{
  "id": 152172,
  "title": "Just a quick idea for beginners...",
  "url": "/competitions/prostate-cancer-grade-assessment/discussion/152172",
  "author_name": "PAB97",
  "post_date": "2020-05-18T17:53:54.459000",
  "votes": 1,
  "comment_count": 0,
  "views": 0,
  "content": "<p>I've posted a kernel that displays how to use GradCAM to analyze the predictions made by your model: <a href=\"https://www.kaggle.com/rftexas/gradcam-comparing-resnet-and-se-resnext\">https://www.kaggle.com/rftexas/gradcam-comparing-resnet-and-se-resnext</a></p>\n\n<p>It is clear that to come up with a great boost in score, one need to zoom in some regions of the biopsy to make sure that there is no blank space left... Indeed, each square inch left blank is a square inch that does not convey information.</p>\n\n<p>By coding a simple heuristic with a sliding window of let's say size 128x128 and comparing the percentage of blank space within the image. One can definitely witness a significant boost in score. </p>",
  "messages": [
    {
      "id": 852831,
      "postDate": "2020-05-18T17:53:54.460Z",
      "content": "<p>I've posted a kernel that displays how to use GradCAM to analyze the predictions made by your model: <a href=\"https://www.kaggle.com/rftexas/gradcam-comparing-resnet-and-se-resnext\">https://www.kaggle.com/rftexas/gradcam-comparing-resnet-and-se-resnext</a></p>\n\n<p>It is clear that to come up with a great boost in score, one need to zoom in some regions of the biopsy to make sure that there is no blank space left... Indeed, each square inch left blank is a square inch that does not convey information.</p>\n\n<p>By coding a simple heuristic with a sliding window of let's say size 128x128 and comparing the percentage of blank space within the image. One can definitely witness a significant boost in score. </p>",
      "rawMarkdown": "I've posted a kernel that displays how to use GradCAM to analyze the predictions made by your model: https://www.kaggle.com/rftexas/gradcam-comparing-resnet-and-se-resnext\n\nIt is clear that to come up with a great boost in score, one need to zoom in some regions of the biopsy to make sure that there is no blank space left... Indeed, each square inch left blank is a square inch that does not convey information.\n\nBy coding a simple heuristic with a sliding window of let's say size 128x128 and comparing the percentage of blank space within the image. One can definitely witness a significant boost in score. ",
      "votes": 1
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "852831": "I've posted a kernel that displays how to use GradCAM to analyze the predictions made by your model: https://www.kaggle.com/rftexas/gradcam-comparing-resnet-and-se-resnext\n\nIt is clear that to come up with a great boost in score, one need to zoom in some regions of the biopsy to make sure that there is no blank space left... Indeed, each square inch left blank is a square inch that does not convey information.\n\nBy coding a simple heuristic with a sliding window of let's say size 128x128 and comparing the percentage of blank space within the image. One can definitely witness a significant boost in score. "
  }
}