{
  "id": 334041,
  "title": "Document Image Enhancement Transformer?",
  "url": "/competitions/denoising-shabby-pages/discussion/334041",
  "author_name": "",
  "post_date": "2022-06-29T13:35:00.473849500Z",
  "votes": null,
  "comment_count": 5,
  "views": 0,
  "content": "<p>Hello everyone,</p>\n<p>I am trying this approach currently. <a href=\"https://arxiv.org/pdf/2201.10252.pdf\" target=\"_blank\">https://arxiv.org/pdf/2201.10252.pdf</a><br>\nDocEnTr: An End-to-End Document Image Enhancement Transformer</p>\n<p>Currently I have ran few epochs training and results seems strange. What do you think? Due to less data it is over fitting maybe? Maybe I should finetune the model instead of training model from scratch that would help. </p>",
  "messages": [
    {
      "id": "1837319",
      "postDate": "06/29/2022 13:35:00",
      "content": "<p>Hello everyone,</p>\n<p>I am trying this approach currently. <a href=\"https://arxiv.org/pdf/2201.10252.pdf\" target=\"_blank\">https://arxiv.org/pdf/2201.10252.pdf</a><br>\nDocEnTr: An End-to-End Document Image Enhancement Transformer</p>\n<p>Currently I have ran few epochs training and results seems strange. What do you think? Due to less data it is over fitting maybe? Maybe I should finetune the model instead of training model from scratch that would help. </p>",
      "rawMarkdown": "Hello everyone,\n\nI am trying this approach currently. https://arxiv.org/pdf/2201.10252.pdf\nDocEnTr: An End-to-End Document Image Enhancement Transformer\n\nCurrently I have ran few epochs training and results seems strange. What do you think? Due to less data it is over fitting maybe? Maybe I should finetune the model instead of training model from scratch that would help.",
      "votes": null
    },
    {
      "id": "1837321",
      "postDate": "06/29/2022 13:35:35",
      "content": "<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F2648341%2Fa09ac478a559c6d5dd6585ff129d0d0b%2F0313-Baylor_91501.pdf-1.png?generation=1656509752586991&amp;alt=media\" alt=\"\"><br>\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F2648341%2F370dad8299f0f28bc81f63ec4da83e4a%2F0144-ascap_100418-fitness2017-blank-agreement.pdf-6.png?generation=1656509773533439&amp;alt=media\" alt=\"\"></p>",
      "rawMarkdown": "![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F2648341%2Fa09ac478a559c6d5dd6585ff129d0d0b%2F0313-Baylor_91501.pdf-1.png?generation=1656509752586991&alt=media)\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F2648341%2F370dad8299f0f28bc81f63ec4da83e4a%2F0144-ascap_100418-fitness2017-blank-agreement.pdf-6.png?generation=1656509773533439&alt=media)",
      "votes": null
    },
    {
      "id": "1838902",
      "postDate": "07/01/2022 01:16:42",
      "content": "<p>What would be the total images for the training? And what would be the epoch and batch size to reproduce this result? Usually i'm getting weird result when the model is underfitting, sometimes that can be resolved by using larger epoch number and different batch size.</p>",
      "rawMarkdown": "What would be the total images for the training? And what would be the epoch and batch size to reproduce this result? Usually i'm getting weird result when the model is underfitting, sometimes that can be resolved by using larger epoch number and different batch size.",
      "votes": null
    },
    {
      "id": "1839270",
      "postDate": "07/01/2022 08:58:52",
      "content": "<p>I am using all training images at the moment. I ran 6 epochs. But let's see maybe more epochs can help. But I think due to slow training kaggle kernels might time out</p>",
      "rawMarkdown": "I am using all training images at the moment. I ran 6 epochs. But let's see maybe more epochs can help. But I think due to slow training kaggle kernels might time out",
      "votes": null
    },
    {
      "id": "1848841",
      "postDate": "07/09/2022 00:43:58",
      "content": "<p>So is it getting better result by using more epochs?</p>",
      "rawMarkdown": "So is it getting better result by using more epochs?",
      "votes": null
    },
    {
      "id": "1850605",
      "postDate": "07/10/2022 14:26:04",
      "content": "<p>No it did not improved..<br>\nI have shifted to another approach now <a href=\"https://github.com/vashiegaran/DE-GAN-Implementation-Using-PyTorch\" target=\"_blank\">https://github.com/vashiegaran/DE-GAN-Implementation-Using-PyTorch</a><br>\n<a href=\"https://www.kaggle.com/code/deepdreamx/de-gan\" target=\"_blank\">https://www.kaggle.com/code/deepdreamx/de-gan</a><br>\nBut still bad results :(</p>",
      "rawMarkdown": "No it did not improved..\nI have shifted to another approach now https://github.com/vashiegaran/DE-GAN-Implementation-Using-PyTorch\nhttps://www.kaggle.com/code/deepdreamx/de-gan\nBut still bad results :(",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1837321,
      "author_name": "deepdreamx",
      "author_url": "",
      "post_date": "06/29/2022 13:35:35",
      "content": "<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F2648341%2Fa09ac478a559c6d5dd6585ff129d0d0b%2F0313-Baylor_91501.pdf-1.png?generation=1656509752586991&amp;alt=media\" alt=\"\"><br>\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F2648341%2F370dad8299f0f28bc81f63ec4da83e4a%2F0144-ascap_100418-fitness2017-blank-agreement.pdf-6.png?generation=1656509773533439&amp;alt=media\" alt=\"\"></p>",
      "votes": null,
      "replies": [
        {
          "id": 1838902,
          "author_name": "weisinx7",
          "author_url": "",
          "post_date": "07/01/2022 01:16:42",
          "content": "<p>What would be the total images for the training? And what would be the epoch and batch size to reproduce this result? Usually i'm getting weird result when the model is underfitting, sometimes that can be resolved by using larger epoch number and different batch size.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1839270,
          "author_name": "deepdreamx",
          "author_url": "",
          "post_date": "07/01/2022 08:58:52",
          "content": "<p>I am using all training images at the moment. I ran 6 epochs. But let's see maybe more epochs can help. But I think due to slow training kaggle kernels might time out</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1848841,
          "author_name": "weisinx7",
          "author_url": "",
          "post_date": "07/09/2022 00:43:58",
          "content": "<p>So is it getting better result by using more epochs?</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1850605,
          "author_name": "deepdreamx",
          "author_url": "",
          "post_date": "07/10/2022 14:26:04",
          "content": "<p>No it did not improved..<br>\nI have shifted to another approach now <a href=\"https://github.com/vashiegaran/DE-GAN-Implementation-Using-PyTorch\" target=\"_blank\">https://github.com/vashiegaran/DE-GAN-Implementation-Using-PyTorch</a><br>\n<a href=\"https://www.kaggle.com/code/deepdreamx/de-gan\" target=\"_blank\">https://www.kaggle.com/code/deepdreamx/de-gan</a><br>\nBut still bad results :(</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1837319": "Hello everyone,\n\nI am trying this approach currently. https://arxiv.org/pdf/2201.10252.pdf\nDocEnTr: An End-to-End Document Image Enhancement Transformer\n\nCurrently I have ran few epochs training and results seems strange. What do you think? Due to less data it is over fitting maybe? Maybe I should finetune the model instead of training model from scratch that would help.",
    "1837321": "![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F2648341%2Fa09ac478a559c6d5dd6585ff129d0d0b%2F0313-Baylor_91501.pdf-1.png?generation=1656509752586991&alt=media)\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F2648341%2F370dad8299f0f28bc81f63ec4da83e4a%2F0144-ascap_100418-fitness2017-blank-agreement.pdf-6.png?generation=1656509773533439&alt=media)",
    "1838902": "What would be the total images for the training? And what would be the epoch and batch size to reproduce this result? Usually i'm getting weird result when the model is underfitting, sometimes that can be resolved by using larger epoch number and different batch size.",
    "1839270": "I am using all training images at the moment. I ran 6 epochs. But let's see maybe more epochs can help. But I think due to slow training kaggle kernels might time out",
    "1848841": "So is it getting better result by using more epochs?",
    "1850605": "No it did not improved..\nI have shifted to another approach now https://github.com/vashiegaran/DE-GAN-Implementation-Using-PyTorch\nhttps://www.kaggle.com/code/deepdreamx/de-gan\nBut still bad results :("
  },
  "source": "meta"
}