{
  "id": 448419,
  "title": "Competition end - write up and compare some results",
  "url": "/competitions/denoising-shabby-pages/discussion/448419",
  "author_name": "Toon",
  "post_date": "2023-10-19T15:44:51.319000",
  "votes": 2,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Hi All,</p>\n<p>Since this competition has ended, i'm curious to know what routes were taken by you all. I myself had the best results with the Restormer architecture. My write up of my progress can be found here: <a href=\"https://toon-beerten.medium.com/denoising-and-reconstructing-dirty-documents-for-optimal-digitalization-ed3a186aa3d6\" target=\"_blank\">https://toon-beerten.medium.com/denoising-and-reconstructing-dirty-documents-for-optimal-digitalization-ed3a186aa3d6</a> .<br>\nWe saw that using postprocessing techniques like gaussian blur, the competition score could be improved. But that doesn't necessarily means a better OCR result in the end. That's why i'm very curious to see some actual examples from everybody.</p>\n<p>I propose to compare subjectively a few competition images. I chose these four:</p>\n<ul>\n<li>0080-lkgp_New-patient-registration-form-280920.pdf-2.png<br>\n0132-PAESTA-denying-apollo-moon-landings.pdf-15.png<br>\n0346-Erie_sPublicSchools_m_e_ch_20.pdf-16.png<br>\n0388-MGMResorts_truth-about-entertainment-infographic.pdf-1.png</li>\n</ul>\n<p>Left is original, right is my model output.</p>\n<table>\n<thead>\n<tr>\n<th><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8892775%2F1fc573f6b4b4a0800e80d785e0640745%2F0080-lkgp_New-patient-registration-form-280920.pdf-2.png?generation=1697729001969554&amp;alt=media\" alt=\"\"></th>\n<th><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8892775%2F8e712a194071e20cbe49d57b3cbea165%2F0080-lkgp_New-patient-registration-form-280920.pdf-2.png?generation=1697729568926852&amp;alt=media\" alt=\"\"></th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8892775%2Fca21f4335a834864ea655a907b9bec95%2F0132-PAESTA-denying-apollo-moon-landings.pdf-15.png?generation=1697729198258940&amp;alt=media\" alt=\"\"></td>\n<td><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8892775%2F827273271b408083953b4d7541cc9bed%2F0132-PAESTA-denying-apollo-moon-landings.pdf-15.png?generation=1697729550786680&amp;alt=media\" alt=\"\"></td>\n</tr>\n<tr>\n<td>---</td>\n<td>---</td>\n</tr>\n<tr>\n<td><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8892775%2F6278750b9374d18ba6a30546121e2eb5%2F0346-Erie_sPublicSchools_m_e_ch_20.pdf-16.png?generation=1697729361825485&amp;alt=media\" alt=\"\"></td>\n<td><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8892775%2F12da4dee093eceff17e9da7ca85a00e5%2F0346-Erie_sPublicSchools_m_e_ch_20.pdf-16.png?generation=1697729526483403&amp;alt=media\" alt=\"\"></td>\n</tr>\n<tr>\n<td>---</td>\n<td>---</td>\n</tr>\n<tr>\n<td><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8892775%2F0ce730ab8f86d0e1cbbef9a67daed7a1%2F0388-MGMResorts_truth-about-entertainment-infographic.pdf-1.png?generation=1697729470012196&amp;alt=media\" alt=\"\"></td>\n<td><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8892775%2F845fb173e3a5416545d0a47ee85a01b6%2F0388-MGMResorts_truth-about-entertainment-infographic.pdf-1.png?generation=1697729495744442&amp;alt=media\" alt=\"\"></td>\n</tr>\n</tbody>\n</table>\n<p>Hope to see a few results from the others as well!</p>",
  "messages": [
    {
      "id": 2488944,
      "postDate": "2023-10-19T15:44:51.320Z",
      "content": "<p>Hi All,</p>\n<p>Since this competition has ended, i'm curious to know what routes were taken by you all. I myself had the best results with the Restormer architecture. My write up of my progress can be found here: <a href=\"https://toon-beerten.medium.com/denoising-and-reconstructing-dirty-documents-for-optimal-digitalization-ed3a186aa3d6\" target=\"_blank\">https://toon-beerten.medium.com/denoising-and-reconstructing-dirty-documents-for-optimal-digitalization-ed3a186aa3d6</a> .<br>\nWe saw that using postprocessing techniques like gaussian blur, the competition score could be improved. But that doesn't necessarily means a better OCR result in the end. That's why i'm very curious to see some actual examples from everybody.</p>\n<p>I propose to compare subjectively a few competition images. I chose these four:</p>\n<ul>\n<li>0080-lkgp_New-patient-registration-form-280920.pdf-2.png<br>\n0132-PAESTA-denying-apollo-moon-landings.pdf-15.png<br>\n0346-Erie_sPublicSchools_m_e_ch_20.pdf-16.png<br>\n0388-MGMResorts_truth-about-entertainment-infographic.pdf-1.png</li>\n</ul>\n<p>Left is original, right is my model output.</p>\n<table>\n<thead>\n<tr>\n<th><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8892775%2F1fc573f6b4b4a0800e80d785e0640745%2F0080-lkgp_New-patient-registration-form-280920.pdf-2.png?generation=1697729001969554&amp;alt=media\" alt=\"\"></th>\n<th><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8892775%2F8e712a194071e20cbe49d57b3cbea165%2F0080-lkgp_New-patient-registration-form-280920.pdf-2.png?generation=1697729568926852&amp;alt=media\" alt=\"\"></th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8892775%2Fca21f4335a834864ea655a907b9bec95%2F0132-PAESTA-denying-apollo-moon-landings.pdf-15.png?generation=1697729198258940&amp;alt=media\" alt=\"\"></td>\n<td><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8892775%2F827273271b408083953b4d7541cc9bed%2F0132-PAESTA-denying-apollo-moon-landings.pdf-15.png?generation=1697729550786680&amp;alt=media\" alt=\"\"></td>\n</tr>\n<tr>\n<td>---</td>\n<td>---</td>\n</tr>\n<tr>\n<td><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8892775%2F6278750b9374d18ba6a30546121e2eb5%2F0346-Erie_sPublicSchools_m_e_ch_20.pdf-16.png?generation=1697729361825485&amp;alt=media\" alt=\"\"></td>\n<td><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8892775%2F12da4dee093eceff17e9da7ca85a00e5%2F0346-Erie_sPublicSchools_m_e_ch_20.pdf-16.png?generation=1697729526483403&amp;alt=media\" alt=\"\"></td>\n</tr>\n<tr>\n<td>---</td>\n<td>---</td>\n</tr>\n<tr>\n<td><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8892775%2F0ce730ab8f86d0e1cbbef9a67daed7a1%2F0388-MGMResorts_truth-about-entertainment-infographic.pdf-1.png?generation=1697729470012196&amp;alt=media\" alt=\"\"></td>\n<td><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8892775%2F845fb173e3a5416545d0a47ee85a01b6%2F0388-MGMResorts_truth-about-entertainment-infographic.pdf-1.png?generation=1697729495744442&amp;alt=media\" alt=\"\"></td>\n</tr>\n</tbody>\n</table>\n<p>Hope to see a few results from the others as well!</p>",
      "rawMarkdown": "Hi All,\n\nSince this competition has ended, i'm curious to know what routes were taken by you all. I myself had the best results with the Restormer architecture. My write up of my progress can be found here: https://toon-beerten.medium.com/denoising-and-reconstructing-dirty-documents-for-optimal-digitalization-ed3a186aa3d6 .\nWe saw that using postprocessing techniques like gaussian blur, the competition score could be improved. But that doesn't necessarily means a better OCR result in the end. That's why i'm very curious to see some actual examples from everybody.\n\nI propose to compare subjectively a few competition images. I chose these four:\n- 0080-lkgp_New-patient-registration-form-280920.pdf-2.png\n0132-PAESTA-denying-apollo-moon-landings.pdf-15.png\n0346-Erie_sPublicSchools_m_e_ch_20.pdf-16.png\n0388-MGMResorts_truth-about-entertainment-infographic.pdf-1.png\n\nLeft is original, right is my model output.\n\n| ![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8892775%2F1fc573f6b4b4a0800e80d785e0640745%2F0080-lkgp_New-patient-registration-form-280920.pdf-2.png?generation=1697729001969554&alt=media) | ![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8892775%2F8e712a194071e20cbe49d57b3cbea165%2F0080-lkgp_New-patient-registration-form-280920.pdf-2.png?generation=1697729568926852&alt=media) |\n| --- | --- |\n| ![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8892775%2Fca21f4335a834864ea655a907b9bec95%2F0132-PAESTA-denying-apollo-moon-landings.pdf-15.png?generation=1697729198258940&alt=media) | ![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8892775%2F827273271b408083953b4d7541cc9bed%2F0132-PAESTA-denying-apollo-moon-landings.pdf-15.png?generation=1697729550786680&alt=media) |\n| --- | --- |\n| ![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8892775%2F6278750b9374d18ba6a30546121e2eb5%2F0346-Erie_sPublicSchools_m_e_ch_20.pdf-16.png?generation=1697729361825485&alt=media) | ![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8892775%2F12da4dee093eceff17e9da7ca85a00e5%2F0346-Erie_sPublicSchools_m_e_ch_20.pdf-16.png?generation=1697729526483403&alt=media) |\n| --- | --- |\n| ![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8892775%2F0ce730ab8f86d0e1cbbef9a67daed7a1%2F0388-MGMResorts_truth-about-entertainment-infographic.pdf-1.png?generation=1697729470012196&alt=media) | ![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8892775%2F845fb173e3a5416545d0a47ee85a01b6%2F0388-MGMResorts_truth-about-entertainment-infographic.pdf-1.png?generation=1697729495744442&alt=media) |\n\nHope to see a few results from the others as well!\n\n\n\n\n\n\n\n\n",
      "votes": 2
    },
    {
      "id": 2498945,
      "postDate": "2023-10-25T16:08:56.747Z",
      "rawMarkdown": "",
      "isDeleted": true
    },
    {
      "id": 2498936,
      "postDate": "2023-10-25T16:06:10.180Z",
      "rawMarkdown": "",
      "isDeleted": true
    }
  ],
  "comments": [
    {
      "id": 2498945,
      "author_name": "",
      "author_url": "",
      "post_date": "2023-10-25T16:08:56.747000",
      "content": "",
      "votes": 0,
      "replies": []
    },
    {
      "id": 2498936,
      "author_name": "",
      "author_url": "",
      "post_date": "2023-10-25T16:06:10.180000",
      "content": "",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2488944": "Hi All,\n\nSince this competition has ended, i'm curious to know what routes were taken by you all. I myself had the best results with the Restormer architecture. My write up of my progress can be found here: https://toon-beerten.medium.com/denoising-and-reconstructing-dirty-documents-for-optimal-digitalization-ed3a186aa3d6 .\nWe saw that using postprocessing techniques like gaussian blur, the competition score could be improved. But that doesn't necessarily means a better OCR result in the end. That's why i'm very curious to see some actual examples from everybody.\n\nI propose to compare subjectively a few competition images. I chose these four:\n- 0080-lkgp_New-patient-registration-form-280920.pdf-2.png\n0132-PAESTA-denying-apollo-moon-landings.pdf-15.png\n0346-Erie_sPublicSchools_m_e_ch_20.pdf-16.png\n0388-MGMResorts_truth-about-entertainment-infographic.pdf-1.png\n\nLeft is original, right is my model output.\n\n| ![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8892775%2F1fc573f6b4b4a0800e80d785e0640745%2F0080-lkgp_New-patient-registration-form-280920.pdf-2.png?generation=1697729001969554&alt=media) | ![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8892775%2F8e712a194071e20cbe49d57b3cbea165%2F0080-lkgp_New-patient-registration-form-280920.pdf-2.png?generation=1697729568926852&alt=media) |\n| --- | --- |\n| ![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8892775%2Fca21f4335a834864ea655a907b9bec95%2F0132-PAESTA-denying-apollo-moon-landings.pdf-15.png?generation=1697729198258940&alt=media) | ![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8892775%2F827273271b408083953b4d7541cc9bed%2F0132-PAESTA-denying-apollo-moon-landings.pdf-15.png?generation=1697729550786680&alt=media) |\n| --- | --- |\n| ![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8892775%2F6278750b9374d18ba6a30546121e2eb5%2F0346-Erie_sPublicSchools_m_e_ch_20.pdf-16.png?generation=1697729361825485&alt=media) | ![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8892775%2F12da4dee093eceff17e9da7ca85a00e5%2F0346-Erie_sPublicSchools_m_e_ch_20.pdf-16.png?generation=1697729526483403&alt=media) |\n| --- | --- |\n| ![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8892775%2F0ce730ab8f86d0e1cbbef9a67daed7a1%2F0388-MGMResorts_truth-about-entertainment-infographic.pdf-1.png?generation=1697729470012196&alt=media) | ![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8892775%2F845fb173e3a5416545d0a47ee85a01b6%2F0388-MGMResorts_truth-about-entertainment-infographic.pdf-1.png?generation=1697729495744442&alt=media) |\n\nHope to see a few results from the others as well!\n\n\n\n\n\n\n\n\n",
    "2498945": "",
    "2498936": ""
  }
}