{
  "id": 458774,
  "title": "freeze of submission score !!",
  "url": "/competitions/UBC-OCEAN/discussion/458774",
  "author_name": "Ahmed Mandour",
  "post_date": "2023-12-01T13:00:45.399000",
  "votes": 2,
  "comment_count": 1,
  "views": 0,
  "content": "<p>why i get the same score 0.16 each time, this is my pipeline :<br>\ni fine tune pretrained model on tiles dataset from <a href=\"https://www.kaggle.com/JIRKA\" target=\"_blank\">@JIRKA</a> BOROVEC (only 100 random tiles from each train image) and got a very good result on testset 0.7 (and of course using class weight ) then during inference i tile test image to 512 tile and predict but observing that my model always predict HGSC so only 0.16 score ,<br>\nDoes anyone know the problem ?</p>",
  "messages": [
    {
      "id": 2545368,
      "postDate": "2023-12-01T13:00:45.400Z",
      "content": "<p>why i get the same score 0.16 each time, this is my pipeline :<br>\ni fine tune pretrained model on tiles dataset from <a href=\"https://www.kaggle.com/JIRKA\" target=\"_blank\">@JIRKA</a> BOROVEC (only 100 random tiles from each train image) and got a very good result on testset 0.7 (and of course using class weight ) then during inference i tile test image to 512 tile and predict but observing that my model always predict HGSC so only 0.16 score ,<br>\nDoes anyone know the problem ?</p>",
      "rawMarkdown": "why i get the same score 0.16 each time, this is my pipeline :\ni fine tune pretrained model on tiles dataset from @JIRKA BOROVEC (only 100 random tiles from each train image) and got a very good result on testset 0.7 (and of course using class weight ) then during inference i tile test image to 512 tile and predict but observing that my model always predict HGSC so only 0.16 score ,\nDoes anyone know the problem ?\n",
      "votes": 2
    },
    {
      "id": 2552482,
      "postDate": "2023-12-07T13:54:19.460Z",
      "content": "<p>I encountered the same problem. Do you know the answer now? <br>\nTrying different models with good CV acc still always yields 0.16 LB results, and what's more confusing is that this is equal the sample_submission.csv score. May be there are some bugs in the inference code, but I haven't found them yet</p>",
      "rawMarkdown": "I encountered the same problem. Do you know the answer now? \nTrying different models with good CV acc still always yields 0.16 LB results, and what's more confusing is that this is equal the sample_submission.csv score. May be there are some bugs in the inference code, but I haven't found them yet"
    }
  ],
  "comments": [
    {
      "id": 2552482,
      "author_name": "kaggleranonymous",
      "author_url": "",
      "post_date": "2023-12-07T13:54:19.460000",
      "content": "<p>I encountered the same problem. Do you know the answer now? <br>\nTrying different models with good CV acc still always yields 0.16 LB results, and what's more confusing is that this is equal the sample_submission.csv score. May be there are some bugs in the inference code, but I haven't found them yet</p>",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2545368": "why i get the same score 0.16 each time, this is my pipeline :\ni fine tune pretrained model on tiles dataset from @JIRKA BOROVEC (only 100 random tiles from each train image) and got a very good result on testset 0.7 (and of course using class weight ) then during inference i tile test image to 512 tile and predict but observing that my model always predict HGSC so only 0.16 score ,\nDoes anyone know the problem ?\n",
    "2552482": "I encountered the same problem. Do you know the answer now? \nTrying different models with good CV acc still always yields 0.16 LB results, and what's more confusing is that this is equal the sample_submission.csv score. May be there are some bugs in the inference code, but I haven't found them yet"
  }
}