{
  "id": 555389,
  "title": "Success with different masking methods?",
  "url": "/competitions/czii-cryo-et-object-identification/discussion/555389",
  "author_name": "Connor",
  "post_date": "2025-01-07T04:33:00.886000",
  "votes": 2,
  "comment_count": 0,
  "views": 0,
  "content": "<p>I've run experiments with the three following mask creation methods for a single stage heatmap model with a point extraction function run on post processing:</p>\n<ol>\n<li>A multi-class binary mask that fills the radius for each particle</li>\n<li>A smaller mask used to fit the centroid (i.e. 10 pixels)</li>\n<li>A gaussian mask with the goal of giving a probability of 1.0 around the centroid of the target and having it fade towards 0.0 at the edge of the target radius to help learn surrounding features.</li>\n</ol>\n<p>Has anyone else experimented with any other masking methods or had more success with one or the other?  I seem to get similar results for options 2 and 3 and need to refine my heatmap to coordinate function for option 1 if I am to use that.</p>",
  "messages": [
    {
      "id": 3090257,
      "postDate": "2025-01-07T04:33:00.887Z",
      "content": "<p>I've run experiments with the three following mask creation methods for a single stage heatmap model with a point extraction function run on post processing:</p>\n<ol>\n<li>A multi-class binary mask that fills the radius for each particle</li>\n<li>A smaller mask used to fit the centroid (i.e. 10 pixels)</li>\n<li>A gaussian mask with the goal of giving a probability of 1.0 around the centroid of the target and having it fade towards 0.0 at the edge of the target radius to help learn surrounding features.</li>\n</ol>\n<p>Has anyone else experimented with any other masking methods or had more success with one or the other?  I seem to get similar results for options 2 and 3 and need to refine my heatmap to coordinate function for option 1 if I am to use that.</p>",
      "rawMarkdown": "I've run experiments with the three following mask creation methods for a single stage heatmap model with a point extraction function run on post processing:\n1. A multi-class binary mask that fills the radius for each particle\n2. A smaller mask used to fit the centroid (i.e. 10 pixels)\n3. A gaussian mask with the goal of giving a probability of 1.0 around the centroid of the target and having it fade towards 0.0 at the edge of the target radius to help learn surrounding features.\n\nHas anyone else experimented with any other masking methods or had more success with one or the other?  I seem to get similar results for options 2 and 3 and need to refine my heatmap to coordinate function for option 1 if I am to use that.",
      "votes": 2
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "3090257": "I've run experiments with the three following mask creation methods for a single stage heatmap model with a point extraction function run on post processing:\n1. A multi-class binary mask that fills the radius for each particle\n2. A smaller mask used to fit the centroid (i.e. 10 pixels)\n3. A gaussian mask with the goal of giving a probability of 1.0 around the centroid of the target and having it fade towards 0.0 at the edge of the target radius to help learn surrounding features.\n\nHas anyone else experimented with any other masking methods or had more success with one or the other?  I seem to get similar results for options 2 and 3 and need to refine my heatmap to coordinate function for option 1 if I am to use that."
  }
}