{
  "id": 376197,
  "title": "LB score is not improving?",
  "url": "/competitions/rsna-breast-cancer-detection/discussion/376197",
  "author_name": "Myo Min Htet(wnp)🇲🇲",
  "post_date": "2023-01-05T06:42:54.024000",
  "votes": 3,
  "comment_count": 2,
  "views": 0,
  "content": "<p>I've tried many times with different size but my LB score is not getting improved. My highest LB score is 0.16<br>\nefficientnet_b2 with 1024 size also not working. I am using pos weight and simple training pipeline.</p>",
  "messages": [
    {
      "id": 2086912,
      "postDate": "2023-01-05T06:42:54.023Z",
      "content": "<p>I've tried many times with different size but my LB score is not getting improved. My highest LB score is 0.16<br>\nefficientnet_b2 with 1024 size also not working. I am using pos weight and simple training pipeline.</p>",
      "rawMarkdown": "I've tried many times with different size but my LB score is not getting improved. My highest LB score is 0.16\nefficientnet_b2 with 1024 size also not working. I am using pos weight and simple training pipeline.",
      "votes": 3
    },
    {
      "id": 2086935,
      "postDate": "2023-01-05T07:02:40.343Z",
      "content": "<p>Hi Myominhtet,</p>\n<p>What for me worked very well was to go from an efficientnet b2 model to an efficientnet b4 model. Other things that you could try that improved my score a lot was to use region of interest images and upsampling of positive samples (together with a positive weight).</p>\n<p>Hope this helps a bit and good luck!</p>",
      "rawMarkdown": "Hi Myominhtet,\n\nWhat for me worked very well was to go from an efficientnet b2 model to an efficientnet b4 model. Other things that you could try that improved my score a lot was to use region of interest images and upsampling of positive samples (together with a positive weight).\n\nHope this helps a bit and good luck!",
      "votes": 1,
      "replies": [
        {
          "id": 2086983,
          "postDate": "2023-01-05T08:00:25.177Z",
          "content": "<p>thanks man</p>",
          "rawMarkdown": "thanks man\n",
          "votes": 1
        }
      ]
    }
  ],
  "comments": [
    {
      "id": 2086935,
      "author_name": "menno",
      "author_url": "",
      "post_date": "2023-01-05T07:02:40.343000",
      "content": "<p>Hi Myominhtet,</p>\n<p>What for me worked very well was to go from an efficientnet b2 model to an efficientnet b4 model. Other things that you could try that improved my score a lot was to use region of interest images and upsampling of positive samples (together with a positive weight).</p>\n<p>Hope this helps a bit and good luck!</p>",
      "votes": 1,
      "replies": [
        {
          "id": 2086983,
          "author_name": "Myo Min Htet(wnp)🇲🇲",
          "author_url": "",
          "post_date": "2023-01-05T08:00:25.177000",
          "content": "<p>thanks man</p>",
          "votes": 1,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "2086912": "I've tried many times with different size but my LB score is not getting improved. My highest LB score is 0.16\nefficientnet_b2 with 1024 size also not working. I am using pos weight and simple training pipeline.",
    "2086935": "Hi Myominhtet,\n\nWhat for me worked very well was to go from an efficientnet b2 model to an efficientnet b4 model. Other things that you could try that improved my score a lot was to use region of interest images and upsampling of positive samples (together with a positive weight).\n\nHope this helps a bit and good luck!"
  }
}