{
  "id": 214999,
  "title": "Segmentation model for catheters",
  "url": "/competitions/ranzcr-clip-catheter-line-classification/discussion/214999",
  "author_name": "ryches",
  "post_date": "2021-01-28T08:49:15.980000",
  "votes": 8,
  "comment_count": 0,
  "views": 0,
  "content": "<p>As mentioned in the previous <a href=\"https://www.kaggle.com/c/ranzcr-clip-catheter-line-classification/discussion/212698\" target=\"_blank\">thread</a> about segmentation, I have created a model that tries to find the position of the catheter by training on the information made available by the annotations. This model and code is available at <a href=\"https://www.kaggle.com/ryches/segmentation-model\" target=\"_blank\">https://www.kaggle.com/ryches/segmentation-model</a></p>\n<p>The details are in the notebook but the general idea is to train on the annotations and then either transfer this knowledge to the classification task or to generate masks that can be used as additional input to a classification model. This notebook only trains for one epoch and one fold, but can be easily extended and customized to do some variations on this idea. </p>\n<p>Some samples of the undertrained models predictions</p>\n<p><img src=\"https://i.imgur.com/ZxKX5iD.png\" alt=\"\"><br>\n<img src=\"https://i.imgur.com/3Qi1AoM.png\" alt=\"\"><br>\n<img src=\"https://i.imgur.com/NAS0L4K.png\" alt=\"\"><br>\n<img src=\"https://i.imgur.com/wfdJBSM.png\" alt=\"\"></p>",
  "messages": [
    {
      "id": 1174079,
      "postDate": "2021-01-28T08:49:15.980Z",
      "content": "<p>As mentioned in the previous <a href=\"https://www.kaggle.com/c/ranzcr-clip-catheter-line-classification/discussion/212698\" target=\"_blank\">thread</a> about segmentation, I have created a model that tries to find the position of the catheter by training on the information made available by the annotations. This model and code is available at <a href=\"https://www.kaggle.com/ryches/segmentation-model\" target=\"_blank\">https://www.kaggle.com/ryches/segmentation-model</a></p>\n<p>The details are in the notebook but the general idea is to train on the annotations and then either transfer this knowledge to the classification task or to generate masks that can be used as additional input to a classification model. This notebook only trains for one epoch and one fold, but can be easily extended and customized to do some variations on this idea. </p>\n<p>Some samples of the undertrained models predictions</p>\n<p><img src=\"https://i.imgur.com/ZxKX5iD.png\" alt=\"\"><br>\n<img src=\"https://i.imgur.com/3Qi1AoM.png\" alt=\"\"><br>\n<img src=\"https://i.imgur.com/NAS0L4K.png\" alt=\"\"><br>\n<img src=\"https://i.imgur.com/wfdJBSM.png\" alt=\"\"></p>",
      "rawMarkdown": "As mentioned in the previous [thread](https://www.kaggle.com/c/ranzcr-clip-catheter-line-classification/discussion/212698) about segmentation, I have created a model that tries to find the position of the catheter by training on the information made available by the annotations. This model and code is available at https://www.kaggle.com/ryches/segmentation-model\n\nThe details are in the notebook but the general idea is to train on the annotations and then either transfer this knowledge to the classification task or to generate masks that can be used as additional input to a classification model. This notebook only trains for one epoch and one fold, but can be easily extended and customized to do some variations on this idea. \n\nSome samples of the undertrained models predictions\n\n![](https://i.imgur.com/ZxKX5iD.png)\n![](https://i.imgur.com/3Qi1AoM.png)\n![](https://i.imgur.com/NAS0L4K.png)\n![](https://i.imgur.com/wfdJBSM.png)\n\n",
      "votes": 8
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "1174079": "As mentioned in the previous [thread](https://www.kaggle.com/c/ranzcr-clip-catheter-line-classification/discussion/212698) about segmentation, I have created a model that tries to find the position of the catheter by training on the information made available by the annotations. This model and code is available at https://www.kaggle.com/ryches/segmentation-model\n\nThe details are in the notebook but the general idea is to train on the annotations and then either transfer this knowledge to the classification task or to generate masks that can be used as additional input to a classification model. This notebook only trains for one epoch and one fold, but can be easily extended and customized to do some variations on this idea. \n\nSome samples of the undertrained models predictions\n\n![](https://i.imgur.com/ZxKX5iD.png)\n![](https://i.imgur.com/3Qi1AoM.png)\n![](https://i.imgur.com/NAS0L4K.png)\n![](https://i.imgur.com/wfdJBSM.png)\n\n"
  }
}