{
  "id": 430485,
  "title": "Did you take any action against the percentage of Contrails?",
  "url": "/competitions/google-research-identify-contrails-reduce-global-warming/discussion/430485",
  "author_name": "teyosan1229",
  "post_date": "2023-08-10T02:46:21.031000",
  "votes": 1,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Thanks for the competition. Congratulations to all the winners!  </p>\n<p>I was new to segmentation and this competition helped me learn the basics!</p>\n<p>Now, I was trying to calculate the CV and determine the threshold in the validation data.<br>\nAs a result, most of the models have a threshold value of 0.01~0.05. However, using this value, LBscore was significantly worse.</p>\n<p>I checked the masks and found a divergence of 0.53% for Train and 0.18% for validation, so I gave up on adjusting the thresholds.</p>\n<p>Have you guys done any efforts to consider the percentage of Contrails?</p>",
  "messages": [
    {
      "id": 2383017,
      "postDate": "2023-08-10T06:22:57.723Z",
      "content": "<p>From the time it took for inference, we found that the percentage of images without masks was close to validation, so we used the same threshold setting as validation.</p>",
      "rawMarkdown": "From the time it took for inference, we found that the percentage of images without masks was close to validation, so we used the same threshold setting as validation.",
      "votes": 1,
      "replies": [
        {
          "id": 2383320,
          "postDate": "2023-08-10T09:59:13.567Z",
          "content": "<p>Thank you for the information!</p>",
          "rawMarkdown": "Thank you for the information!"
        }
      ]
    },
    {
      "id": 2382763,
      "postDate": "2023-08-10T02:46:21.030Z",
      "content": "<p>Thanks for the competition. Congratulations to all the winners!  </p>\n<p>I was new to segmentation and this competition helped me learn the basics!</p>\n<p>Now, I was trying to calculate the CV and determine the threshold in the validation data.<br>\nAs a result, most of the models have a threshold value of 0.01~0.05. However, using this value, LBscore was significantly worse.</p>\n<p>I checked the masks and found a divergence of 0.53% for Train and 0.18% for validation, so I gave up on adjusting the thresholds.</p>\n<p>Have you guys done any efforts to consider the percentage of Contrails?</p>",
      "rawMarkdown": "Thanks for the competition. Congratulations to all the winners!  \n\n\nI was new to segmentation and this competition helped me learn the basics!\n\nNow, I was trying to calculate the CV and determine the threshold in the validation data.\nAs a result, most of the models have a threshold value of 0.01~0.05. However, using this value, LBscore was significantly worse.\n\nI checked the masks and found a divergence of 0.53% for Train and 0.18% for validation, so I gave up on adjusting the thresholds.\n\nHave you guys done any efforts to consider the percentage of Contrails?"
    }
  ],
  "comments": [
    {
      "id": 2383017,
      "author_name": "tattaka",
      "author_url": "",
      "post_date": "2023-08-10T06:22:57.723000",
      "content": "<p>From the time it took for inference, we found that the percentage of images without masks was close to validation, so we used the same threshold setting as validation.</p>",
      "votes": 1,
      "replies": [
        {
          "id": 2383320,
          "author_name": "teyosan1229",
          "author_url": "",
          "post_date": "2023-08-10T09:59:13.567000",
          "content": "<p>Thank you for the information!</p>",
          "votes": 0,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "2383017": "From the time it took for inference, we found that the percentage of images without masks was close to validation, so we used the same threshold setting as validation.",
    "2382763": "Thanks for the competition. Congratulations to all the winners!  \n\n\nI was new to segmentation and this competition helped me learn the basics!\n\nNow, I was trying to calculate the CV and determine the threshold in the validation data.\nAs a result, most of the models have a threshold value of 0.01~0.05. However, using this value, LBscore was significantly worse.\n\nI checked the masks and found a divergence of 0.53% for Train and 0.18% for validation, so I gave up on adjusting the thresholds.\n\nHave you guys done any efforts to consider the percentage of Contrails?"
  }
}