/03/2021 16:23:12",
          "content": "<p>Thanks for the clarification. If this is the case, I suppose it is better to 'finetune' models in the original dataset(GLDv2) in order to catch those landmarks. </p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1501927,
          "author_name": "enric1296",
          "author_url": "",
          "post_date": "09/03/2021 16:42:03",
          "content": "<p>Take care of non-landmarks </p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1516291,
          "author_name": "hdsk38",
          "author_url": "",
          "post_date": "09/18/2021 06:51:56",
          "content": "<p>Oh, is that true? Based on the description of this competition below, I thought the private training set are made from the public training set.</p>\n<blockquote>\n  <p>To facilitate recognition-by-retrieval approaches, the private training set contains only a 100k subset of the total public training set.</p>\n</blockquote>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1515711,
      "author_name": "prtmmishra7",
      "author_url": "",
      "post_date": "09/17/2021 12:04:41",
      "content": "<p>Hi, can you make inference on 512x512 images with GPU on kaggle? I'm getting OOM error while extracting embeddings from the index images? can anyone help? how to resolve this problem?</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1501389": "From the data description in the paper:\n\n> eval dataset\n This contains images of the landmarks that are present in the\nabove (recognition challenge) index dataset and nonlandmark (distractor) images.\n\n>index\nIndex dataset (recognition challenge): 100,000 images sampled from the GLDv2 training dataset.\n\nSince GLDv2 has more than 81313 landmarks, does this mean the test set will contain landmarks outside of the 81313 ones?",
    "1501830": "I have checked that they are outside the public training ones",
    "1501905": "Thanks for the clarification. If this is the case, I suppose it is better to 'finetune' models in the original dataset(GLDv2) in order to catch those landmarks.",
    "1501927": "Take care of non-landmarks",
    "1515711": "Hi, can you make inference on 512x512 images with GPU on kaggle? I'm getting OOM error while extracting embeddings from the index images? can anyone help? how to resolve this problem?",
    "1516291": "Oh, is that true? Based on the description of this competition below, I thought the private training set are made from the public training set.\n\n> To facilitate recognition-by-retrieval approaches, the private training set contains only a 100k subset of the total public training set."
  },
  "source": "meta"
}