{
  "id": 313120,
  "title": "New Data: Initial Impressions ",
  "url": "/competitions/ultra-mnist/discussion/313120",
  "author_name": "",
  "post_date": "2022-03-15T17:54:37.211623100Z",
  "votes": 1,
  "comment_count": 1,
  "views": 0,
  "content": "<p>The new Data seems a lot more complicated.</p>\n<p>Before we try to remove the background, i think its helpful to understand how the data was created in the first place.</p>\n<p>Here are my guesses/ Initial line of thoughts.</p>\n<p>The dataset seems like combination of <br>\n<strong>Randomly Linear transformed rectangular grids(seems like the same one as before) ⊕ Random Circles and triangles ⊕ MNIST digits</strong></p>\n<p>The trick to remove the noise would probably involve use of segmentation, measuring area or \"edginess of a contour?\"</p>\n<p>These are just my initial thoughts though, feel free to add in yours aswell!</p>",
  "messages": [
    {
      "id": "1723773",
      "postDate": "03/15/2022 17:54:37",
      "content": "<p>The new Data seems a lot more complicated.</p>\n<p>Before we try to remove the background, i think its helpful to understand how the data was created in the first place.</p>\n<p>Here are my guesses/ Initial line of thoughts.</p>\n<p>The dataset seems like combination of <br>\n<strong>Randomly Linear transformed rectangular grids(seems like the same one as before) ⊕ Random Circles and triangles ⊕ MNIST digits</strong></p>\n<p>The trick to remove the noise would probably involve use of segmentation, measuring area or \"edginess of a contour?\"</p>\n<p>These are just my initial thoughts though, feel free to add in yours aswell!</p>",
      "rawMarkdown": "The new Data seems a lot more complicated.\n\nBefore we try to remove the background, i think its helpful to understand how the data was created in the first place.\n\nHere are my guesses/ Initial line of thoughts.\n\nThe dataset seems like combination of \n**Randomly Linear transformed rectangular grids(seems like the same one as before) ⊕ Random Circles and triangles ⊕ MNIST digits**\n\nThe trick to remove the noise would probably involve use of segmentation, measuring area or \"edginess of a contour?\"\n\nThese are just my initial thoughts though, feel free to add in yours aswell!",
      "votes": null
    },
    {
      "id": "1723816",
      "postDate": "03/15/2022 18:46:42",
      "content": "<p><a href=\"https://www.kaggle.com/imams2000\" target=\"_blank\">@imams2000</a> Nice thoughts :)<br>\nI would recommend focusing more on the deep learning part of the problem, and trying to come up with a solution for such issues. Reverse engineering the background might help you figure out a way to counter our efforts, but that wouldn't really be a learning towards deep learning :)</p>",
      "rawMarkdown": "imams2000 Nice thoughts :)\nI would recommend focusing more on the deep learning part of the problem, and trying to come up with a solution for such issues. Reverse engineering the background might help you figure out a way to counter our efforts, but that wouldn't really be a learning towards deep learning :)",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1723816,
      "author_name": "dkgupta90",
      "author_url": "",
      "post_date": "03/15/2022 18:46:42",
      "content": "<p><a href=\"https://www.kaggle.com/imams2000\" target=\"_blank\">@imams2000</a> Nice thoughts :)<br>\nI would recommend focusing more on the deep learning part of the problem, and trying to come up with a solution for such issues. Reverse engineering the background might help you figure out a way to counter our efforts, but that wouldn't really be a learning towards deep learning :)</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1723773": "The new Data seems a lot more complicated.\n\nBefore we try to remove the background, i think its helpful to understand how the data was created in the first place.\n\nHere are my guesses/ Initial line of thoughts.\n\nThe dataset seems like combination of \n**Randomly Linear transformed rectangular grids(seems like the same one as before) ⊕ Random Circles and triangles ⊕ MNIST digits**\n\nThe trick to remove the noise would probably involve use of segmentation, measuring area or \"edginess of a contour?\"\n\nThese are just my initial thoughts though, feel free to add in yours aswell!",
    "1723816": "imams2000 Nice thoughts :)\nI would recommend focusing more on the deep learning part of the problem, and trying to come up with a solution for such issues. Reverse engineering the background might help you figure out a way to counter our efforts, but that wouldn't really be a learning towards deep learning :)"
  },
  "source": "meta"
}