{
  "id": 579443,
  "title": "tomo_5764_d6 and tomo_3a8480",
  "url": "/competitions/byu-locating-bacterial-flagellar-motors-2025/discussion/579443",
  "author_name": "",
  "post_date": "2025-05-17T13:46:26.271948700Z",
  "votes": null,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Any idea to preprocess them somehow to get valuable information out of these type of tomograms? </p>\n<p>Other suggestions and topics are welcomed</p>\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F25917519%2F42306ac2fe89e398f06f453a29904aca%2Ftomo_3a8480_z0227.jpg?generation=1747489750779844&amp;alt=media\" alt=\"\"></p>\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F25917519%2F42eed623f15d43267ff60ee6eea33706%2Ftomo_5764d6_z0096.jpg?generation=1747490191927800&amp;alt=media\" alt=\"\"></p>",
  "messages": [
    {
      "id": "3203920",
      "postDate": "05/17/2025 13:46:26",
      "content": "<p>Any idea to preprocess them somehow to get valuable information out of these type of tomograms? </p>\n<p>Other suggestions and topics are welcomed</p>\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F25917519%2F42306ac2fe89e398f06f453a29904aca%2Ftomo_3a8480_z0227.jpg?generation=1747489750779844&amp;alt=media\" alt=\"\"></p>\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F25917519%2F42eed623f15d43267ff60ee6eea33706%2Ftomo_5764d6_z0096.jpg?generation=1747490191927800&amp;alt=media\" alt=\"\"></p>",
      "rawMarkdown": "Any idea to preprocess them somehow to get valuable information out of these type of tomograms? \n\nOther suggestions and topics are welcomed\n\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F25917519%2F42306ac2fe89e398f06f453a29904aca%2Ftomo_3a8480_z0227.jpg?generation=1747489750779844&alt=media)\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F25917519%2F42eed623f15d43267ff60ee6eea33706%2Ftomo_5764d6_z0096.jpg?generation=1747490191927800&alt=media)",
      "votes": null
    },
    {
      "id": "3203948",
      "postDate": "05/17/2025 14:38:29",
      "content": "<p>To ignore domain specific data, you can use training-irrelevant method like classic keypoint detection to capture this.</p>",
      "rawMarkdown": "To ignore domain specific data, you can use training-irrelevant method like classic keypoint detection to capture this.",
      "votes": null
    },
    {
      "id": "3203994",
      "postDate": "05/17/2025 16:43:50",
      "content": "<p>Keypoint generation is the key, what do you suggest here ??  </p>\n<p>orbs? <br>\nrandom sampling (dont think it will work pretty well (keypoints have small localization)</p>\n<p>CNN feature maps?? (I m trying this (2d (your sppf post) + 3d with DEdgeConvs) (too much training time but high hopes) </p>\n<p>also why I m having hard time converging my yolo's ??</p>",
      "rawMarkdown": "Keypoint generation is the key, what do you suggest here ??  \n\norbs? \nrandom sampling (dont think it will work pretty well (keypoints have small localization)\n\nCNN feature maps?? (I m trying this (2d (your sppf post) + 3d with DEdgeConvs) (too much training time but high hopes) \n\nalso why I m having hard time converging my yolo's ??",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3203948,
      "author_name": "tom99763",
      "author_url": "",
      "post_date": "05/17/2025 14:38:29",
      "content": "<p>To ignore domain specific data, you can use training-irrelevant method like classic keypoint detection to capture this.</p>",
      "votes": null,
      "replies": [
        {
          "id": 3203994,
          "author_name": "suryatrainer",
          "author_url": "",
          "post_date": "05/17/2025 16:43:50",
          "content": "<p>Keypoint generation is the key, what do you suggest here ??  </p>\n<p>orbs? <br>\nrandom sampling (dont think it will work pretty well (keypoints have small localization)</p>\n<p>CNN feature maps?? (I m trying this (2d (your sppf post) + 3d with DEdgeConvs) (too much training time but high hopes) </p>\n<p>also why I m having hard time converging my yolo's ??</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "3203920": "Any idea to preprocess them somehow to get valuable information out of these type of tomograms? \n\nOther suggestions and topics are welcomed\n\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F25917519%2F42306ac2fe89e398f06f453a29904aca%2Ftomo_3a8480_z0227.jpg?generation=1747489750779844&alt=media)\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F25917519%2F42eed623f15d43267ff60ee6eea33706%2Ftomo_5764d6_z0096.jpg?generation=1747490191927800&alt=media)",
    "3203948": "To ignore domain specific data, you can use training-irrelevant method like classic keypoint detection to capture this.",
    "3203994": "Keypoint generation is the key, what do you suggest here ??  \n\norbs? \nrandom sampling (dont think it will work pretty well (keypoints have small localization)\n\nCNN feature maps?? (I m trying this (2d (your sppf post) + 3d with DEdgeConvs) (too much training time but high hopes) \n\nalso why I m having hard time converging my yolo's ??"
  },
  "source": "meta"
}