{
  "id": 475201,
  "title": "Has anyone tried mask based augmentation?",
  "url": "/competitions/blood-vessel-segmentation/discussion/475201",
  "author_name": "samshipengs",
  "post_date": "2024-02-07T13:50:51.931000",
  "votes": 1,
  "comment_count": 0,
  "views": 0,
  "content": "<p>I was inspired by one of Heng's <a href=\"https://www.kaggle.com/competitions/blood-vessel-segmentation/discussion/456118#2552494\" target=\"_blank\">thread</a>, I thought it makes sense to add some variations in the mask region where the model likely pay most if not all of its attention to, so I tried creating some difference (such as brightness fading from inner boundary to the center, or partial gray region)<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F1878085%2F36a76f14431fddd9fb7e6698c96e710e%2Fdemo.png?generation=1707313626023928&amp;alt=media\" alt=\"partial gray region\"> <br>\n(left: original, right: partial grayed region)</p>\n<p>but from what I observed, this distablized training and validation performances gets worse, and seems model became easier to overfit. Has anyone tried similar, what are your experience and thoughts?</p>",
  "messages": [
    {
      "id": 2641463,
      "postDate": "2024-02-07T13:50:51.933Z",
      "content": "<p>I was inspired by one of Heng's <a href=\"https://www.kaggle.com/competitions/blood-vessel-segmentation/discussion/456118#2552494\" target=\"_blank\">thread</a>, I thought it makes sense to add some variations in the mask region where the model likely pay most if not all of its attention to, so I tried creating some difference (such as brightness fading from inner boundary to the center, or partial gray region)<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F1878085%2F36a76f14431fddd9fb7e6698c96e710e%2Fdemo.png?generation=1707313626023928&amp;alt=media\" alt=\"partial gray region\"> <br>\n(left: original, right: partial grayed region)</p>\n<p>but from what I observed, this distablized training and validation performances gets worse, and seems model became easier to overfit. Has anyone tried similar, what are your experience and thoughts?</p>",
      "rawMarkdown": "I was inspired by one of Heng's [thread](https://www.kaggle.com/competitions/blood-vessel-segmentation/discussion/456118#2552494), I thought it makes sense to add some variations in the mask region where the model likely pay most if not all of its attention to, so I tried creating some difference (such as brightness fading from inner boundary to the center, or partial gray region)![partial gray region](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F1878085%2F36a76f14431fddd9fb7e6698c96e710e%2Fdemo.png?generation=1707313626023928&alt=media) \n(left: original, right: partial grayed region)\n\nbut from what I observed, this distablized training and validation performances gets worse, and seems model became easier to overfit. Has anyone tried similar, what are your experience and thoughts?",
      "votes": 1
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "2641463": "I was inspired by one of Heng's [thread](https://www.kaggle.com/competitions/blood-vessel-segmentation/discussion/456118#2552494), I thought it makes sense to add some variations in the mask region where the model likely pay most if not all of its attention to, so I tried creating some difference (such as brightness fading from inner boundary to the center, or partial gray region)![partial gray region](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F1878085%2F36a76f14431fddd9fb7e6698c96e710e%2Fdemo.png?generation=1707313626023928&alt=media) \n(left: original, right: partial grayed region)\n\nbut from what I observed, this distablized training and validation performances gets worse, and seems model became easier to overfit. Has anyone tried similar, what are your experience and thoughts?"
  }
}