{
  "id": 243556,
  "title": "Secondary labels are effective for bronze medal (76th solution)",
  "url": "/competitions/birdclef-2021/writeups/csb-secondary-labels-are-effective-for-bronze-meda",
  "author_name": "",
  "post_date": "2021-06-03T06:34:19.195128100Z",
  "votes": 4,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Thank you for the interesting competition, and congratulations to the winning teams!</p>\n<p>This is our 76th (Private: 0.6109, Public: 0.6861) solution.</p>\n<h3>Our solution summary</h3>\n<p>(1) Baseline framework is kkiller’s public notebook (Thanks to <a href=\"https://www.kaggle.com/kneroma\" target=\"_blank\">@kneroma</a>!)<br>\n<a href=\"https://www.kaggle.com/kneroma/clean-fast-simple-bird-identifier-training-colab\" target=\"_blank\">https://www.kaggle.com/kneroma/clean-fast-simple-bird-identifier-training-colab</a></p>\n<p>(2) Just adding secondary labels information</p>\n<ul>\n<li>primary_label: 0.995</li>\n<li>secondary_labels: 0.5 (Using 0.7 or 0.995 is also same performance)</li>\n<li>others: 0.0025  </li>\n</ul>\n<p>(3) Choose threshold that is optimal for train soundscapes</p>\n<p>We also tried data augmentation (noise, mixup … ), test time augmentation, and so on. However, these things did not work in our environment.</p>\n<p>Thank you!</p>",
  "messages": [
    {
      "id": "1333919",
      "postDate": "06/03/2021 06:34:19",
      "content": "<p>Thank you for the interesting competition, and congratulations to the winning teams!</p>\n<p>This is our 76th (Private: 0.6109, Public: 0.6861) solution.</p>\n<h3>Our solution summary</h3>\n<p>(1) Baseline framework is kkiller’s public notebook (Thanks to <a href=\"https://www.kaggle.com/kneroma\" target=\"_blank\">@kneroma</a>!)<br>\n<a href=\"https://www.kaggle.com/kneroma/clean-fast-simple-bird-identifier-training-colab\" target=\"_blank\">https://www.kaggle.com/kneroma/clean-fast-simple-bird-identifier-training-colab</a></p>\n<p>(2) Just adding secondary labels information</p>\n<ul>\n<li>primary_label: 0.995</li>\n<li>secondary_labels: 0.5 (Using 0.7 or 0.995 is also same performance)</li>\n<li>others: 0.0025  </li>\n</ul>\n<p>(3) Choose threshold that is optimal for train soundscapes</p>\n<p>We also tried data augmentation (noise, mixup … ), test time augmentation, and so on. However, these things did not work in our environment.</p>\n<p>Thank you!</p>",
      "rawMarkdown": "Thank you for the interesting competition, and congratulations to the winning teams!\n\nThis is our 76th (Private: 0.6109, Public: 0.6861) solution.\n\n### Our solution summary\n\n(1) Baseline framework is kkiller’s public notebook (Thanks to @kneroma!)\nhttps://www.kaggle.com/kneroma/clean-fast-simple-bird-identifier-training-colab\n\n(2) Just adding secondary labels information\n- primary_label: 0.995\n- secondary_labels: 0.5 (Using 0.7 or 0.995 is also same performance)\n- others: 0.0025  \n\n(3) Choose threshold that is optimal for train soundscapes\n\nWe also tried data augmentation (noise, mixup … ), test time augmentation, and so on. However, these things did not work in our environment.\n\nThank you!",
      "votes": null
    },
    {
      "id": "1333964",
      "postDate": "06/03/2021 07:14:54",
      "content": "<p>Congrats for your first bronze medal <a href=\"https://www.kaggle.com/asasan\" target=\"_blank\">@asasan</a> </p>",
      "rawMarkdown": "Congrats for your first bronze medal @asasan",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1333964,
      "author_name": "kneroma",
      "author_url": "",
      "post_date": "06/03/2021 07:14:54",
      "content": "<p>Congrats for your first bronze medal <a href=\"https://www.kaggle.com/asasan\" target=\"_blank\">@asasan</a> </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1333919": "Thank you for the interesting competition, and congratulations to the winning teams!\n\nThis is our 76th (Private: 0.6109, Public: 0.6861) solution.\n\n### Our solution summary\n\n(1) Baseline framework is kkiller’s public notebook (Thanks to @kneroma!)\nhttps://www.kaggle.com/kneroma/clean-fast-simple-bird-identifier-training-colab\n\n(2) Just adding secondary labels information\n- primary_label: 0.995\n- secondary_labels: 0.5 (Using 0.7 or 0.995 is also same performance)\n- others: 0.0025  \n\n(3) Choose threshold that is optimal for train soundscapes\n\nWe also tried data augmentation (noise, mixup … ), test time augmentation, and so on. However, these things did not work in our environment.\n\nThank you!",
    "1333964": "Congrats for your first bronze medal @asasan"
  },
  "source": "meta"
}