{
  "id": 54348,
  "title": "Strange challenge",
  "url": "/competitions/data-science-bowl-2018/discussion/54348",
  "author_name": "",
  "post_date": "2018-04-12T11:17:06.541523900Z",
  "votes": 5,
  "comment_count": 4,
  "views": 0,
  "content": "<ul>\n<li>670 images in stage1 training data</li>\n<li>67 images in stage 1 test data</li>\n<li>3019 images in stage 2 test data !!!</li>\n</ul>\n\n<p>I m really wondering what is going on here. I have participated in many competitions here and I got some medails.  Some where very challenging, some were a bit strange.  But this challenge is really strange with so less training data and so much test data in stage 2.  it is difficult to create a accurate and robust model with less training data.  </p>\n\n<p>I spend some time in this challenge with UNets and Mask RCNNs but not as much as some of the leaders and I really hope that they will perform well on the new quite different stage 2 test data as well.</p>\n\n<p>Luckily I  not spend so much effort in this challenge and I learned a lot about Mask RCNNS but I not sure that I will participate again in a Data Science Bowl.</p>",
  "messages": [
    {
      "id": "312807",
      "postDate": "04/12/2018 11:17:06",
      "content": "<ul>\n<li>670 images in stage1 training data</li>\n<li>67 images in stage 1 test data</li>\n<li>3019 images in stage 2 test data !!!</li>\n</ul>\n\n<p>I m really wondering what is going on here. I have participated in many competitions here and I got some medails.  Some where very challenging, some were a bit strange.  But this challenge is really strange with so less training data and so much test data in stage 2.  it is difficult to create a accurate and robust model with less training data.  </p>\n\n<p>I spend some time in this challenge with UNets and Mask RCNNs but not as much as some of the leaders and I really hope that they will perform well on the new quite different stage 2 test data as well.</p>\n\n<p>Luckily I  not spend so much effort in this challenge and I learned a lot about Mask RCNNS but I not sure that I will participate again in a Data Science Bowl.</p>",
      "rawMarkdown": "670 images in stage1 training data\n - 67 images in stage 1 test data\n - 3019 images in stage 2 test data !!!\n\nI m really wondering what is going on here. I have participated in many competitions here and I got some medails.  Some where very challenging, some were a bit strange.  But this challenge is really strange with so less training data and so much test data in stage 2.  it is difficult to create a accurate and robust model with less training data.  \n\nI spend some time in this challenge with UNets and Mask RCNNs but not as much as some of the leaders and I really hope that they will perform well on the new quite different stage 2 test data as well.\n\nLuckily I  not spend so much effort in this challenge and I learned a lot about Mask RCNNS but I not sure that I will participate again in a Data Science Bowl.",
      "votes": null
    },
    {
      "id": "312812",
      "postDate": "04/12/2018 11:33:06",
      "content": "<p>I think the description in “Data Description” section might be concerned.</p>\n\n<blockquote>\n  <p>To deter hand labeling, it will also contain images that are ignored in scoring. </p>\n</blockquote>\n\n<p>This probably means that not all of 3019 images have ground truth.</p>",
      "rawMarkdown": "I think the description in “Data Description” section might be concerned.\n\n&gt; To deter hand labeling, it will also contain images that are ignored in scoring. \n\n This probably means that not all of 3019 images have ground truth.",
      "votes": null
    },
    {
      "id": "312815",
      "postDate": "04/12/2018 11:38:34",
      "content": "<p>I m aware of that - it is really not my first challenge.  But normally you have a much larger test set from the very beginning with a lot of \"fake\" data in to prevent hand labeling.  And this stage 2 test data contains a number of images which are very different from the stage 1 images</p>",
      "rawMarkdown": "I m aware of that - it is really not my first challenge.  But normally you have a much larger test set from the very beginning with a lot of \"fake\" data in to prevent hand labeling.  And this stage 2 test data contains a number of images which are very different from the stage 1 images",
      "votes": null
    },
    {
      "id": "313044",
      "postDate": "04/12/2018 17:43:27",
      "content": "<p>And not as many and not as different as I've expected. </p>",
      "rawMarkdown": "And not as many and not as different as I've expected.",
      "votes": null
    },
    {
      "id": "313319",
      "postDate": "04/13/2018 05:24:30",
      "content": "<p>Just learning reading Heng's post makes time spent on this challenge extremely valuable IMHO.</p>",
      "rawMarkdown": "Just learning reading Heng's post makes time spent on this challenge extremely valuable IMHO.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 312812,
      "author_name": "tepppeikanayama",
      "author_url": "",
      "post_date": "04/12/2018 11:33:06",
      "content": "<p>I think the description in “Data Description” section might be concerned.</p>\n\n<blockquote>\n  <p>To deter hand labeling, it will also contain images that are ignored in scoring. </p>\n</blockquote>\n\n<p>This probably means that not all of 3019 images have ground truth.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 312815,
      "author_name": "thomastilli",
      "author_url": "",
      "post_date": "04/12/2018 11:38:34",
      "content": "<p>I m aware of that - it is really not my first challenge.  But normally you have a much larger test set from the very beginning with a lot of \"fake\" data in to prevent hand labeling.  And this stage 2 test data contains a number of images which are very different from the stage 1 images</p>",
      "votes": null,
      "replies": [
        {
          "id": 313044,
          "author_name": "thenickname",
          "author_url": "",
          "post_date": "04/12/2018 17:43:27",
          "content": "<p>And not as many and not as different as I've expected. </p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 313319,
      "author_name": "cpmpml",
      "author_url": "",
      "post_date": "04/13/2018 05:24:30",
      "content": "<p>Just learning reading Heng's post makes time spent on this challenge extremely valuable IMHO.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "312807": "670 images in stage1 training data\n - 67 images in stage 1 test data\n - 3019 images in stage 2 test data !!!\n\nI m really wondering what is going on here. I have participated in many competitions here and I got some medails.  Some where very challenging, some were a bit strange.  But this challenge is really strange with so less training data and so much test data in stage 2.  it is difficult to create a accurate and robust model with less training data.  \n\nI spend some time in this challenge with UNets and Mask RCNNs but not as much as some of the leaders and I really hope that they will perform well on the new quite different stage 2 test data as well.\n\nLuckily I  not spend so much effort in this challenge and I learned a lot about Mask RCNNS but I not sure that I will participate again in a Data Science Bowl.",
    "312812": "I think the description in “Data Description” section might be concerned.\n\n&gt; To deter hand labeling, it will also contain images that are ignored in scoring. \n\n This probably means that not all of 3019 images have ground truth.",
    "312815": "I m aware of that - it is really not my first challenge.  But normally you have a much larger test set from the very beginning with a lot of \"fake\" data in to prevent hand labeling.  And this stage 2 test data contains a number of images which are very different from the stage 1 images",
    "313044": "And not as many and not as different as I've expected.",
    "313319": "Just learning reading Heng's post makes time spent on this challenge extremely valuable IMHO."
  },
  "source": "meta"
}