{
  "id": 71094,
  "title": "Best acc of detection on no ship picture",
  "url": "/competitions/airbus-ship-detection/discussion/71094",
  "author_name": "",
  "post_date": "2018-11-10T04:12:35.607301700Z",
  "votes": 2,
  "comment_count": 14,
  "views": 0,
  "content": "<p>I am just curious of the best loc acc of dection on no ship picture. resolution, model, etc.\nFor me, 0.974 on 384 x 384 using Res34</p>",
  "messages": [
    {
      "id": "418536",
      "postDate": "11/10/2018 04:12:35",
      "content": "<p>I am just curious of the best loc acc of dection on no ship picture. resolution, model, etc.\nFor me, 0.974 on 384 x 384 using Res34</p>",
      "rawMarkdown": "I am just curious of the best loc acc of dection on no ship picture. resolution, model, etc.\nFor me, 0.974 on 384 x 384 using Res34",
      "votes": null
    },
    {
      "id": "418931",
      "postDate": "11/10/2018 22:39:06",
      "content": "<p>update: 0.983 on 384 * 384 using Res34</p>",
      "rawMarkdown": "update: 0.983 on 384 * 384 using Res34",
      "votes": null
    },
    {
      "id": "420093",
      "postDate": "11/13/2018 03:34:53",
      "content": "<p>no one answer my first discussion. T T...sad...0.0...QAQ...(╯‵□′)╯︵┴─┴ </p>",
      "rawMarkdown": "no one answer my first discussion. T T...sad...0.0...QAQ...(╯‵□′)╯︵┴─┴",
      "votes": null
    },
    {
      "id": "420096",
      "postDate": "11/13/2018 03:39:08",
      "content": "<p>haha，0.99+ on 768*768 </p>",
      "rawMarkdown": "haha，0.99+ on 768*768",
      "votes": null
    },
    {
      "id": "420100",
      "postDate": "11/13/2018 03:59:05",
      "content": "<p>Thank you for ur answering!!! good man!(○^～^○)∠※∠※</p>",
      "rawMarkdown": "Thank you for ur answering!!! good man!(○^～^○)∠※∠※",
      "votes": null
    },
    {
      "id": "420137",
      "postDate": "11/13/2018 05:44:42",
      "content": "<p>0.99+ on 384*384  lb 0.517, but the better lb of detection, the worse final submission.</p>",
      "rawMarkdown": "0.99+ on 384*384  lb 0.517, but the better lb of detection, the worse final submission.",
      "votes": null
    },
    {
      "id": "420140",
      "postDate": "11/13/2018 05:59:54",
      "content": "<p>Been experiencing this also. 0.984 cv classifier does better on lb than 0.99+ cv classifier</p>",
      "rawMarkdown": "Been experiencing this also. 0.984 cv classifier does better on lb than 0.99+ cv classifier",
      "votes": null
    },
    {
      "id": "420194",
      "postDate": "11/13/2018 08:40:26",
      "content": "<p>SeuTao, Is it on non-leak split?</p>",
      "rawMarkdown": "SeuTao, Is it on non-leak split?",
      "votes": null
    },
    {
      "id": "420217",
      "postDate": "11/13/2018 09:53:55",
      "content": "<p>You can try merge different classify model.</p>",
      "rawMarkdown": "You can try merge different classify model.",
      "votes": null
    },
    {
      "id": "420288",
      "postDate": "11/13/2018 12:07:37",
      "content": "<p>yes</p>",
      "rawMarkdown": "yes",
      "votes": null
    },
    {
      "id": "420344",
      "postDate": "11/13/2018 13:31:39",
      "content": "<p>I believe this is because when you submit an empty prediction you could get a good result if your model is good or if if predicts more empty masks in general. By submitting empty pred only you are only measuring recall whilst ignoring precision.</p>\n\n<p>A better way to measure perhaps is what lafoss said here <a href=\"https://www.kaggle.com/c/airbus-ship-detection/discussion/70221#413815\">https://www.kaggle.com/c/airbus-ship-detection/discussion/70221#413815</a></p>",
      "rawMarkdown": "I believe this is because when you submit an empty prediction you could get a good result if your model is good or if if predicts more empty masks in general. By submitting empty pred only you are only measuring recall whilst ignoring precision.\n\nA better way to measure perhaps is what lafoss said here https://www.kaggle.com/c/airbus-ship-detection/discussion/70221#413815",
      "votes": null
    },
    {
      "id": "420381",
      "postDate": "11/13/2018 14:24:34",
      "content": "<p>Thanks</p>",
      "rawMarkdown": "Thanks",
      "votes": null
    },
    {
      "id": "421398",
      "postDate": "11/15/2018 00:39:01",
      "content": "<p>Congratulations for the golden medal! but at the same time feel sorry for not in the money region. </p>",
      "rawMarkdown": "Congratulations for the golden medal! but at the same time feel sorry for not in the money region.",
      "votes": null
    },
    {
      "id": "421408",
      "postDate": "11/15/2018 00:53:16",
      "content": "<p>Thank u！It's really lucky for us to win a gold in this comp. Congratulations to u too!</p>",
      "rawMarkdown": "Thank u！It's really lucky for us to win a gold in this comp. Congratulations to u too!",
      "votes": null
    },
    {
      "id": "421419",
      "postDate": "11/15/2018 01:16:03",
      "content": "<p>acc 0.9917 on 384*384 resnet 34,5% validation considering BigImageId.</p>",
      "rawMarkdown": "acc 0.9917 on 384*384 resnet 34,5% validation considering BigImageId.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 418931,
      "author_name": "marcyuezhao",
      "author_url": "",
      "post_date": "11/10/2018 22:39:06",
      "content": "<p>update: 0.983 on 384 * 384 using Res34</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 420093,
      "author_name": "marcyuezhao",
      "author_url": "",
      "post_date": "11/13/2018 03:34:53",
      "content": "<p>no one answer my first discussion. T T...sad...0.0...QAQ...(╯‵□′)╯︵┴─┴ </p>",
      "votes": null,
      "replies": [
        {
          "id": 420096,
          "author_name": "shentao",
          "author_url": "",
          "post_date": "11/13/2018 03:39:08",
          "content": "<p>haha，0.99+ on 768*768 </p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 420100,
          "author_name": "marcyuezhao",
          "author_url": "",
          "post_date": "11/13/2018 03:59:05",
          "content": "<p>Thank you for ur answering!!! good man!(○^～^○)∠※∠※</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 420194,
          "author_name": "zfturbo",
          "author_url": "",
          "post_date": "11/13/2018 08:40:26",
          "content": "<p>SeuTao, Is it on non-leak split?</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 420288,
          "author_name": "shentao",
          "author_url": "",
          "post_date": "11/13/2018 12:07:37",
          "content": "<p>yes</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 421398,
          "author_name": "marcyuezhao",
          "author_url": "",
          "post_date": "11/15/2018 00:39:01",
          "content": "<p>Congratulations for the golden medal! but at the same time feel sorry for not in the money region. </p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 421408,
          "author_name": "shentao",
          "author_url": "",
          "post_date": "11/15/2018 00:53:16",
          "content": "<p>Thank u！It's really lucky for us to win a gold in this comp. Congratulations to u too!</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 420137,
      "author_name": "hdzheng",
      "author_url": "",
      "post_date": "11/13/2018 05:44:42",
      "content": "<p>0.99+ on 384*384  lb 0.517, but the better lb of detection, the worse final submission.</p>",
      "votes": null,
      "replies": [
        {
          "id": 420140,
          "author_name": "zitorelova",
          "author_url": "",
          "post_date": "11/13/2018 05:59:54",
          "content": "<p>Been experiencing this also. 0.984 cv classifier does better on lb than 0.99+ cv classifier</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 420217,
          "author_name": "hdzheng",
          "author_url": "",
          "post_date": "11/13/2018 09:53:55",
          "content": "<p>You can try merge different classify model.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 420344,
          "author_name": "arc144",
          "author_url": "",
          "post_date": "11/13/2018 13:31:39",
          "content": "<p>I believe this is because when you submit an empty prediction you could get a good result if your model is good or if if predicts more empty masks in general. By submitting empty pred only you are only measuring recall whilst ignoring precision.</p>\n\n<p>A better way to measure perhaps is what lafoss said here <a href=\"https://www.kaggle.com/c/airbus-ship-detection/discussion/70221#413815\">https://www.kaggle.com/c/airbus-ship-detection/discussion/70221#413815</a></p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 420381,
          "author_name": "hdzheng",
          "author_url": "",
          "post_date": "11/13/2018 14:24:34",
          "content": "<p>Thanks</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 421419,
      "author_name": "soonhwankwon",
      "author_url": "",
      "post_date": "11/15/2018 01:16:03",
      "content": "<p>acc 0.9917 on 384*384 resnet 34,5% validation considering BigImageId.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "418536": "I am just curious of the best loc acc of dection on no ship picture. resolution, model, etc.\nFor me, 0.974 on 384 x 384 using Res34",
    "418931": "update: 0.983 on 384 * 384 using Res34",
    "420093": "no one answer my first discussion. T T...sad...0.0...QAQ...(╯‵□′)╯︵┴─┴",
    "420096": "haha，0.99+ on 768*768",
    "420100": "Thank you for ur answering!!! good man!(○^～^○)∠※∠※",
    "420137": "0.99+ on 384*384  lb 0.517, but the better lb of detection, the worse final submission.",
    "420140": "Been experiencing this also. 0.984 cv classifier does better on lb than 0.99+ cv classifier",
    "420194": "SeuTao, Is it on non-leak split?",
    "420217": "You can try merge different classify model.",
    "420288": "yes",
    "420344": "I believe this is because when you submit an empty prediction you could get a good result if your model is good or if if predicts more empty masks in general. By submitting empty pred only you are only measuring recall whilst ignoring precision.\n\nA better way to measure perhaps is what lafoss said here https://www.kaggle.com/c/airbus-ship-detection/discussion/70221#413815",
    "420381": "Thanks",
    "421398": "Congratulations for the golden medal! but at the same time feel sorry for not in the money region.",
    "421408": "Thank u！It's really lucky for us to win a gold in this comp. Congratulations to u too!",
    "421419": "acc 0.9917 on 384*384 resnet 34,5% validation considering BigImageId."
  },
  "source": "meta"
}