{
  "id": 307617,
  "title": "115th (would have been ~48th ) - WBF ensemble (same model)",
  "url": "/competitions/tensorflow-great-barrier-reef/discussion/307617",
  "author_name": "",
  "post_date": "2022-02-15T01:51:44.426518900Z",
  "votes": 9,
  "comment_count": 5,
  "views": 0,
  "content": "<p>Hi all!<br>\nCongratulation to all the winners and learners! </p>\n<p>It's 3:50 AM in the morning my time so I am a bit sleepy but super happy with my results! </p>\n<p>My best solution got me 0.678 which would get me 0.692 if I selected better,&nbsp; it is a very simple idea (at least for my humble experience !) </p>\n<p>I used a YOLOv5 shared model (which was better than my own) but ensembled it with itself (on different resolutions) - high, medium, and low + tracking</p>\n<p>The difference between my current place and the silver zone place that I would have got is just a small fraction in tunning image sizes. </p>\n<p>One of the things that I was excited about and did work (but not my best score) was the layer of binary classification.&nbsp;I will share the training code once I have the time. </p>\n<p>Thanks to all of you and a special thank you go to <a href=\"https://www.kaggle.com/cdeotte\" target=\"_blank\">@cdeotte</a> for his life-saving advice in \"pinning the original environment\" when I had the norfair issue. </p>\n<p>Also, a special thank you to <a href=\"https://www.kaggle.com/freshair1996\" target=\"_blank\">@freshair1996</a> for publically sharing the model, it was better than all the training I did on my limited hardware. </p>\n<p>Also, a special thank you to <a href=\"https://www.kaggle.com/alexteboul\" target=\"_blank\">@alexteboul</a> for providing the \"COTS v NotCOTS Cropped Crown of Thorns\" dataset which I used in my binary classification. </p>\n<p>And finally \"once the host approves the results\" I will upgrade to Competition Expert Level! WhoooHoooo party time!🎉 &nbsp;or sleep time now🤠! </p>\n<p>Keep Learning everyone and thank you!&nbsp;</p>",
  "messages": [
    {
      "id": "1690537",
      "postDate": "02/15/2022 01:51:44",
      "content": "<p>Hi all!<br>\nCongratulation to all the winners and learners! </p>\n<p>It's 3:50 AM in the morning my time so I am a bit sleepy but super happy with my results! </p>\n<p>My best solution got me 0.678 which would get me 0.692 if I selected better,&nbsp; it is a very simple idea (at least for my humble experience !) </p>\n<p>I used a YOLOv5 shared model (which was better than my own) but ensembled it with itself (on different resolutions) - high, medium, and low + tracking</p>\n<p>The difference between my current place and the silver zone place that I would have got is just a small fraction in tunning image sizes. </p>\n<p>One of the things that I was excited about and did work (but not my best score) was the layer of binary classification.&nbsp;I will share the training code once I have the time. </p>\n<p>Thanks to all of you and a special thank you go to <a href=\"https://www.kaggle.com/cdeotte\" target=\"_blank\">@cdeotte</a> for his life-saving advice in \"pinning the original environment\" when I had the norfair issue. </p>\n<p>Also, a special thank you to <a href=\"https://www.kaggle.com/freshair1996\" target=\"_blank\">@freshair1996</a> for publically sharing the model, it was better than all the training I did on my limited hardware. </p>\n<p>Also, a special thank you to <a href=\"https://www.kaggle.com/alexteboul\" target=\"_blank\">@alexteboul</a> for providing the \"COTS v NotCOTS Cropped Crown of Thorns\" dataset which I used in my binary classification. </p>\n<p>And finally \"once the host approves the results\" I will upgrade to Competition Expert Level! WhoooHoooo party time!🎉 &nbsp;or sleep time now🤠! </p>\n<p>Keep Learning everyone and thank you!&nbsp;</p>",
      "rawMarkdown": "Hi all!\nCongratulation to all the winners and learners! \n\nIt's 3:50 AM in the morning my time so I am a bit sleepy but super happy with my results! \n\nMy best solution got me 0.678 which would get me 0.692 if I selected better,  it is a very simple idea (at least for my humble experience !) \n\nI used a YOLOv5 shared model (which was better than my own) but ensembled it with itself (on different resolutions) - high, medium, and low + tracking\n\nThe difference between my current place and the silver zone place that I would have got is just a small fraction in tunning image sizes. \n\nOne of the things that I was excited about and did work (but not my best score) was the layer of binary classification. I will share the training code once I have the time. \n\nThanks to all of you and a special thank you go to @cdeotte for his life-saving advice in \"pinning the original environment\" when I had the norfair issue. \n\nAlso, a special thank you to @freshair1996 for publically sharing the model, it was better than all the training I did on my limited hardware. \n\nAlso, a special thank you to @alexteboul for providing the \"COTS v NotCOTS Cropped Crown of Thorns\" dataset which I used in my binary classification. \n\nAnd finally \"once the host approves the results\" I will upgrade to Competition Expert Level! WhoooHoooo party time!🎉  or sleep time now🤠! \n\nKeep Learning everyone and thank you!",
      "votes": null
    },
    {
      "id": "1690777",
      "postDate": "02/15/2022 05:21:19",
      "content": "<p>It's a meaningful journey for us. I am happy to see that my best single model can let some kagglers (with or without hardware) to develop their creativity.</p>",
      "rawMarkdown": "It's a meaningful journey for us. I am happy to see that my best single model can let some kagglers (with or without hardware) to develop their creativity.",
      "votes": null
    },
    {
      "id": "1690880",
      "postDate": "02/15/2022 06:24:05",
      "content": "<p>thank you for your sharing.</p>",
      "rawMarkdown": "thank you for your sharing.",
      "votes": null
    },
    {
      "id": "1691001",
      "postDate": "02/15/2022 07:24:25",
      "content": "<p>haha，a nice model finished by all of us！it seems get some good scores （with some tricks on it as I have said）.Enjoy it! 😆😆</p>",
      "rawMarkdown": "haha，a nice model finished by all of us！it seems get some good scores （with some tricks on it as I have said）.Enjoy it! 😆😆",
      "votes": null
    },
    {
      "id": "1691388",
      "postDate": "02/15/2022 11:39:56",
      "content": "<p>Yes it is a nice model, Thank you for sharing :-)</p>",
      "rawMarkdown": "Yes it is a nice model, Thank you for sharing :-)",
      "votes": null
    },
    {
      "id": "1691407",
      "postDate": "02/15/2022 11:53:31",
      "content": "<p>Haha！😬😬🤝🤝</p>",
      "rawMarkdown": "Haha！😬😬🤝🤝",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1690777,
      "author_name": "freshair1996",
      "author_url": "",
      "post_date": "02/15/2022 05:21:19",
      "content": "<p>It's a meaningful journey for us. I am happy to see that my best single model can let some kagglers (with or without hardware) to develop their creativity.</p>",
      "votes": null,
      "replies": [
        {
          "id": 1690880,
          "author_name": "dragonzhang",
          "author_url": "",
          "post_date": "02/15/2022 06:24:05",
          "content": "<p>thank you for your sharing.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1691001,
          "author_name": "freshair1996",
          "author_url": "",
          "post_date": "02/15/2022 07:24:25",
          "content": "<p>haha，a nice model finished by all of us！it seems get some good scores （with some tricks on it as I have said）.Enjoy it! 😆😆</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1691388,
          "author_name": "asalhi",
          "author_url": "",
          "post_date": "02/15/2022 11:39:56",
          "content": "<p>Yes it is a nice model, Thank you for sharing :-)</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1691407,
          "author_name": "freshair1996",
          "author_url": "",
          "post_date": "02/15/2022 11:53:31",
          "content": "<p>Haha！😬😬🤝🤝</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1690537": "Hi all!\nCongratulation to all the winners and learners! \n\nIt's 3:50 AM in the morning my time so I am a bit sleepy but super happy with my results! \n\nMy best solution got me 0.678 which would get me 0.692 if I selected better,  it is a very simple idea (at least for my humble experience !) \n\nI used a YOLOv5 shared model (which was better than my own) but ensembled it with itself (on different resolutions) - high, medium, and low + tracking\n\nThe difference between my current place and the silver zone place that I would have got is just a small fraction in tunning image sizes. \n\nOne of the things that I was excited about and did work (but not my best score) was the layer of binary classification. I will share the training code once I have the time. \n\nThanks to all of you and a special thank you go to @cdeotte for his life-saving advice in \"pinning the original environment\" when I had the norfair issue. \n\nAlso, a special thank you to @freshair1996 for publically sharing the model, it was better than all the training I did on my limited hardware. \n\nAlso, a special thank you to @alexteboul for providing the \"COTS v NotCOTS Cropped Crown of Thorns\" dataset which I used in my binary classification. \n\nAnd finally \"once the host approves the results\" I will upgrade to Competition Expert Level! WhoooHoooo party time!🎉  or sleep time now🤠! \n\nKeep Learning everyone and thank you!",
    "1690777": "It's a meaningful journey for us. I am happy to see that my best single model can let some kagglers (with or without hardware) to develop their creativity.",
    "1690880": "thank you for your sharing.",
    "1691001": "haha，a nice model finished by all of us！it seems get some good scores （with some tricks on it as I have said）.Enjoy it! 😆😆",
    "1691388": "Yes it is a nice model, Thank you for sharing :-)",
    "1691407": "Haha！😬😬🤝🤝"
  },
  "source": "meta"
}