{
  "id": 39530,
  "title": "Need to know if we can add our own target labels",
  "url": "/competitions/passenger-screening-algorithm-challenge/discussion/39530",
  "author_name": "",
  "post_date": "2017-09-15T18:40:43.642517800Z",
  "votes": 2,
  "comment_count": 3,
  "views": 0,
  "content": "<p>Are we allowed to go go through each sample and add more labels, E.G. hand-label the pixel-level location of body parts?  </p>\n\n<p>The rules say we can only use \"publicly available external data\". </p>\n\n<ul>\n<li>Would the new labels be \"external data\"? </li>\n<li>The rules also say we're not allowed to share the dataset. Would external labels not be allowed because we can't make them \"publicly available\"? </li>\n<li>If it is allowed could it be unfair to people who don't have access to the modified dataset?</li>\n</ul>\n\n<p>I feel this is a pretty important issue. I see that it's been asked before but I can't find an official answer.</p>",
  "messages": [
    {
      "id": "221587",
      "postDate": "09/15/2017 18:40:43",
      "content": "<p>Are we allowed to go go through each sample and add more labels, E.G. hand-label the pixel-level location of body parts?  </p>\n\n<p>The rules say we can only use \"publicly available external data\". </p>\n\n<ul>\n<li>Would the new labels be \"external data\"? </li>\n<li>The rules also say we're not allowed to share the dataset. Would external labels not be allowed because we can't make them \"publicly available\"? </li>\n<li>If it is allowed could it be unfair to people who don't have access to the modified dataset?</li>\n</ul>\n\n<p>I feel this is a pretty important issue. I see that it's been asked before but I can't find an official answer.</p>",
      "rawMarkdown": "Are we allowed to go go through each sample and add more labels, E.G. hand-label the pixel-level location of body parts?  \n\nThe rules say we can only use \"publicly available external data\". \n\n - Would the new labels be \"external data\"? \n - The rules also say we're not allowed to share the dataset. Would external labels not be allowed because we can't make them \"publicly available\"? \n - If it is allowed could it be unfair to people who don't have access to the modified dataset?\n\nI feel this is a pretty important issue. I see that it's been asked before but I can't find an official answer.",
      "votes": null
    },
    {
      "id": "221606",
      "postDate": "09/15/2017 19:54:56",
      "content": "<p>You <em>may</em> hand annotate the training set. This is not considered external data and thus does not have to be shared.</p>\n\n<p>You <em>may not</em> hand annotate the test set. Your algorithm must be capable of predicting probabilities for a given image without human intervention.</p>",
      "rawMarkdown": "You *may* hand annotate the training set. This is not considered external data and thus does not have to be shared.\n\nYou *may not* hand annotate the test set. Your algorithm must be capable of predicting probabilities for a given image without human intervention.",
      "votes": null
    },
    {
      "id": "255012",
      "postDate": "12/08/2017 03:34:24",
      "content": "<p>Hi William, </p>\n\n<p>Following up on this question, since stage 1 test labels will be released for stage 2, then in this case the definition of \"test data\" as you have used it here means stage 2 test data and not stage 1 test data right? </p>\n\n<p>In other words, for stage 2, I believe we should be able to treat all stage 1 data (train &amp; test) all as training data, correct?</p>\n\n<blockquote>\n  <p><strong>William Cukierski wrote</strong></p>\n  \n  <blockquote>\n    <p>You <em>may</em> hand annotate the training set. This is not considered external data and thus does not have to be shared.</p>\n  </blockquote>\n  \n  <p>You <em>may not</em> hand annotate the test set. Your algorithm must be capable of predicting probabilities for a given image without human intervention.</p>\n</blockquote>",
      "rawMarkdown": "Hi William, \n\nFollowing up on this question, since stage 1 test labels will be released for stage 2, then in this case the definition of \"test data\" as you have used it here means stage 2 test data and not stage 1 test data right? \n\nIn other words, for stage 2, I believe we should be able to treat all stage 1 data (train &amp; test) all as training data, correct?\n\n&gt; **William Cukierski wrote**\n&gt; \n&gt; &gt; You *may* hand annotate the training set. This is not considered external data and thus does not have to be shared.\n&gt; \n&gt; You *may not* hand annotate the test set. Your algorithm must be capable of predicting probabilities for a given image without human intervention.",
      "votes": null
    },
    {
      "id": "255173",
      "postDate": "12/08/2017 14:37:47",
      "content": "<p>That's correct James.</p>",
      "rawMarkdown": "That's correct James.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 221606,
      "author_name": "wcukierski",
      "author_url": "",
      "post_date": "09/15/2017 19:54:56",
      "content": "<p>You <em>may</em> hand annotate the training set. This is not considered external data and thus does not have to be shared.</p>\n\n<p>You <em>may not</em> hand annotate the test set. Your algorithm must be capable of predicting probabilities for a given image without human intervention.</p>",
      "votes": null,
      "replies": [
        {
          "id": 255012,
          "author_name": "jamesrequa",
          "author_url": "",
          "post_date": "12/08/2017 03:34:24",
          "content": "<p>Hi William, </p>\n\n<p>Following up on this question, since stage 1 test labels will be released for stage 2, then in this case the definition of \"test data\" as you have used it here means stage 2 test data and not stage 1 test data right? </p>\n\n<p>In other words, for stage 2, I believe we should be able to treat all stage 1 data (train &amp; test) all as training data, correct?</p>\n\n<blockquote>\n  <p><strong>William Cukierski wrote</strong></p>\n  \n  <blockquote>\n    <p>You <em>may</em> hand annotate the training set. This is not considered external data and thus does not have to be shared.</p>\n  </blockquote>\n  \n  <p>You <em>may not</em> hand annotate the test set. Your algorithm must be capable of predicting probabilities for a given image without human intervention.</p>\n</blockquote>",
          "votes": null,
          "replies": []
        },
        {
          "id": 255173,
          "author_name": "wcukierski",
          "author_url": "",
          "post_date": "12/08/2017 14:37:47",
          "content": "<p>That's correct James.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "221587": "Are we allowed to go go through each sample and add more labels, E.G. hand-label the pixel-level location of body parts?  \n\nThe rules say we can only use \"publicly available external data\". \n\n - Would the new labels be \"external data\"? \n - The rules also say we're not allowed to share the dataset. Would external labels not be allowed because we can't make them \"publicly available\"? \n - If it is allowed could it be unfair to people who don't have access to the modified dataset?\n\nI feel this is a pretty important issue. I see that it's been asked before but I can't find an official answer.",
    "221606": "You *may* hand annotate the training set. This is not considered external data and thus does not have to be shared.\n\nYou *may not* hand annotate the test set. Your algorithm must be capable of predicting probabilities for a given image without human intervention.",
    "255012": "Hi William, \n\nFollowing up on this question, since stage 1 test labels will be released for stage 2, then in this case the definition of \"test data\" as you have used it here means stage 2 test data and not stage 1 test data right? \n\nIn other words, for stage 2, I believe we should be able to treat all stage 1 data (train &amp; test) all as training data, correct?\n\n&gt; **William Cukierski wrote**\n&gt; \n&gt; &gt; You *may* hand annotate the training set. This is not considered external data and thus does not have to be shared.\n&gt; \n&gt; You *may not* hand annotate the test set. Your algorithm must be capable of predicting probabilities for a given image without human intervention.",
    "255173": "That's correct James."
  },
  "source": "meta"
}