{
  "id": 305718,
  "title": "This repo will make Tensorflow great again!",
  "url": "/competitions/happy-whale-and-dolphin/discussion/305718",
  "author_name": "",
  "post_date": "2022-02-06T16:08:43.694902700Z",
  "votes": 64,
  "comment_count": 5,
  "views": 0,
  "content": "<p>For Image classification task, most of the new pretrained models are in Pytorch. <br>\nThanks to <a href=\"https://github.com/rwightman/pytorch-image-models\" target=\"_blank\">timm</a>, we can use all these models in Pytorch easily. However we need a lot of GPU resources when using Pytorch.</p>\n<p>For Tensorflow, we can use very powerful TPU while it's lack of new models.😭😭<br>\nNow I found an amazing project <a href=\"https://github.com/martinsbruveris/tensorflow-image-models\" target=\"_blank\">tfimm</a>, it's very just easy to use just as timm. I've tested some backbone models like swin-transformer and vit on recent petfiner competition and achieved similar results. </p>\n<p>This project is still on a very early stage, hope we can contribute to it more.😛</p>",
  "messages": [
    {
      "id": "1678555",
      "postDate": "02/06/2022 16:08:43",
      "content": "<p>For Image classification task, most of the new pretrained models are in Pytorch. <br>\nThanks to <a href=\"https://github.com/rwightman/pytorch-image-models\" target=\"_blank\">timm</a>, we can use all these models in Pytorch easily. However we need a lot of GPU resources when using Pytorch.</p>\n<p>For Tensorflow, we can use very powerful TPU while it's lack of new models.😭😭<br>\nNow I found an amazing project <a href=\"https://github.com/martinsbruveris/tensorflow-image-models\" target=\"_blank\">tfimm</a>, it's very just easy to use just as timm. I've tested some backbone models like swin-transformer and vit on recent petfiner competition and achieved similar results. </p>\n<p>This project is still on a very early stage, hope we can contribute to it more.😛</p>",
      "rawMarkdown": "For Image classification task, most of the new pretrained models are in Pytorch. \nThanks to [timm](https://github.com/rwightman/pytorch-image-models), we can use all these models in Pytorch easily. However we need a lot of GPU resources when using Pytorch.\n\nFor Tensorflow, we can use very powerful TPU while it's lack of new models.😭😭\nNow I found an amazing project [tfimm](https://github.com/martinsbruveris/tensorflow-image-models), it's very just easy to use just as timm. I've tested some backbone models like swin-transformer and vit on recent petfiner competition and achieved similar results. \n\nThis project is still on a very early stage, hope we can contribute to it more.😛",
      "votes": null
    },
    {
      "id": "1680716",
      "postDate": "02/08/2022 01:29:31",
      "content": "<p>Very useful</p>",
      "rawMarkdown": "Very useful",
      "votes": null
    },
    {
      "id": "1682402",
      "postDate": "02/09/2022 06:33:53",
      "content": "<p>Just wanted to say thank you for this amazing recourse! I am using Tensorflow and Keras, and this library provides an enormous amount of opportunities without using Pytorch.</p>",
      "rawMarkdown": "Just wanted to say thank you for this amazing recourse! I am using Tensorflow and Keras, and this library provides an enormous amount of opportunities without using Pytorch.",
      "votes": null
    },
    {
      "id": "1685675",
      "postDate": "02/11/2022 13:08:28",
      "content": "<p>Very useful</p>",
      "rawMarkdown": "Very useful",
      "votes": null
    },
    {
      "id": "1686735",
      "postDate": "02/12/2022 10:02:23",
      "content": "<p>Wow. Very useful.</p>",
      "rawMarkdown": "Wow. Very useful.",
      "votes": null
    },
    {
      "id": "1696685",
      "postDate": "02/19/2022 02:59:21",
      "content": "<p>make tensorflow great again!</p>",
      "rawMarkdown": "make tensorflow great again!",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1680716,
      "author_name": "hanxiao2007",
      "author_url": "",
      "post_date": "02/08/2022 01:29:31",
      "content": "<p>Very useful</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1682402,
      "author_name": "antonzv",
      "author_url": "",
      "post_date": "02/09/2022 06:33:53",
      "content": "<p>Just wanted to say thank you for this amazing recourse! I am using Tensorflow and Keras, and this library provides an enormous amount of opportunities without using Pytorch.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1685675,
      "author_name": "abdulmanankhalid",
      "author_url": "",
      "post_date": "02/11/2022 13:08:28",
      "content": "<p>Very useful</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1686735,
      "author_name": "dinowun",
      "author_url": "",
      "post_date": "02/12/2022 10:02:23",
      "content": "<p>Wow. Very useful.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1696685,
      "author_name": "fss331",
      "author_url": "",
      "post_date": "02/19/2022 02:59:21",
      "content": "<p>make tensorflow great again!</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1678555": "For Image classification task, most of the new pretrained models are in Pytorch. \nThanks to [timm](https://github.com/rwightman/pytorch-image-models), we can use all these models in Pytorch easily. However we need a lot of GPU resources when using Pytorch.\n\nFor Tensorflow, we can use very powerful TPU while it's lack of new models.😭😭\nNow I found an amazing project [tfimm](https://github.com/martinsbruveris/tensorflow-image-models), it's very just easy to use just as timm. I've tested some backbone models like swin-transformer and vit on recent petfiner competition and achieved similar results. \n\nThis project is still on a very early stage, hope we can contribute to it more.😛",
    "1680716": "Very useful",
    "1682402": "Just wanted to say thank you for this amazing recourse! I am using Tensorflow and Keras, and this library provides an enormous amount of opportunities without using Pytorch.",
    "1685675": "Very useful",
    "1686735": "Wow. Very useful.",
    "1696685": "make tensorflow great again!"
  },
  "source": "meta"
}