{
  "id": 108648,
  "title": "Could the data be augmentated ?",
  "url": "/competitions/kuzushiji-recognition/discussion/108648",
  "author_name": "",
  "post_date": "2019-09-13T04:45:38.469231600Z",
  "votes": 2,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Hello everyone, I am new to deep learning and I wonder how the data of this competition be augmentated ?</p>",
  "messages": [
    {
      "id": "625430",
      "postDate": "09/13/2019 04:45:38",
      "content": "<p>Hello everyone, I am new to deep learning and I wonder how the data of this competition be augmentated ?</p>",
      "rawMarkdown": "Hello everyone, I am new to deep learning and I wonder how the data of this competition be augmentated ?",
      "votes": null
    },
    {
      "id": "625623",
      "postDate": "09/13/2019 08:30:21",
      "content": "<p>Since paper has different colors and meaning does not change depending on the paper color, you can try adjusting the whole image colors and intensity, e.g. hue, saturation, value. Also changing the size does not change the meaning, but it can be a bit trickier to integrate properly, it depends on your pipeline. In theory, small elastic transforms (as here <a href=\"https://github.com/albu/albumentations/\">https://github.com/albu/albumentations/</a>) should also work but I didn't try them yet.</p>",
      "rawMarkdown": "Since paper has different colors and meaning does not change depending on the paper color, you can try adjusting the whole image colors and intensity, e.g. hue, saturation, value. Also changing the size does not change the meaning, but it can be a bit trickier to integrate properly, it depends on your pipeline. In theory, small elastic transforms (as here https://github.com/albu/albumentations/) should also work but I didn't try them yet.",
      "votes": null
    },
    {
      "id": "634178",
      "postDate": "09/26/2019 01:54:03",
      "content": "<p>thank you</p>",
      "rawMarkdown": "thank you",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 625623,
      "author_name": "lopuhin",
      "author_url": "",
      "post_date": "09/13/2019 08:30:21",
      "content": "<p>Since paper has different colors and meaning does not change depending on the paper color, you can try adjusting the whole image colors and intensity, e.g. hue, saturation, value. Also changing the size does not change the meaning, but it can be a bit trickier to integrate properly, it depends on your pipeline. In theory, small elastic transforms (as here <a href=\"https://github.com/albu/albumentations/\">https://github.com/albu/albumentations/</a>) should also work but I didn't try them yet.</p>",
      "votes": null,
      "replies": [
        {
          "id": 634178,
          "author_name": "thanhtinqn97",
          "author_url": "",
          "post_date": "09/26/2019 01:54:03",
          "content": "<p>thank you</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "625430": "Hello everyone, I am new to deep learning and I wonder how the data of this competition be augmentated ?",
    "625623": "Since paper has different colors and meaning does not change depending on the paper color, you can try adjusting the whole image colors and intensity, e.g. hue, saturation, value. Also changing the size does not change the meaning, but it can be a bit trickier to integrate properly, it depends on your pipeline. In theory, small elastic transforms (as here https://github.com/albu/albumentations/) should also work but I didn't try them yet.",
    "634178": "thank you"
  },
  "source": "meta"
}