{
  "id": 324572,
  "title": "What models are you using in this competition? Has anyone tried transformer?",
  "url": "/competitions/birdclef-2022/discussion/324572",
  "author_name": "",
  "post_date": "2022-05-12T10:08:43.450052400Z",
  "votes": null,
  "comment_count": 2,
  "views": 0,
  "content": "<p>I used EfficientNet in this competition, I wonder if the model has a great influence on the score of the competition?</p>",
  "messages": [
    {
      "id": "1785668",
      "postDate": "05/12/2022 10:08:43",
      "content": "<p>I used EfficientNet in this competition, I wonder if the model has a great influence on the score of the competition?</p>",
      "rawMarkdown": "I used EfficientNet in this competition, I wonder if the model has a great influence on the score of the competition?",
      "votes": null
    },
    {
      "id": "1785673",
      "postDate": "05/12/2022 10:12:32",
      "content": "<p>I found that large batch hurt the competition score, What's the reason? Shouldn't the BN layer use large batch?</p>",
      "rawMarkdown": "I found that large batch hurt the competition score, What's the reason? Shouldn't the BN layer use large batch?",
      "votes": null
    },
    {
      "id": "1786367",
      "postDate": "05/12/2022 20:00:07",
      "content": "<p>I've been using EfficientNet, too. Intuitively, transformers might have an edge because the features you're looking for are a little more diffuse / \"disconnected\" than most image classification problems and I think xformers are a little better for that type of data. I came into this contest really late, though, and there's very little time to experiment, at least for those of us who are below the medal rank! </p>",
      "rawMarkdown": "I've been using EfficientNet, too. Intuitively, transformers might have an edge because the features you're looking for are a little more diffuse / \"disconnected\" than most image classification problems and I think xformers are a little better for that type of data. I came into this contest really late, though, and there's very little time to experiment, at least for those of us who are below the medal rank!",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1785673,
      "author_name": "liuxwww",
      "author_url": "",
      "post_date": "05/12/2022 10:12:32",
      "content": "<p>I found that large batch hurt the competition score, What's the reason? Shouldn't the BN layer use large batch?</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1786367,
      "author_name": "lobrien",
      "author_url": "",
      "post_date": "05/12/2022 20:00:07",
      "content": "<p>I've been using EfficientNet, too. Intuitively, transformers might have an edge because the features you're looking for are a little more diffuse / \"disconnected\" than most image classification problems and I think xformers are a little better for that type of data. I came into this contest really late, though, and there's very little time to experiment, at least for those of us who are below the medal rank! </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1785668": "I used EfficientNet in this competition, I wonder if the model has a great influence on the score of the competition?",
    "1785673": "I found that large batch hurt the competition score, What's the reason? Shouldn't the BN layer use large batch?",
    "1786367": "I've been using EfficientNet, too. Intuitively, transformers might have an edge because the features you're looking for are a little more diffuse / \"disconnected\" than most image classification problems and I think xformers are a little better for that type of data. I came into this contest really late, though, and there's very little time to experiment, at least for those of us who are below the medal rank!"
  },
  "source": "meta"
}