{
  "id": 291261,
  "title": "SOTA model: Swin Transformer V2",
  "url": "/competitions/tensorflow-great-barrier-reef/discussion/291261",
  "author_name": "",
  "post_date": "2021-11-28T13:56:47.554719800Z",
  "votes": 6,
  "comment_count": 2,
  "views": 0,
  "content": "<p>According to <a href=\"https://paperswithcode.com/sota/object-detection-on-coco\" target=\"_blank\">Object Detection on COCO test-dev</a>, <a href=\"https://paperswithcode.com/paper/swin-transformer-v2-scaling-up-capacity-and\" target=\"_blank\">Swin Transformer V2</a> acheived sota.  <br>\nI prepare the train notebook for EfficientDet because it was used in <a href=\"https://www.kaggle.com/c/global-wheat-detection\" target=\"_blank\">the past wheat competition</a>. But it outperforms EfficientDet D7! I was surprised at rapid advance in technology:)  <br>\nThe code is available from <a href=\"https://github.com/microsoft/Swin-Transformer\" target=\"_blank\">here</a>.<br>\nSo I want to compare the accuracy between the two.</p>",
  "messages": [
    {
      "id": "1598479",
      "postDate": "11/28/2021 13:56:47",
      "content": "<p>According to <a href=\"https://paperswithcode.com/sota/object-detection-on-coco\" target=\"_blank\">Object Detection on COCO test-dev</a>, <a href=\"https://paperswithcode.com/paper/swin-transformer-v2-scaling-up-capacity-and\" target=\"_blank\">Swin Transformer V2</a> acheived sota.  <br>\nI prepare the train notebook for EfficientDet because it was used in <a href=\"https://www.kaggle.com/c/global-wheat-detection\" target=\"_blank\">the past wheat competition</a>. But it outperforms EfficientDet D7! I was surprised at rapid advance in technology:)  <br>\nThe code is available from <a href=\"https://github.com/microsoft/Swin-Transformer\" target=\"_blank\">here</a>.<br>\nSo I want to compare the accuracy between the two.</p>",
      "rawMarkdown": "According to [Object Detection on COCO test-dev](https://paperswithcode.com/sota/object-detection-on-coco), [Swin Transformer V2](https://paperswithcode.com/paper/swin-transformer-v2-scaling-up-capacity-and) acheived sota.  \nI prepare the train notebook for EfficientDet because it was used in [the past wheat competition](https://www.kaggle.com/c/global-wheat-detection). But it outperforms EfficientDet D7! I was surprised at rapid advance in technology:)  \nThe code is available from [here](https://github.com/microsoft/Swin-Transformer).\nSo I want to compare the accuracy between the two.",
      "votes": null
    },
    {
      "id": "1618816",
      "postDate": "12/15/2021 11:34:22",
      "content": "<p>I considered Swin Transformer at first, but I realised that it has really big number of parameters and kaggle notebooks only has 16 gb gpu. When I have got much time, will try to implement it. If it works, I would definitely publish that notebook :)</p>",
      "rawMarkdown": "I considered Swin Transformer at first, but I realised that it has really big number of parameters and kaggle notebooks only has 16 gb gpu. When I have got much time, will try to implement it. If it works, I would definitely publish that notebook :)",
      "votes": null
    },
    {
      "id": "1618833",
      "postDate": "12/15/2021 11:49:45",
      "content": "<p>I am waitin for it:)</p>",
      "rawMarkdown": "I am waitin for it:)",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1618816,
      "author_name": "sarperyurttas",
      "author_url": "",
      "post_date": "12/15/2021 11:34:22",
      "content": "<p>I considered Swin Transformer at first, but I realised that it has really big number of parameters and kaggle notebooks only has 16 gb gpu. When I have got much time, will try to implement it. If it works, I would definitely publish that notebook :)</p>",
      "votes": null,
      "replies": [
        {
          "id": 1618833,
          "author_name": "osamurai",
          "author_url": "",
          "post_date": "12/15/2021 11:49:45",
          "content": "<p>I am waitin for it:)</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1598479": "According to [Object Detection on COCO test-dev](https://paperswithcode.com/sota/object-detection-on-coco), [Swin Transformer V2](https://paperswithcode.com/paper/swin-transformer-v2-scaling-up-capacity-and) acheived sota.  \nI prepare the train notebook for EfficientDet because it was used in [the past wheat competition](https://www.kaggle.com/c/global-wheat-detection). But it outperforms EfficientDet D7! I was surprised at rapid advance in technology:)  \nThe code is available from [here](https://github.com/microsoft/Swin-Transformer).\nSo I want to compare the accuracy between the two.",
    "1618816": "I considered Swin Transformer at first, but I realised that it has really big number of parameters and kaggle notebooks only has 16 gb gpu. When I have got much time, will try to implement it. If it works, I would definitely publish that notebook :)",
    "1618833": "I am waitin for it:)"
  },
  "source": "meta"
}