{
  "id": 463944,
  "title": "Local score(CV) ? Leaderboard score(LB) ?",
  "url": "/competitions/aocr2024/discussion/463944",
  "author_name": "",
  "post_date": "2023-12-28T06:48:54.144140400Z",
  "votes": 1,
  "comment_count": 3,
  "views": 0,
  "content": "<p>I'm curious about why so many teams are achieving such high scores on the leaderboard. Despite training on the entire dataset and overfitting it, I haven't been able to achieve a local score greater than 0.93. Would you be willing to share your local  score along with its corresponding leaderboard score?</p>\n<p>In my best-performing scenarios:</p>\n<p>Single fold (1/5)<br>\n2D SwinV2 model: CV 0.62 / LB 0.67<br>\n2.5D CoatLiteMini model: CV 0.72 / LB 0.74<br>\n3D UNet: CV 0.87 / LB 0.85\"</p>",
  "messages": [
    {
      "id": "2576836",
      "postDate": "12/28/2023 06:48:54",
      "content": "<p>I'm curious about why so many teams are achieving such high scores on the leaderboard. Despite training on the entire dataset and overfitting it, I haven't been able to achieve a local score greater than 0.93. Would you be willing to share your local  score along with its corresponding leaderboard score?</p>\n<p>In my best-performing scenarios:</p>\n<p>Single fold (1/5)<br>\n2D SwinV2 model: CV 0.62 / LB 0.67<br>\n2.5D CoatLiteMini model: CV 0.72 / LB 0.74<br>\n3D UNet: CV 0.87 / LB 0.85\"</p>",
      "rawMarkdown": "I'm curious about why so many teams are achieving such high scores on the leaderboard. Despite training on the entire dataset and overfitting it, I haven't been able to achieve a local score greater than 0.93. Would you be willing to share your local  score along with its corresponding leaderboard score?\n\nIn my best-performing scenarios:\n\nSingle fold (1/5)\n2D SwinV2 model: CV 0.62 / LB 0.67\n2.5D CoatLiteMini model: CV 0.72 / LB 0.74\n3D UNet: CV 0.87 / LB 0.85\"",
      "votes": null
    },
    {
      "id": "2587833",
      "postDate": "01/05/2024 04:19:51",
      "content": "<p>Our cross-validation score looks the same as you. +- 0.5 comparing to learderboard score</p>",
      "rawMarkdown": "Our cross-validation score looks the same as you. +- 0.5 comparing to learderboard score",
      "votes": null
    },
    {
      "id": "2591448",
      "postDate": "01/08/2024 01:11:59",
      "content": "<p>1)  I'm not sure if I understand your question; perhaps you could elaborate further.</p>\n<p>2) As for how to achieve a score higher than 0.9, I don't know either, but I think you need to incorporate medical knowledge into the algorithm. This probably isn't a topic where you can achieve the highest score by simply using deep learning training.</p>",
      "rawMarkdown": "1)  I'm not sure if I understand your question; perhaps you could elaborate further.\n\n2) As for how to achieve a score higher than 0.9, I don't know either, but I think you need to incorporate medical knowledge into the algorithm. This probably isn't a topic where you can achieve the highest score by simply using deep learning training.",
      "votes": null
    },
    {
      "id": "2593783",
      "postDate": "01/09/2024 12:59:54",
      "content": "<p><a href=\"https://www.kaggle.com/atom1231\" target=\"_blank\">@atom1231</a> is asking for validation scores (local score) as compared to test set (leaderboard score)</p>",
      "rawMarkdown": "atom1231 is asking for validation scores (local score) as compared to test set (leaderboard score)",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2587833,
      "author_name": "thangngoc89",
      "author_url": "",
      "post_date": "01/05/2024 04:19:51",
      "content": "<p>Our cross-validation score looks the same as you. +- 0.5 comparing to learderboard score</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 2591448,
      "author_name": "goman178",
      "author_url": "",
      "post_date": "01/08/2024 01:11:59",
      "content": "<p>1)  I'm not sure if I understand your question; perhaps you could elaborate further.</p>\n<p>2) As for how to achieve a score higher than 0.9, I don't know either, but I think you need to incorporate medical knowledge into the algorithm. This probably isn't a topic where you can achieve the highest score by simply using deep learning training.</p>",
      "votes": null,
      "replies": [
        {
          "id": 2593783,
          "author_name": "thangngoc89",
          "author_url": "",
          "post_date": "01/09/2024 12:59:54",
          "content": "<p><a href=\"https://www.kaggle.com/atom1231\" target=\"_blank\">@atom1231</a> is asking for validation scores (local score) as compared to test set (leaderboard score)</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "2576836": "I'm curious about why so many teams are achieving such high scores on the leaderboard. Despite training on the entire dataset and overfitting it, I haven't been able to achieve a local score greater than 0.93. Would you be willing to share your local  score along with its corresponding leaderboard score?\n\nIn my best-performing scenarios:\n\nSingle fold (1/5)\n2D SwinV2 model: CV 0.62 / LB 0.67\n2.5D CoatLiteMini model: CV 0.72 / LB 0.74\n3D UNet: CV 0.87 / LB 0.85\"",
    "2587833": "Our cross-validation score looks the same as you. +- 0.5 comparing to learderboard score",
    "2591448": "1)  I'm not sure if I understand your question; perhaps you could elaborate further.\n\n2) As for how to achieve a score higher than 0.9, I don't know either, but I think you need to incorporate medical knowledge into the algorithm. This probably isn't a topic where you can achieve the highest score by simply using deep learning training.",
    "2593783": "atom1231 is asking for validation scores (local score) as compared to test set (leaderboard score)"
  },
  "source": "meta"
}