{
  "id": 119724,
  "title": "Grey bar in some examples",
  "url": "/competitions/Kannada-MNIST/discussion/119724",
  "author_name": "",
  "post_date": "2019-11-30T22:25:05.066302600Z",
  "votes": null,
  "comment_count": 2,
  "views": 0,
  "content": "<p>I noticed in some of the examples I manually looked at that there are some random grey \"bars\" (if you look at them in gray scale) is this some artifact of data collection or intentionally added noise?</p>",
  "messages": [
    {
      "id": "685059",
      "postDate": "11/30/2019 22:25:05",
      "content": "<p>I noticed in some of the examples I manually looked at that there are some random grey \"bars\" (if you look at them in gray scale) is this some artifact of data collection or intentionally added noise?</p>",
      "rawMarkdown": "I noticed in some of the examples I manually looked at that there are some random grey \"bars\" (if you look at them in gray scale) is this some artifact of data collection or intentionally added noise?",
      "votes": null
    },
    {
      "id": "685061",
      "postDate": "11/30/2019 22:35:33",
      "content": "<p>Thanks for the question <a href=\"/nirvaux\">@nirvaux</a> !\nThese are the outlines of the sheet on which the numerals were authored. \nYou may see the raw scan images I have open-sourced here: <a href=\"https://github.com/vinayprabhu/Kannada_MNIST/blob/master/data/input_scans/\">https://github.com/vinayprabhu/Kannada_MNIST/blob/master/data/input_scans/</a>\nFor this version, I left them in the dataset in order to reflect the real-world noise effects, but you are also welcome to remove them in the pre-processing phase should you deem fit.\nHere are the notebooks that will walk you through the process:\n<a href=\"https://github.com/vinayprabhu/Kannada_MNIST/tree/master/colab_notebooks\">https://github.com/vinayprabhu/Kannada_MNIST/tree/master/colab_notebooks</a></p>",
      "rawMarkdown": "Thanks for the question @nirvaux !\nThese are the outlines of the sheet on which the numerals were authored. \nYou may see the raw scan images I have open-sourced here: https://github.com/vinayprabhu/Kannada_MNIST/blob/master/data/input_scans/\nFor this version, I left them in the dataset in order to reflect the real-world noise effects, but you are also welcome to remove them in the pre-processing phase should you deem fit.\nHere are the notebooks that will walk you through the process:\nhttps://github.com/vinayprabhu/Kannada_MNIST/tree/master/colab_notebooks",
      "votes": null
    },
    {
      "id": "685062",
      "postDate": "11/30/2019 22:42:35",
      "content": "<p><a href=\"/higgstachyon\">@higgstachyon</a> Ah I see. That makes sense why there will be the bars. Though, they don't seem to affect the performance (at least not that I can see). Also, it might be possible that the test set will have these artifacts so I'll leave them in since my models would need to be able to handle such noise. </p>\n\n<p>Anyway. Thanks for the quick reply! </p>",
      "rawMarkdown": "higgstachyon Ah I see. That makes sense why there will be the bars. Though, they don't seem to affect the performance (at least not that I can see). Also, it might be possible that the test set will have these artifacts so I'll leave them in since my models would need to be able to handle such noise. \n\nAnyway. Thanks for the quick reply!",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 685061,
      "author_name": "higgstachyon",
      "author_url": "",
      "post_date": "11/30/2019 22:35:33",
      "content": "<p>Thanks for the question <a href=\"/nirvaux\">@nirvaux</a> !\nThese are the outlines of the sheet on which the numerals were authored. \nYou may see the raw scan images I have open-sourced here: <a href=\"https://github.com/vinayprabhu/Kannada_MNIST/blob/master/data/input_scans/\">https://github.com/vinayprabhu/Kannada_MNIST/blob/master/data/input_scans/</a>\nFor this version, I left them in the dataset in order to reflect the real-world noise effects, but you are also welcome to remove them in the pre-processing phase should you deem fit.\nHere are the notebooks that will walk you through the process:\n<a href=\"https://github.com/vinayprabhu/Kannada_MNIST/tree/master/colab_notebooks\">https://github.com/vinayprabhu/Kannada_MNIST/tree/master/colab_notebooks</a></p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 685062,
      "author_name": "nirvaux",
      "author_url": "",
      "post_date": "11/30/2019 22:42:35",
      "content": "<p><a href=\"/higgstachyon\">@higgstachyon</a> Ah I see. That makes sense why there will be the bars. Though, they don't seem to affect the performance (at least not that I can see). Also, it might be possible that the test set will have these artifacts so I'll leave them in since my models would need to be able to handle such noise. </p>\n\n<p>Anyway. Thanks for the quick reply! </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "685059": "I noticed in some of the examples I manually looked at that there are some random grey \"bars\" (if you look at them in gray scale) is this some artifact of data collection or intentionally added noise?",
    "685061": "Thanks for the question @nirvaux !\nThese are the outlines of the sheet on which the numerals were authored. \nYou may see the raw scan images I have open-sourced here: https://github.com/vinayprabhu/Kannada_MNIST/blob/master/data/input_scans/\nFor this version, I left them in the dataset in order to reflect the real-world noise effects, but you are also welcome to remove them in the pre-processing phase should you deem fit.\nHere are the notebooks that will walk you through the process:\nhttps://github.com/vinayprabhu/Kannada_MNIST/tree/master/colab_notebooks",
    "685062": "higgstachyon Ah I see. That makes sense why there will be the bars. Though, they don't seem to affect the performance (at least not that I can see). Also, it might be possible that the test set will have these artifacts so I'll leave them in since my models would need to be able to handle such noise. \n\nAnyway. Thanks for the quick reply!"
  },
  "source": "meta"
}