{
  "id": 211545,
  "title": "Medical image augmentation ",
  "url": "/competitions/ranzcr-clip-catheter-line-classification/discussion/211545",
  "author_name": "",
  "post_date": "2021-01-15T15:29:20.975139400Z",
  "votes": 2,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Dear kagglers,<br>\nCurrent competition train/validation AUC diverges after 15-20 epochs, and I am unable to control and improve training model to reflect better validation data. Since, image augmentation provides additional noise and signal, I tried horizontal-flip, brightness, contrast, and  even vertical flip using <code>tensorflow.image</code> APIs. However, can't see much improvement on leader board. <br>\nAm I missing something here; how and what kind of augmentation in medical images show better results? Please share your thoughts from your experience. </p>\n<p>Regards<br>\nCH</p>",
  "messages": [
    {
      "id": "1154337",
      "postDate": "01/15/2021 15:29:20",
      "content": "<p>Dear kagglers,<br>\nCurrent competition train/validation AUC diverges after 15-20 epochs, and I am unable to control and improve training model to reflect better validation data. Since, image augmentation provides additional noise and signal, I tried horizontal-flip, brightness, contrast, and  even vertical flip using <code>tensorflow.image</code> APIs. However, can't see much improvement on leader board. <br>\nAm I missing something here; how and what kind of augmentation in medical images show better results? Please share your thoughts from your experience. </p>\n<p>Regards<br>\nCH</p>",
      "rawMarkdown": "Dear kagglers,\nCurrent competition train/validation AUC diverges after 15-20 epochs, and I am unable to control and improve training model to reflect better validation data. Since, image augmentation provides additional noise and signal, I tried horizontal-flip, brightness, contrast, and  even vertical flip using `tensorflow.image` APIs. However, can't see much improvement on leader board. \nAm I missing something here; how and what kind of augmentation in medical images show better results? Please share your thoughts from your experience. \n\nRegards\nCH",
      "votes": null
    },
    {
      "id": "1154424",
      "postDate": "01/15/2021 16:16:49",
      "content": "<p>There is a work called <a href=\"https://www.kaggle.com/khoongweihao/x-ray-needle-augmentation-et-al\" target=\"_blank\">X-ray Needle Augmentation et al.</a> that could be useful for you.</p>",
      "rawMarkdown": "There is a work called [X-ray Needle Augmentation et al.](https://www.kaggle.com/khoongweihao/x-ray-needle-augmentation-et-al) that could be useful for you.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1154424,
      "author_name": "hiramcho",
      "author_url": "",
      "post_date": "01/15/2021 16:16:49",
      "content": "<p>There is a work called <a href=\"https://www.kaggle.com/khoongweihao/x-ray-needle-augmentation-et-al\" target=\"_blank\">X-ray Needle Augmentation et al.</a> that could be useful for you.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1154337": "Dear kagglers,\nCurrent competition train/validation AUC diverges after 15-20 epochs, and I am unable to control and improve training model to reflect better validation data. Since, image augmentation provides additional noise and signal, I tried horizontal-flip, brightness, contrast, and  even vertical flip using `tensorflow.image` APIs. However, can't see much improvement on leader board. \nAm I missing something here; how and what kind of augmentation in medical images show better results? Please share your thoughts from your experience. \n\nRegards\nCH",
    "1154424": "There is a work called [X-ray Needle Augmentation et al.](https://www.kaggle.com/khoongweihao/x-ray-needle-augmentation-et-al) that could be useful for you."
  },
  "source": "meta"
}