{
  "id": 315306,
  "title": "Asking about idea to improving the dataset.",
  "url": "/competitions/happy-whale-and-dolphin/discussion/315306",
  "author_name": "",
  "post_date": "2022-03-27T12:36:46.985296400Z",
  "votes": 2,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Hi everyone, I'm currently using backfins public dataset (which I believe is the best public dataset), and with my experiments on EfficientNet v1 B5 and image size is 640 and cross-fold validation got me cv 0.83 and lb is 0.747.  I think it's my upperbound in my approach.<br>\nI believe all the top have tricks to improve their own dataset. When it's come to improve dataset I have no idea to improve it. Can you give me suggests on how to do it myself.<br>\nThanks for reading. </p>",
  "messages": [
    {
      "id": "1736542",
      "postDate": "03/27/2022 12:36:46",
      "content": "<p>Hi everyone, I'm currently using backfins public dataset (which I believe is the best public dataset), and with my experiments on EfficientNet v1 B5 and image size is 640 and cross-fold validation got me cv 0.83 and lb is 0.747.  I think it's my upperbound in my approach.<br>\nI believe all the top have tricks to improve their own dataset. When it's come to improve dataset I have no idea to improve it. Can you give me suggests on how to do it myself.<br>\nThanks for reading. </p>",
      "rawMarkdown": "Hi everyone, I'm currently using backfins public dataset (which I believe is the best public dataset), and with my experiments on EfficientNet v1 B5 and image size is 640 and cross-fold validation got me cv 0.83 and lb is 0.747.  I think it's my upperbound in my approach.\nI believe all the top have tricks to improve their own dataset. When it's come to improve dataset I have no idea to improve it. Can you give me suggests on how to do it myself.\nThanks for reading.",
      "votes": null
    },
    {
      "id": "1737517",
      "postDate": "03/28/2022 13:54:52",
      "content": "<p>There are only few weeks before end and I don't think someone will share game changer datasets, but the answer is in your question: you have baseline dataset, I'd recommend to find problems of it at first, and fix these problems at second == you will have better dataset :)</p>",
      "rawMarkdown": "There are only few weeks before end and I don't think someone will share game changer datasets, but the answer is in your question: you have baseline dataset, I'd recommend to find problems of it at first, and fix these problems at second == you will have better dataset :)",
      "votes": null
    },
    {
      "id": "1737684",
      "postDate": "03/28/2022 16:26:56",
      "content": "<p>I'm just realize I was so dumb when ask this question when competitions near the end. Anyway, thanks  for your reply. :)</p>",
      "rawMarkdown": "I'm just realize I was so dumb when ask this question when competitions near the end. Anyway, thanks  for your reply. :)",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1737517,
      "author_name": "kwentar",
      "author_url": "",
      "post_date": "03/28/2022 13:54:52",
      "content": "<p>There are only few weeks before end and I don't think someone will share game changer datasets, but the answer is in your question: you have baseline dataset, I'd recommend to find problems of it at first, and fix these problems at second == you will have better dataset :)</p>",
      "votes": null,
      "replies": [
        {
          "id": 1737684,
          "author_name": "locbaop",
          "author_url": "",
          "post_date": "03/28/2022 16:26:56",
          "content": "<p>I'm just realize I was so dumb when ask this question when competitions near the end. Anyway, thanks  for your reply. :)</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1736542": "Hi everyone, I'm currently using backfins public dataset (which I believe is the best public dataset), and with my experiments on EfficientNet v1 B5 and image size is 640 and cross-fold validation got me cv 0.83 and lb is 0.747.  I think it's my upperbound in my approach.\nI believe all the top have tricks to improve their own dataset. When it's come to improve dataset I have no idea to improve it. Can you give me suggests on how to do it myself.\nThanks for reading.",
    "1737517": "There are only few weeks before end and I don't think someone will share game changer datasets, but the answer is in your question: you have baseline dataset, I'd recommend to find problems of it at first, and fix these problems at second == you will have better dataset :)",
    "1737684": "I'm just realize I was so dumb when ask this question when competitions near the end. Anyway, thanks  for your reply. :)"
  },
  "source": "meta"
}