{
  "id": 375943,
  "title": "Did anyone else notice \"unreasonable effectiveness\" of convnext in this competition?",
  "url": "/competitions/g2net-detecting-continuous-gravitational-waves/discussion/375943",
  "author_name": "DennisSakva",
  "post_date": "2023-01-04T06:05:01.444000",
  "votes": 1,
  "comment_count": 4,
  "views": 0,
  "content": "<p>I tried many backbones with both CNN and transformer architectures and all of them performed worse than the ConvNext Base. I wonder if other participants noticed this and if they have any explanation of why this is the case.</p>\n<p>Oh, and congratulations to the winners. One of the most \"non-orthodox\" competitions I've ever participated in. The diversity of solutions is simply amazing.</p>",
  "messages": [
    {
      "id": 2085401,
      "postDate": "2023-01-04T06:30:30.973Z",
      "content": "<p>yes I also used convnext small and that makes me up!</p>",
      "rawMarkdown": "yes I also used convnext small and that makes me up!",
      "votes": 1
    },
    {
      "id": 2085385,
      "postDate": "2023-01-04T06:05:01.443Z",
      "content": "<p>I tried many backbones with both CNN and transformer architectures and all of them performed worse than the ConvNext Base. I wonder if other participants noticed this and if they have any explanation of why this is the case.</p>\n<p>Oh, and congratulations to the winners. One of the most \"non-orthodox\" competitions I've ever participated in. The diversity of solutions is simply amazing.</p>",
      "rawMarkdown": "I tried many backbones with both CNN and transformer architectures and all of them performed worse than the ConvNext Base. I wonder if other participants noticed this and if they have any explanation of why this is the case.\n\nOh, and congratulations to the winners. One of the most \"non-orthodox\" competitions I've ever participated in. The diversity of solutions is simply amazing.\n",
      "votes": 1
    },
    {
      "id": 2085645,
      "postDate": "2023-01-04T09:45:18.243Z",
      "content": "<p>ResNeSt worked pretty well for me</p>",
      "rawMarkdown": "ResNeSt worked pretty well for me",
      "replies": [
        {
          "id": 2085699,
          "postDate": "2023-01-04T10:35:14.913Z",
          "content": "<p>Did you try convnext?</p>",
          "rawMarkdown": "Did you try convnext?",
          "replies": [
            {
              "id": 2085706,
              "postDate": "2023-01-04T10:39:02.690Z",
              "content": "<p>unfortunately - not, among all backbones I tried I never came up to convnext :( Don't know why</p>",
              "rawMarkdown": "unfortunately - not, among all backbones I tried I never came up to convnext :( Don't know why"
            }
          ]
        }
      ]
    }
  ],
  "comments": [
    {
      "id": 2085401,
      "author_name": "Jungwoo Park",
      "author_url": "",
      "post_date": "2023-01-04T06:30:30.973000",
      "content": "<p>yes I also used convnext small and that makes me up!</p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 2085645,
      "author_name": "Ubique",
      "author_url": "",
      "post_date": "2023-01-04T09:45:18.243000",
      "content": "<p>ResNeSt worked pretty well for me</p>",
      "votes": 0,
      "replies": [
        {
          "id": 2085699,
          "author_name": "DennisSakva",
          "author_url": "",
          "post_date": "2023-01-04T10:35:14.913000",
          "content": "<p>Did you try convnext?</p>",
          "votes": 0,
          "replies": [
            {
              "id": 2085706,
              "author_name": "Ubique",
              "author_url": "",
              "post_date": "2023-01-04T10:39:02.690000",
              "content": "<p>unfortunately - not, among all backbones I tried I never came up to convnext :( Don't know why</p>",
              "votes": 0,
              "replies": []
            }
          ]
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "2085401": "yes I also used convnext small and that makes me up!",
    "2085385": "I tried many backbones with both CNN and transformer architectures and all of them performed worse than the ConvNext Base. I wonder if other participants noticed this and if they have any explanation of why this is the case.\n\nOh, and congratulations to the winners. One of the most \"non-orthodox\" competitions I've ever participated in. The diversity of solutions is simply amazing.\n",
    "2085645": "ResNeSt worked pretty well for me"
  }
}