{
  "id": 217295,
  "title": "New Design of ResNet by DeepMind",
  "url": "/competitions/cassava-leaf-disease-classification/discussion/217295",
  "author_name": "",
  "post_date": "2021-02-06T08:00:05.954309500Z",
  "votes": 13,
  "comment_count": 9,
  "views": 0,
  "content": "<p>In this paper, a normalizer free method is proposed to design a deep residual network without activating the normalization layer! It can be directly applied to ResNet , RegNet and other networks. Under the same flops, it is comparable to efficientnet! Open source code! Author: deepmind<br>\ncode Link：<a href=\"https://github.com/rwightman/pytorch-image-models\" target=\"_blank\">https://github.com/rwightman/pytorch-image-models</a><br>\npaper Link：<a href=\"https://arxiv.org/abs/2101.08692\" target=\"_blank\">https://arxiv.org/abs/2101.08692</a><br>\n![<a href=\"https://www.kaggle.com/hanson0910/temppic?select=_20210206155308.png](url\" target=\"_blank\">https://www.kaggle.com/hanson0910/temppic?select=_20210206155308.png](url</a> to embed)</p>",
  "messages": [
    {
      "id": "1188403",
      "postDate": "02/06/2021 08:00:05",
      "content": "<p>In this paper, a normalizer free method is proposed to design a deep residual network without activating the normalization layer! It can be directly applied to ResNet , RegNet and other networks. Under the same flops, it is comparable to efficientnet! Open source code! Author: deepmind<br>\ncode Link：<a href=\"https://github.com/rwightman/pytorch-image-models\" target=\"_blank\">https://github.com/rwightman/pytorch-image-models</a><br>\npaper Link：<a href=\"https://arxiv.org/abs/2101.08692\" target=\"_blank\">https://arxiv.org/abs/2101.08692</a><br>\n![<a href=\"https://www.kaggle.com/hanson0910/temppic?select=_20210206155308.png](url\" target=\"_blank\">https://www.kaggle.com/hanson0910/temppic?select=_20210206155308.png](url</a> to embed)</p>",
      "rawMarkdown": "In this paper, a normalizer free method is proposed to design a deep residual network without activating the normalization layer! It can be directly applied to ResNet , RegNet and other networks. Under the same flops, it is comparable to efficientnet! Open source code! Author: deepmind\ncode Link：https://github.com/rwightman/pytorch-image-models\npaper Link：https://arxiv.org/abs/2101.08692\n![https://www.kaggle.com/hanson0910/temppic?select=_20210206155308.png](url to embed)",
      "votes": null
    },
    {
      "id": "1188759",
      "postDate": "02/06/2021 13:52:26",
      "content": "<p>Another google's work implemented on <strong>PyTorch</strong> instead of <strong>TF 2.x</strong>…</p>",
      "rawMarkdown": "Another google's work implemented on **PyTorch** instead of **TF 2.x**...",
      "votes": null
    },
    {
      "id": "1188763",
      "postDate": "02/06/2021 13:57:09",
      "content": "<p>Yes,you can try as much as you like</p>",
      "rawMarkdown": "Yes,you can try as much as you like",
      "votes": null
    },
    {
      "id": "1188785",
      "postDate": "02/06/2021 14:22:20",
      "content": "<p>There is no official implementation yet but yes that will likely be either TF1 or JAX or Pytorch once code released. <br>\nAnd yes Google researchers are still reluctant to use TF2</p>",
      "rawMarkdown": "There is no official implementation yet but yes that will likely be either TF1 or JAX or Pytorch once code released. \nAnd yes Google researchers are still reluctant to use TF2",
      "votes": null
    },
    {
      "id": "1188790",
      "postDate": "02/06/2021 14:24:50",
      "content": "<p>Although not the official code, but also enough to learn</p>",
      "rawMarkdown": "Although not the official code, but also enough to learn",
      "votes": null
    },
    {
      "id": "1189318",
      "postDate": "02/06/2021 22:51:25",
      "content": "<p>I wonder why they released TF2 if none of their recent model are on officially on TF2 …</p>",
      "rawMarkdown": "I wonder why they released TF2 if none of their recent model are on officially on TF2 ...",
      "votes": null
    },
    {
      "id": "1189403",
      "postDate": "02/07/2021 01:58:32",
      "content": "<p>I have the same question, <strong>TabNet</strong> could be a great example.</p>",
      "rawMarkdown": "I have the same question, **TabNet** could be a great example.",
      "votes": null
    },
    {
      "id": "1189614",
      "postDate": "02/07/2021 06:30:12",
      "content": "<p>Wow, thanks for your sharing, do you use it in this competition?</p>",
      "rawMarkdown": "Wow, thanks for your sharing, do you use it in this competition?",
      "votes": null
    },
    {
      "id": "1189930",
      "postDate": "02/07/2021 10:49:57",
      "content": "<p>Not Yet！！！！！</p>",
      "rawMarkdown": "Not Yet！！！！！",
      "votes": null
    },
    {
      "id": "1199581",
      "postDate": "02/14/2021 00:03:05",
      "content": "<p><img src=\"https://i.imgflip.com/4xy4fu.jpg\" alt=\"\"></p>",
      "rawMarkdown": "![](https://i.imgflip.com/4xy4fu.jpg)",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1188759,
      "author_name": "hiramcho",
      "author_url": "",
      "post_date": "02/06/2021 13:52:26",
      "content": "<p>Another google's work implemented on <strong>PyTorch</strong> instead of <strong>TF 2.x</strong>…</p>",
      "votes": null,
      "replies": [
        {
          "id": 1188763,
          "author_name": "hanson0910",
          "author_url": "",
          "post_date": "02/06/2021 13:57:09",
          "content": "<p>Yes,you can try as much as you like</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1188785,
          "author_name": "serigne",
          "author_url": "",
          "post_date": "02/06/2021 14:22:20",
          "content": "<p>There is no official implementation yet but yes that will likely be either TF1 or JAX or Pytorch once code released. <br>\nAnd yes Google researchers are still reluctant to use TF2</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1188790,
          "author_name": "hanson0910",
          "author_url": "",
          "post_date": "02/06/2021 14:24:50",
          "content": "<p>Although not the official code, but also enough to learn</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1189318,
          "author_name": "ludovick",
          "author_url": "",
          "post_date": "02/06/2021 22:51:25",
          "content": "<p>I wonder why they released TF2 if none of their recent model are on officially on TF2 …</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1189403,
          "author_name": "hiramcho",
          "author_url": "",
          "post_date": "02/07/2021 01:58:32",
          "content": "<p>I have the same question, <strong>TabNet</strong> could be a great example.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1189614,
      "author_name": "chenlongwang",
      "author_url": "",
      "post_date": "02/07/2021 06:30:12",
      "content": "<p>Wow, thanks for your sharing, do you use it in this competition?</p>",
      "votes": null,
      "replies": [
        {
          "id": 1189930,
          "author_name": "hanson0910",
          "author_url": "",
          "post_date": "02/07/2021 10:49:57",
          "content": "<p>Not Yet！！！！！</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1199581,
      "author_name": "killimi",
      "author_url": "",
      "post_date": "02/14/2021 00:03:05",
      "content": "<p><img src=\"https://i.imgflip.com/4xy4fu.jpg\" alt=\"\"></p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1188403": "In this paper, a normalizer free method is proposed to design a deep residual network without activating the normalization layer! It can be directly applied to ResNet , RegNet and other networks. Under the same flops, it is comparable to efficientnet! Open source code! Author: deepmind\ncode Link：https://github.com/rwightman/pytorch-image-models\npaper Link：https://arxiv.org/abs/2101.08692\n![https://www.kaggle.com/hanson0910/temppic?select=_20210206155308.png](url to embed)",
    "1188759": "Another google's work implemented on **PyTorch** instead of **TF 2.x**...",
    "1188763": "Yes,you can try as much as you like",
    "1188785": "There is no official implementation yet but yes that will likely be either TF1 or JAX or Pytorch once code released. \nAnd yes Google researchers are still reluctant to use TF2",
    "1188790": "Although not the official code, but also enough to learn",
    "1189318": "I wonder why they released TF2 if none of their recent model are on officially on TF2 ...",
    "1189403": "I have the same question, **TabNet** could be a great example.",
    "1189614": "Wow, thanks for your sharing, do you use it in this competition?",
    "1189930": "Not Yet！！！！！",
    "1199581": "![](https://i.imgflip.com/4xy4fu.jpg)"
  },
  "source": "meta"
}