{
  "id": 174076,
  "title": "About Scoring Script",
  "url": "/competitions/landmark-retrieval-2020/discussion/174076",
  "author_name": "",
  "post_date": "2020-08-12T07:52:33.499810500Z",
  "votes": 2,
  "comment_count": 2,
  "views": 0,
  "content": "<p>I implemented the <a href=\"https://www.kaggle.com/c/landmark-retrieval-2020/discussion/165400\" target=\"_blank\">official scoring script</a> with <a href=\"https://www.kaggle.com/camaskew/baseline-landmark-retrieval-model\" target=\"_blank\">official baseline model</a>. But the score of public test dataset is only 0.02 (not 0.271). What's wrong with <a href=\"https://www.kaggle.com/wuliaokaola/glr2020-metric\" target=\"_blank\">my implementation</a>? <br>\nAny help would be greatly appreciated.</p>",
  "messages": [
    {
      "id": "967369",
      "postDate": "08/12/2020 07:52:33",
      "content": "<p>I implemented the <a href=\"https://www.kaggle.com/c/landmark-retrieval-2020/discussion/165400\" target=\"_blank\">official scoring script</a> with <a href=\"https://www.kaggle.com/camaskew/baseline-landmark-retrieval-model\" target=\"_blank\">official baseline model</a>. But the score of public test dataset is only 0.02 (not 0.271). What's wrong with <a href=\"https://www.kaggle.com/wuliaokaola/glr2020-metric\" target=\"_blank\">my implementation</a>? <br>\nAny help would be greatly appreciated.</p>",
      "rawMarkdown": "I implemented the [official scoring script](https://www.kaggle.com/c/landmark-retrieval-2020/discussion/165400) with [official baseline model](https://www.kaggle.com/camaskew/baseline-landmark-retrieval-model). But the score of public test dataset is only 0.02 (not 0.271). What's wrong with [my implementation](https://www.kaggle.com/wuliaokaola/glr2020-metric)? \nAny help would be greatly appreciated.",
      "votes": null
    },
    {
      "id": "968614",
      "postDate": "08/13/2020 06:25:42",
      "content": "<p>Your implementation should be ok. But if you use the csv,  <a href=\"https://s3.amazonaws.com/google-landmark/ground_truth/retrieval_solution_v2.1.csv\" target=\"_blank\">https://s3.amazonaws.com/google-landmark/ground_truth/retrieval_solution_v2.1.csv</a>, you need to add test/index images from <a href=\"https://github.com/cvdfoundation/google-landmark\" target=\"_blank\">original dataset</a> on the competition dataset. Because the test/index competiton dataset is a sub set of original dataset, and the csv contains images which are not found on the competition dataset. I think this is the reason. </p>",
      "rawMarkdown": "Your implementation should be ok. But if you use the csv,  https://s3.amazonaws.com/google-landmark/ground_truth/retrieval_solution_v2.1.csv, you need to add test/index images from [original dataset](https://github.com/cvdfoundation/google-landmark) on the competition dataset. Because the test/index competiton dataset is a sub set of original dataset, and the csv contains images which are not found on the competition dataset. I think this is the reason.",
      "votes": null
    },
    {
      "id": "968814",
      "postDate": "08/13/2020 09:01:04",
      "content": "<p>Thank you for your reply. <br>\nThere are 1129 / 762K images described in <a href=\"https://arxiv.org/pdf/2001.05027.pdf\" target=\"_blank\">the paper</a>, The same number of public test dataset we have. And no missing image found.</p>",
      "rawMarkdown": "Thank you for your reply. \nThere are 1129 / 762K images described in [the paper](https://arxiv.org/pdf/2001.05027.pdf), The same number of public test dataset we have. And no missing image found.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 968614,
      "author_name": "sai11fkaneko",
      "author_url": "",
      "post_date": "08/13/2020 06:25:42",
      "content": "<p>Your implementation should be ok. But if you use the csv,  <a href=\"https://s3.amazonaws.com/google-landmark/ground_truth/retrieval_solution_v2.1.csv\" target=\"_blank\">https://s3.amazonaws.com/google-landmark/ground_truth/retrieval_solution_v2.1.csv</a>, you need to add test/index images from <a href=\"https://github.com/cvdfoundation/google-landmark\" target=\"_blank\">original dataset</a> on the competition dataset. Because the test/index competiton dataset is a sub set of original dataset, and the csv contains images which are not found on the competition dataset. I think this is the reason. </p>",
      "votes": null,
      "replies": [
        {
          "id": 968814,
          "author_name": "wuliaokaola",
          "author_url": "",
          "post_date": "08/13/2020 09:01:04",
          "content": "<p>Thank you for your reply. <br>\nThere are 1129 / 762K images described in <a href=\"https://arxiv.org/pdf/2001.05027.pdf\" target=\"_blank\">the paper</a>, The same number of public test dataset we have. And no missing image found.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "967369": "I implemented the [official scoring script](https://www.kaggle.com/c/landmark-retrieval-2020/discussion/165400) with [official baseline model](https://www.kaggle.com/camaskew/baseline-landmark-retrieval-model). But the score of public test dataset is only 0.02 (not 0.271). What's wrong with [my implementation](https://www.kaggle.com/wuliaokaola/glr2020-metric)? \nAny help would be greatly appreciated.",
    "968614": "Your implementation should be ok. But if you use the csv,  https://s3.amazonaws.com/google-landmark/ground_truth/retrieval_solution_v2.1.csv, you need to add test/index images from [original dataset](https://github.com/cvdfoundation/google-landmark) on the competition dataset. Because the test/index competiton dataset is a sub set of original dataset, and the csv contains images which are not found on the competition dataset. I think this is the reason.",
    "968814": "Thank you for your reply. \nThere are 1129 / 762K images described in [the paper](https://arxiv.org/pdf/2001.05027.pdf), The same number of public test dataset we have. And no missing image found."
  },
  "source": "meta"
}