{
  "id": 209492,
  "title": "Using Erosion and Thresholds to Improve MNIST Accuracy",
  "url": "/competitions/digit-recognizer/discussion/209492",
  "author_name": "SnowyOwl",
  "post_date": "2021-01-07T18:05:34.972000",
  "votes": 0,
  "comment_count": 0,
  "views": null,
  "content": "<p>Hello,</p>\n<p>I've made a new notebook on using erosion to improve MNIST Accuracy. It includes a pre-processing function that erodes MNIST images, to get rid of any noise/messy lines on the images, and then cleans them up.</p>\n<p>If I get enough interest/upvotes, I can publish a function/class to add for a tensorflow/pytorch transform layer.</p>\n<p>Link: <a href=\"https://www.kaggle.com/kyleberdy/basic-mnist-visualization\" target=\"_blank\">https://www.kaggle.com/kyleberdy/basic-mnist-visualization</a></p>",
  "messages": [
    {
      "id": 1143011,
      "postDate": "2021-01-07T18:05:34.973Z",
      "content": "<p>Hello,</p>\n<p>I've made a new notebook on using erosion to improve MNIST Accuracy. It includes a pre-processing function that erodes MNIST images, to get rid of any noise/messy lines on the images, and then cleans them up.</p>\n<p>If I get enough interest/upvotes, I can publish a function/class to add for a tensorflow/pytorch transform layer.</p>\n<p>Link: <a href=\"https://www.kaggle.com/kyleberdy/basic-mnist-visualization\" target=\"_blank\">https://www.kaggle.com/kyleberdy/basic-mnist-visualization</a></p>",
      "rawMarkdown": "Hello,\n\nI've made a new notebook on using erosion to improve MNIST Accuracy. It includes a pre-processing function that erodes MNIST images, to get rid of any noise/messy lines on the images, and then cleans them up.\n\nIf I get enough interest/upvotes, I can publish a function/class to add for a tensorflow/pytorch transform layer.\n\nLink: https://www.kaggle.com/kyleberdy/basic-mnist-visualization"
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "1143011": "Hello,\n\nI've made a new notebook on using erosion to improve MNIST Accuracy. It includes a pre-processing function that erodes MNIST images, to get rid of any noise/messy lines on the images, and then cleans them up.\n\nIf I get enough interest/upvotes, I can publish a function/class to add for a tensorflow/pytorch transform layer.\n\nLink: https://www.kaggle.com/kyleberdy/basic-mnist-visualization"
  }
}