{
  "id": 428376,
  "title": "Things I have learned",
  "url": "/competitions/hubmap-hacking-the-human-vasculature/discussion/428376",
  "author_name": "Yassine Alouini",
  "post_date": "2023-08-01T07:17:16.624000",
  "votes": 6,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Even though the result isn't that great, I have managed to learn a few things:</p>\n<ul>\n<li>More medical domain knowledge</li>\n<li><strong>UNet</strong> could be used as a <strong>semantic segmentation model first</strong> and then separate the segmentation mask into the instances.</li>\n<li><a href=\"https://mmdetection.readthedocs.io/en/latest/\" target=\"_blank\"><strong>MMDet</strong></a> is a powerful set of tools even though the API needs to get used to first (especially if you are migrating from 2.x to 3.x)</li>\n<li>Various techniques to <strong>ensemble predicted segmentations</strong> especially weighted boxes (and masks) fusion <a href=\"https://github.com/ZFTurbo/Weighted-Boxes-Fusion\" target=\"_blank\">https://github.com/ZFTurbo/Weighted-Boxes-Fusion</a> </li>\n</ul>\n<p>Please share the things you have learned, thanks!</p>",
  "messages": [
    {
      "id": 2368456,
      "postDate": "2023-08-01T07:17:16.623Z",
      "content": "<p>Even though the result isn't that great, I have managed to learn a few things:</p>\n<ul>\n<li>More medical domain knowledge</li>\n<li><strong>UNet</strong> could be used as a <strong>semantic segmentation model first</strong> and then separate the segmentation mask into the instances.</li>\n<li><a href=\"https://mmdetection.readthedocs.io/en/latest/\" target=\"_blank\"><strong>MMDet</strong></a> is a powerful set of tools even though the API needs to get used to first (especially if you are migrating from 2.x to 3.x)</li>\n<li>Various techniques to <strong>ensemble predicted segmentations</strong> especially weighted boxes (and masks) fusion <a href=\"https://github.com/ZFTurbo/Weighted-Boxes-Fusion\" target=\"_blank\">https://github.com/ZFTurbo/Weighted-Boxes-Fusion</a> </li>\n</ul>\n<p>Please share the things you have learned, thanks!</p>",
      "rawMarkdown": "Even though the result isn't that great, I have managed to learn a few things:\n\n- More medical domain knowledge\n- **UNet** could be used as a **semantic segmentation model first** and then separate the segmentation mask into the instances.\n- [**MMDet**](https://mmdetection.readthedocs.io/en/latest/) is a powerful set of tools even though the API needs to get used to first (especially if you are migrating from 2.x to 3.x)\n- Various techniques to **ensemble predicted segmentations** especially weighted boxes (and masks) fusion https://github.com/ZFTurbo/Weighted-Boxes-Fusion \n\nPlease share the things you have learned, thanks!",
      "votes": 5
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "2368456": "Even though the result isn't that great, I have managed to learn a few things:\n\n- More medical domain knowledge\n- **UNet** could be used as a **semantic segmentation model first** and then separate the segmentation mask into the instances.\n- [**MMDet**](https://mmdetection.readthedocs.io/en/latest/) is a powerful set of tools even though the API needs to get used to first (especially if you are migrating from 2.x to 3.x)\n- Various techniques to **ensemble predicted segmentations** especially weighted boxes (and masks) fusion https://github.com/ZFTurbo/Weighted-Boxes-Fusion \n\nPlease share the things you have learned, thanks!"
  }
}