{
  "id": 427540,
  "title": "New loss function to potentially boost the prediction",
  "url": "/competitions/google-research-identify-contrails-reduce-global-warming/discussion/427540",
  "author_name": "junzi",
  "post_date": "2023-07-28T11:22:17.209000",
  "votes": 9,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Hey, we recently released some open-source code on contrail segmentation, with a new loss function that seems to boost the contrail detection performance (based on our limited GOES).</p>\n<p><a href=\"https://arxiv.org/abs/2307.12032\" target=\"_blank\">https://github.com/junzis/contrail-net</a></p>\n<p>I am curious to see if anyone is interested and willing to give it a try :)</p>\n<p>The detailed explanation of the code is in paper: <a href=\"https://arxiv.org/abs/2307.12032\" target=\"_blank\">https://arxiv.org/abs/2307.12032</a></p>",
  "messages": [
    {
      "id": 2362918,
      "postDate": "2023-07-28T11:22:17.210Z",
      "content": "<p>Hey, we recently released some open-source code on contrail segmentation, with a new loss function that seems to boost the contrail detection performance (based on our limited GOES).</p>\n<p><a href=\"https://arxiv.org/abs/2307.12032\" target=\"_blank\">https://github.com/junzis/contrail-net</a></p>\n<p>I am curious to see if anyone is interested and willing to give it a try :)</p>\n<p>The detailed explanation of the code is in paper: <a href=\"https://arxiv.org/abs/2307.12032\" target=\"_blank\">https://arxiv.org/abs/2307.12032</a></p>",
      "rawMarkdown": "Hey, we recently released some open-source code on contrail segmentation, with a new loss function that seems to boost the contrail detection performance (based on our limited GOES).\n\n[https://github.com/junzis/contrail-net](https://arxiv.org/abs/2307.12032)\n\nI am curious to see if anyone is interested and willing to give it a try :)\n\nThe detailed explanation of the code is in paper: [https://arxiv.org/abs/2307.12032](https://arxiv.org/abs/2307.12032)",
      "votes": 8
    },
    {
      "id": 2363022,
      "postDate": "2023-07-28T13:04:22.260Z",
      "content": "<p>Did you train a model on 20 images only?</p>",
      "rawMarkdown": "Did you train a model on 20 images only?",
      "votes": 1
    },
    {
      "id": 2362929,
      "postDate": "2023-07-28T11:33:00.707Z",
      "content": "<p>Thanks for sharing, let me try it</p>",
      "rawMarkdown": "Thanks for sharing, let me try it"
    }
  ],
  "comments": [
    {
      "id": 2363022,
      "author_name": "Optimo",
      "author_url": "",
      "post_date": "2023-07-28T13:04:22.260000",
      "content": "<p>Did you train a model on 20 images only?</p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 2362929,
      "author_name": "william.wu",
      "author_url": "",
      "post_date": "2023-07-28T11:33:00.707000",
      "content": "<p>Thanks for sharing, let me try it</p>",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2362918": "Hey, we recently released some open-source code on contrail segmentation, with a new loss function that seems to boost the contrail detection performance (based on our limited GOES).\n\n[https://github.com/junzis/contrail-net](https://arxiv.org/abs/2307.12032)\n\nI am curious to see if anyone is interested and willing to give it a try :)\n\nThe detailed explanation of the code is in paper: [https://arxiv.org/abs/2307.12032](https://arxiv.org/abs/2307.12032)",
    "2363022": "Did you train a model on 20 images only?",
    "2362929": "Thanks for sharing, let me try it"
  }
}