{
  "id": 66542,
  "title": "Effectively augmenting Chest X-rays",
  "url": "/competitions/rsna-pneumonia-detection-challenge/discussion/66542",
  "author_name": "",
  "post_date": "2018-09-22T12:09:11.693756700Z",
  "votes": 27,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Hi Kaggle community,</p>\n\n<p>In my previous competitions and daily job, I encountered a lot of difficulties augmenting the data, which consists of images with keypoints / bounding boxes. Recently, I got fed up with the existing libraries and implemented a new fast library for data augmentation: <a href=\"https://github.com/MIPT-Oulu/solt\">https://github.com/MIPT-Oulu/solt</a>.</p>\n\n<p>This library is thoroughly tested and I also claim it to be faster than the existing libraries for Python in certain cases, e.g. when you apply Rotate, Scale and Shear consecutively. Additional feature of my library is a possibility to effectively serialise your augmentation pipelines - therefore you can see in your logs how your augmentations have changed.</p>\n\n<p>Considering this competition, I find it useful to share one of the examples from my library, where I augmented the images with the bounding boxes corresponding to different radiological findings from CXR14 dataset:\n<a href=\"https://mipt-oulu.github.io/solt/Medical_Data_Augmentation_CXR14.html\">https://mipt-oulu.github.io/solt/Medical_Data_Augmentation_CXR14.html</a>. I hope you will find it useful in this challenge!</p>\n\n<p>One example picture: <img src=\"https://mipt-oulu.github.io/solt/_images/Medical_Data_Augmentation_CXR14_22_2.png\" alt=\"augmentation_cxr\"></p>\n\n<p>If you find this library useful - please support it by a GitHub \"star\".</p>\n\n<p>Cheers,\n- Aleksei.</p>",
  "messages": [
    {
      "id": "391775",
      "postDate": "09/22/2018 12:09:11",
      "content": "<p>Hi Kaggle community,</p>\n\n<p>In my previous competitions and daily job, I encountered a lot of difficulties augmenting the data, which consists of images with keypoints / bounding boxes. Recently, I got fed up with the existing libraries and implemented a new fast library for data augmentation: <a href=\"https://github.com/MIPT-Oulu/solt\">https://github.com/MIPT-Oulu/solt</a>.</p>\n\n<p>This library is thoroughly tested and I also claim it to be faster than the existing libraries for Python in certain cases, e.g. when you apply Rotate, Scale and Shear consecutively. Additional feature of my library is a possibility to effectively serialise your augmentation pipelines - therefore you can see in your logs how your augmentations have changed.</p>\n\n<p>Considering this competition, I find it useful to share one of the examples from my library, where I augmented the images with the bounding boxes corresponding to different radiological findings from CXR14 dataset:\n<a href=\"https://mipt-oulu.github.io/solt/Medical_Data_Augmentation_CXR14.html\">https://mipt-oulu.github.io/solt/Medical_Data_Augmentation_CXR14.html</a>. I hope you will find it useful in this challenge!</p>\n\n<p>One example picture: <img src=\"https://mipt-oulu.github.io/solt/_images/Medical_Data_Augmentation_CXR14_22_2.png\" alt=\"augmentation_cxr\"></p>\n\n<p>If you find this library useful - please support it by a GitHub \"star\".</p>\n\n<p>Cheers,\n- Aleksei.</p>",
      "rawMarkdown": "Hi Kaggle community,\n\nIn my previous competitions and daily job, I encountered a lot of difficulties augmenting the data, which consists of images with keypoints / bounding boxes. Recently, I got fed up with the existing libraries and implemented a new fast library for data augmentation: [https://github.com/MIPT-Oulu/solt][1].\n\n This library is thoroughly tested and I also claim it to be faster than the existing libraries for Python in certain cases, e.g. when you apply Rotate, Scale and Shear consecutively. Additional feature of my library is a possibility to effectively serialise your augmentation pipelines - therefore you can see in your logs how your augmentations have changed.\n\nConsidering this competition, I find it useful to share one of the examples from my library, where I augmented the images with the bounding boxes corresponding to different radiological findings from CXR14 dataset:\n[https://mipt-oulu.github.io/solt/Medical_Data_Augmentation_CXR14.html][2]. I hope you will find it useful in this challenge!\n\nOne example picture: ![augmentation_cxr][3]\n\nIf you find this library useful - please support it by a GitHub \"star\".\n\nCheers,\n- Aleksei.\n\n  [1]: https://github.com/MIPT-Oulu/solt\n  [2]: https://mipt-oulu.github.io/solt/Medical_Data_Augmentation_CXR14.html\n  [3]: https://mipt-oulu.github.io/solt/_images/Medical_Data_Augmentation_CXR14_22_2.png",
      "votes": null
    },
    {
      "id": "393286",
      "postDate": "09/25/2018 05:42:45",
      "content": "<p>Hi Aleksei! Thanks for your sharing. Also, could you explain why your library more powerful then imgaug library? (<a href=\"https://github.com/aleju/imgaug\">https://github.com/aleju/imgaug</a>)</p>",
      "rawMarkdown": "Hi Aleksei! Thanks for your sharing. Also, could you explain why your library more powerful then imgaug library? (https://github.com/aleju/imgaug)",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 393286,
      "author_name": "serart",
      "author_url": "",
      "post_date": "09/25/2018 05:42:45",
      "content": "<p>Hi Aleksei! Thanks for your sharing. Also, could you explain why your library more powerful then imgaug library? (<a href=\"https://github.com/aleju/imgaug\">https://github.com/aleju/imgaug</a>)</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "391775": "Hi Kaggle community,\n\nIn my previous competitions and daily job, I encountered a lot of difficulties augmenting the data, which consists of images with keypoints / bounding boxes. Recently, I got fed up with the existing libraries and implemented a new fast library for data augmentation: [https://github.com/MIPT-Oulu/solt][1].\n\n This library is thoroughly tested and I also claim it to be faster than the existing libraries for Python in certain cases, e.g. when you apply Rotate, Scale and Shear consecutively. Additional feature of my library is a possibility to effectively serialise your augmentation pipelines - therefore you can see in your logs how your augmentations have changed.\n\nConsidering this competition, I find it useful to share one of the examples from my library, where I augmented the images with the bounding boxes corresponding to different radiological findings from CXR14 dataset:\n[https://mipt-oulu.github.io/solt/Medical_Data_Augmentation_CXR14.html][2]. I hope you will find it useful in this challenge!\n\nOne example picture: ![augmentation_cxr][3]\n\nIf you find this library useful - please support it by a GitHub \"star\".\n\nCheers,\n- Aleksei.\n\n  [1]: https://github.com/MIPT-Oulu/solt\n  [2]: https://mipt-oulu.github.io/solt/Medical_Data_Augmentation_CXR14.html\n  [3]: https://mipt-oulu.github.io/solt/_images/Medical_Data_Augmentation_CXR14_22_2.png",
    "393286": "Hi Aleksei! Thanks for your sharing. Also, could you explain why your library more powerful then imgaug library? (https://github.com/aleju/imgaug)"
  },
  "source": "meta"
}