{
  "id": 471055,
  "title": "A bug that could lead to 0 score in lb",
  "url": "/competitions/blood-vessel-segmentation/discussion/471055",
  "author_name": "",
  "post_date": "2024-01-26T18:30:27.207413600Z",
  "votes": 5,
  "comment_count": 1,
  "views": 0,
  "content": "<p>I am still learning about segmentation, and I want to share some of what I've learned along the way, maybe that can help someone struggling with same issues I encountered. <br>\nOne issue I encountered was consistently receiving a 0 score on the leaderboard while achieving high scores during validation. The reason was that I used Albumentations.resize during training and validation, but used cv2.resize for inference. <br>\nI hadn't considered that they might have really different internal implementations😅(though I am using same params). <br>\nUsing same resizing function fixed the issue.</p>",
  "messages": [
    {
      "id": "2621371",
      "postDate": "01/26/2024 18:30:27",
      "content": "<p>I am still learning about segmentation, and I want to share some of what I've learned along the way, maybe that can help someone struggling with same issues I encountered. <br>\nOne issue I encountered was consistently receiving a 0 score on the leaderboard while achieving high scores during validation. The reason was that I used Albumentations.resize during training and validation, but used cv2.resize for inference. <br>\nI hadn't considered that they might have really different internal implementations😅(though I am using same params). <br>\nUsing same resizing function fixed the issue.</p>",
      "rawMarkdown": "I am still learning about segmentation, and I want to share some of what I've learned along the way, maybe that can help someone struggling with same issues I encountered. \nOne issue I encountered was consistently receiving a 0 score on the leaderboard while achieving high scores during validation. The reason was that I used Albumentations.resize during training and validation, but used cv2.resize for inference. \nI hadn't considered that they might have really different internal implementations😅(though I am using same params). \nUsing same resizing function fixed the issue.",
      "votes": null
    },
    {
      "id": "2621515",
      "postDate": "01/26/2024 19:50:24",
      "content": "<p>this is probably because <code>cv2.resize(img, (w, h) )</code> expects (w, h) whereas albumentations expects (h, w)</p>",
      "rawMarkdown": "this is probably because `cv2.resize(img, (w, h) )` expects (w, h) whereas albumentations expects (h, w)",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2621515,
      "author_name": "ihebch",
      "author_url": "",
      "post_date": "01/26/2024 19:50:24",
      "content": "<p>this is probably because <code>cv2.resize(img, (w, h) )</code> expects (w, h) whereas albumentations expects (h, w)</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2621371": "I am still learning about segmentation, and I want to share some of what I've learned along the way, maybe that can help someone struggling with same issues I encountered. \nOne issue I encountered was consistently receiving a 0 score on the leaderboard while achieving high scores during validation. The reason was that I used Albumentations.resize during training and validation, but used cv2.resize for inference. \nI hadn't considered that they might have really different internal implementations😅(though I am using same params). \nUsing same resizing function fixed the issue.",
    "2621515": "this is probably because `cv2.resize(img, (w, h) )` expects (w, h) whereas albumentations expects (h, w)"
  },
  "source": "meta"
}