{
  "id": 165644,
  "title": "What should I do? (About current issues)",
  "url": "/competitions/global-wheat-detection/discussion/165644",
  "author_name": "",
  "post_date": "2020-07-10T13:02:35.177784400Z",
  "votes": 1,
  "comment_count": 4,
  "views": 0,
  "content": "<p>Hi, all!</p>\n\n<p>Various issues seem to be complicated in this competition.\nI've read the discussion as much as possible, but I wasn't sure whether it was clear or not.</p>\n\n<p>So I'm going to write this to make it clearer one more time.</p>\n\n<h1>1. GPL-3 licence issues(e.g. Yolov5)</h1>\n\n<ul>\n<li>Use or Not?</li>\n<li>If I use some open sources have GPL-3 licence issues, can I get medals?\n(I am not interested in prizes. So, It is possible ['Leaderboard Paradox' ].)(<a href=\"https://www.kaggle.com/c/global-wheat-detection/discussion/165294\">https://www.kaggle.com/c/global-wheat-detection/discussion/165294</a>)</li>\n</ul>\n\n<h1>2. Large bounding boxes in data set</h1>\n\n<p>I heard that the private data set will be fixed.\n - Fix manually or Not in public data set?\n - I want to know what is the best way in this situation.</p>\n\n<p>I am worried about what to do.\nBecause of the above issues, LB is unreliable.</p>\n\n<p>There is no perfect solution, but I want to hear from others. 😂</p>",
  "messages": [
    {
      "id": "922967",
      "postDate": "07/10/2020 13:02:35",
      "content": "<p>Hi, all!</p>\n\n<p>Various issues seem to be complicated in this competition.\nI've read the discussion as much as possible, but I wasn't sure whether it was clear or not.</p>\n\n<p>So I'm going to write this to make it clearer one more time.</p>\n\n<h1>1. GPL-3 licence issues(e.g. Yolov5)</h1>\n\n<ul>\n<li>Use or Not?</li>\n<li>If I use some open sources have GPL-3 licence issues, can I get medals?\n(I am not interested in prizes. So, It is possible ['Leaderboard Paradox' ].)(<a href=\"https://www.kaggle.com/c/global-wheat-detection/discussion/165294\">https://www.kaggle.com/c/global-wheat-detection/discussion/165294</a>)</li>\n</ul>\n\n<h1>2. Large bounding boxes in data set</h1>\n\n<p>I heard that the private data set will be fixed.\n - Fix manually or Not in public data set?\n - I want to know what is the best way in this situation.</p>\n\n<p>I am worried about what to do.\nBecause of the above issues, LB is unreliable.</p>\n\n<p>There is no perfect solution, but I want to hear from others. 😂</p>",
      "rawMarkdown": "Hi, all!\n\nVarious issues seem to be complicated in this competition.\nI've read the discussion as much as possible, but I wasn't sure whether it was clear or not.\n\nSo I'm going to write this to make it clearer one more time.\n\n#1. GPL-3 licence issues(e.g. Yolov5)\n - Use or Not?\n - If I use some open sources have GPL-3 licence issues, can I get medals?\n   (I am not interested in prizes. So, It is possible ['Leaderboard Paradox' ].)(https://www.kaggle.com/c/global-wheat-detection/discussion/165294)\n \n\n#2. Large bounding boxes in data set\n   I heard that the private data set will be fixed.\n - Fix manually or Not in public data set?\n - I want to know what is the best way in this situation.\n\n\nI am worried about what to do.\nBecause of the above issues, LB is unreliable.\n\n\nThere is no perfect solution, but I want to hear from others. 😂",
      "votes": null
    },
    {
      "id": "923014",
      "postDate": "07/10/2020 13:36:18",
      "content": "<p>About the YoloV5 issue, as far as I understand, hosts said that ending up in 1st-3rd spot using YoloV5 will get you disqualifed because of it GPL License, but they did not declare what will happen to the other teams at the medal spots.\nI hope one of the hosts will clarify it soon so teams who aim to medals will know if they can use it or not.</p>",
      "rawMarkdown": "About the YoloV5 issue, as far as I understand, hosts said that ending up in 1st-3rd spot using YoloV5 will get you disqualifed because of it GPL License, but they did not declare what will happen to the other teams at the medal spots.\nI hope one of the hosts will clarify it soon so teams who aim to medals will know if they can use it or not.",
      "votes": null
    },
    {
      "id": "923019",
      "postDate": "07/10/2020 13:38:32",
      "content": "<p>I hope this will be resolved quickly. 😫 </p>",
      "rawMarkdown": "I hope this will be resolved quickly. 😫",
      "votes": null
    },
    {
      "id": "923020",
      "postDate": "07/10/2020 13:38:49",
      "content": "<ol>\n<li><p>For me, I already have a 5-fold Yolov5 model trained but I decided to not use them. Though I'm more interested in the medals, I feel like I should comply with the rules.</p></li>\n<li><p>You should always clean your data if possible. There's a saying like something along the line of \"garbage in, garbage out,\" and this is generally true for machine learning algorithms. Even if you don't have the time to clean all the incorrect annotations, do treat them differently in your pipeline.</p></li>\n</ol>",
      "rawMarkdown": "1. For me, I already have a 5-fold Yolov5 model trained but I decided to not use them. Though I'm more interested in the medals, I feel like I should comply with the rules.\n\n2. You should always clean your data if possible. There's a saying like something along the line of \"garbage in, garbage out,\" and this is generally true for machine learning algorithms. Even if you don't have the time to clean all the incorrect annotations, do treat them differently in your pipeline.",
      "votes": null
    },
    {
      "id": "923043",
      "postDate": "07/10/2020 13:52:41",
      "content": "<ol>\n<li>😂 (I am in a similar situation. orz)</li>\n<li>You are right. Dataset I or others have confirmed will be clean. However, I think that 'clean' can be different depending on the situation and sometimes a slightly wrong value can make sense.(e.g. If the host says that it cannot modify the private data.). And it is difficult to complete this all data clean perfectly, and it takes a lot of effort and time. if the model baseline needs to be re-established for some issues, This competition may be unfair not only for me, but for everyone. 😪 </li>\n</ol>",
      "rawMarkdown": "1. 😂 (I am in a similar situation. orz)\n2. You are right. Dataset I or others have confirmed will be clean. However, I think that 'clean' can be different depending on the situation and sometimes a slightly wrong value can make sense.(e.g. If the host says that it cannot modify the private data.). And it is difficult to complete this all data clean perfectly, and it takes a lot of effort and time. if the model baseline needs to be re-established for some issues, This competition may be unfair not only for me, but for everyone. 😪",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 923014,
      "author_name": "dudkmanron",
      "author_url": "",
      "post_date": "07/10/2020 13:36:18",
      "content": "<p>About the YoloV5 issue, as far as I understand, hosts said that ending up in 1st-3rd spot using YoloV5 will get you disqualifed because of it GPL License, but they did not declare what will happen to the other teams at the medal spots.\nI hope one of the hosts will clarify it soon so teams who aim to medals will know if they can use it or not.</p>",
      "votes": null,
      "replies": [
        {
          "id": 923019,
          "author_name": "piantic",
          "author_url": "",
          "post_date": "07/10/2020 13:38:32",
          "content": "<p>I hope this will be resolved quickly. 😫 </p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 923020,
      "author_name": "alexanderliao",
      "author_url": "",
      "post_date": "07/10/2020 13:38:49",
      "content": "<ol>\n<li><p>For me, I already have a 5-fold Yolov5 model trained but I decided to not use them. Though I'm more interested in the medals, I feel like I should comply with the rules.</p></li>\n<li><p>You should always clean your data if possible. There's a saying like something along the line of \"garbage in, garbage out,\" and this is generally true for machine learning algorithms. Even if you don't have the time to clean all the incorrect annotations, do treat them differently in your pipeline.</p></li>\n</ol>",
      "votes": null,
      "replies": [
        {
          "id": 923043,
          "author_name": "piantic",
          "author_url": "",
          "post_date": "07/10/2020 13:52:41",
          "content": "<ol>\n<li>😂 (I am in a similar situation. orz)</li>\n<li>You are right. Dataset I or others have confirmed will be clean. However, I think that 'clean' can be different depending on the situation and sometimes a slightly wrong value can make sense.(e.g. If the host says that it cannot modify the private data.). And it is difficult to complete this all data clean perfectly, and it takes a lot of effort and time. if the model baseline needs to be re-established for some issues, This competition may be unfair not only for me, but for everyone. 😪 </li>\n</ol>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "922967": "Hi, all!\n\nVarious issues seem to be complicated in this competition.\nI've read the discussion as much as possible, but I wasn't sure whether it was clear or not.\n\nSo I'm going to write this to make it clearer one more time.\n\n#1. GPL-3 licence issues(e.g. Yolov5)\n - Use or Not?\n - If I use some open sources have GPL-3 licence issues, can I get medals?\n   (I am not interested in prizes. So, It is possible ['Leaderboard Paradox' ].)(https://www.kaggle.com/c/global-wheat-detection/discussion/165294)\n \n\n#2. Large bounding boxes in data set\n   I heard that the private data set will be fixed.\n - Fix manually or Not in public data set?\n - I want to know what is the best way in this situation.\n\n\nI am worried about what to do.\nBecause of the above issues, LB is unreliable.\n\n\nThere is no perfect solution, but I want to hear from others. 😂",
    "923014": "About the YoloV5 issue, as far as I understand, hosts said that ending up in 1st-3rd spot using YoloV5 will get you disqualifed because of it GPL License, but they did not declare what will happen to the other teams at the medal spots.\nI hope one of the hosts will clarify it soon so teams who aim to medals will know if they can use it or not.",
    "923019": "I hope this will be resolved quickly. 😫",
    "923020": "1. For me, I already have a 5-fold Yolov5 model trained but I decided to not use them. Though I'm more interested in the medals, I feel like I should comply with the rules.\n\n2. You should always clean your data if possible. There's a saying like something along the line of \"garbage in, garbage out,\" and this is generally true for machine learning algorithms. Even if you don't have the time to clean all the incorrect annotations, do treat them differently in your pipeline.",
    "923043": "1. 😂 (I am in a similar situation. orz)\n2. You are right. Dataset I or others have confirmed will be clean. However, I think that 'clean' can be different depending on the situation and sometimes a slightly wrong value can make sense.(e.g. If the host says that it cannot modify the private data.). And it is difficult to complete this all data clean perfectly, and it takes a lot of effort and time. if the model baseline needs to be re-established for some issues, This competition may be unfair not only for me, but for everyone. 😪"
  },
  "source": "meta"
}