{
  "id": 420744,
  "title": "Removing small blood vessel predictions",
  "url": "/competitions/hubmap-hacking-the-human-vasculature/discussion/420744",
  "author_name": "",
  "post_date": "2023-07-02T10:06:43.197462300Z",
  "votes": 1,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Looking at the typical blood vessel sizes in the training set, they tend to be on the order of a thousand pixels, with a very small proportion being smaller than e.g. 100 pixels. Perhaps certain models which have a tendency to predict a lot of very small regions could benefit from simply removing regions with a size below a certain threshold, such as 100.</p>\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F14274288%2Fc325c5a3358885265f151bab810764b9%2Fbloodvesselsize.PNG?generation=1688292167216775&amp;alt=media\" alt=\"\"></p>",
  "messages": [
    {
      "id": "2326672",
      "postDate": "07/02/2023 10:06:43",
      "content": "<p>Looking at the typical blood vessel sizes in the training set, they tend to be on the order of a thousand pixels, with a very small proportion being smaller than e.g. 100 pixels. Perhaps certain models which have a tendency to predict a lot of very small regions could benefit from simply removing regions with a size below a certain threshold, such as 100.</p>\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F14274288%2Fc325c5a3358885265f151bab810764b9%2Fbloodvesselsize.PNG?generation=1688292167216775&amp;alt=media\" alt=\"\"></p>",
      "rawMarkdown": "Looking at the typical blood vessel sizes in the training set, they tend to be on the order of a thousand pixels, with a very small proportion being smaller than e.g. 100 pixels. Perhaps certain models which have a tendency to predict a lot of very small regions could benefit from simply removing regions with a size below a certain threshold, such as 100.\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F14274288%2Fc325c5a3358885265f151bab810764b9%2Fbloodvesselsize.PNG?generation=1688292167216775&alt=media)",
      "votes": null
    },
    {
      "id": "2329232",
      "postDate": "07/04/2023 06:49:03",
      "content": "<p>thank for sharing！Have you conducted any experiments on this？</p>",
      "rawMarkdown": "thank for sharing！Have you conducted any experiments on this？",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2329232,
      "author_name": "fanshutou",
      "author_url": "",
      "post_date": "07/04/2023 06:49:03",
      "content": "<p>thank for sharing！Have you conducted any experiments on this？</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2326672": "Looking at the typical blood vessel sizes in the training set, they tend to be on the order of a thousand pixels, with a very small proportion being smaller than e.g. 100 pixels. Perhaps certain models which have a tendency to predict a lot of very small regions could benefit from simply removing regions with a size below a certain threshold, such as 100.\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F14274288%2Fc325c5a3358885265f151bab810764b9%2Fbloodvesselsize.PNG?generation=1688292167216775&alt=media)",
    "2329232": "thank for sharing！Have you conducted any experiments on this？"
  },
  "source": "meta"
}