{
  "id": 67079,
  "title": "Information Contained in Stroke Order and Length",
  "url": "/competitions/quickdraw-doodle-recognition/discussion/67079",
  "author_name": "Simmo",
  "post_date": "2018-09-28T13:58:51.726000",
  "votes": 2,
  "comment_count": 3,
  "views": 0,
  "content": "<p>Do you think there is a significant amount of information contained in the length and order of the strokes? </p>\n\n<p>Perhaps from the length and order of the first few stroked you can make an educated guess of what the final class will be, or at least refine the prediction coming out of your CNN? This could probably be done with an LSTM or something similar in conjunction with a CNN?</p>",
  "messages": [
    {
      "id": 395399,
      "postDate": "2018-09-28T13:58:51.727Z",
      "content": "<p>Do you think there is a significant amount of information contained in the length and order of the strokes? </p>\n\n<p>Perhaps from the length and order of the first few stroked you can make an educated guess of what the final class will be, or at least refine the prediction coming out of your CNN? This could probably be done with an LSTM or something similar in conjunction with a CNN?</p>",
      "rawMarkdown": "Do you think there is a significant amount of information contained in the length and order of the strokes? \n\nPerhaps from the length and order of the first few stroked you can make an educated guess of what the final class will be, or at least refine the prediction coming out of your CNN? This could probably be done with an LSTM or something similar in conjunction with a CNN?",
      "votes": 2
    },
    {
      "id": 395452,
      "postDate": "2018-09-28T15:53:12.443Z",
      "content": "<p>Length and order of strokes is full of noise related to mouse use: manual dexterity, pointer speed, mouse pad size, workspace clutter and ergonomics, etc.</p>\n\n<p>OTOH, most people probably draw in top down fashion, starting with the outline and filling the details later. In such case, the first couple of long strokes could be more essential to object recognition.</p>",
      "rawMarkdown": "Length and order of strokes is full of noise related to mouse use: manual dexterity, pointer speed, mouse pad size, workspace clutter and ergonomics, etc.\n\nOTOH, most people probably draw in top down fashion, starting with the outline and filling the details later. In such case, the first couple of long strokes could be more essential to object recognition.",
      "votes": 2,
      "replies": [
        {
          "id": 395504,
          "postDate": "2018-09-28T17:45:23.293Z",
          "content": "<p>Is there any advantage though in us thinking of the problem in this way given that we already have the full set of strokes? The full shape will always be more descriptive than the first couple of lines</p>",
          "rawMarkdown": "Is there any advantage though in us thinking of the problem in this way given that we already have the full set of strokes? The full shape will always be more descriptive than the first couple of lines"
        },
        {
          "id": 395572,
          "postDate": "2018-09-28T21:05:36.337Z",
          "content": "<p>There may be some advantage if your shape analysis software is not perfect in distinguishing an outline (general shape) of an object from the details. I think, I may take advantage of it.</p>",
          "rawMarkdown": "There may be some advantage if your shape analysis software is not perfect in distinguishing an outline (general shape) of an object from the details. I think, I may take advantage of it."
        }
      ]
    }
  ],
  "comments": [
    {
      "id": 395452,
      "author_name": "Paul Jurczak",
      "author_url": "",
      "post_date": "2018-09-28T15:53:12.443000",
      "content": "<p>Length and order of strokes is full of noise related to mouse use: manual dexterity, pointer speed, mouse pad size, workspace clutter and ergonomics, etc.</p>\n\n<p>OTOH, most people probably draw in top down fashion, starting with the outline and filling the details later. In such case, the first couple of long strokes could be more essential to object recognition.</p>",
      "votes": 2,
      "replies": [
        {
          "id": 395504,
          "author_name": "Simmo",
          "author_url": "",
          "post_date": "2018-09-28T17:45:23.293000",
          "content": "<p>Is there any advantage though in us thinking of the problem in this way given that we already have the full set of strokes? The full shape will always be more descriptive than the first couple of lines</p>",
          "votes": 0,
          "replies": []
        },
        {
          "id": 395572,
          "author_name": "Paul Jurczak",
          "author_url": "",
          "post_date": "2018-09-28T21:05:36.337000",
          "content": "<p>There may be some advantage if your shape analysis software is not perfect in distinguishing an outline (general shape) of an object from the details. I think, I may take advantage of it.</p>",
          "votes": 0,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "395399": "Do you think there is a significant amount of information contained in the length and order of the strokes? \n\nPerhaps from the length and order of the first few stroked you can make an educated guess of what the final class will be, or at least refine the prediction coming out of your CNN? This could probably be done with an LSTM or something similar in conjunction with a CNN?",
    "395452": "Length and order of strokes is full of noise related to mouse use: manual dexterity, pointer speed, mouse pad size, workspace clutter and ergonomics, etc.\n\nOTOH, most people probably draw in top down fashion, starting with the outline and filling the details later. In such case, the first couple of long strokes could be more essential to object recognition."
  }
}