{
  "id": 206345,
  "title": "Data-efficient image Transformers: Beating EfficientNet",
  "url": "/competitions/cassava-leaf-disease-classification/discussion/206345",
  "author_name": "",
  "post_date": "2020-12-24T07:04:42.049006500Z",
  "votes": 5,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Hi everyone. Facebook just released Data-efficient image Transformers yesterday which beats Efficient Net.  </p>\n<p>Link to Paper : <a href=\"https://arxiv.org/abs/2012.12877?fbclid=IwAR0_TgUnnUE08lX1fMjVDQXYW-OQfH_wPy5g_hpB3jSlZwGcnnWUkuN1Ias\" target=\"_blank\">https://arxiv.org/abs/2012.12877?fbclid=IwAR0_TgUnnUE08lX1fMjVDQXYW-OQfH_wPy5g_hpB3jSlZwGcnnWUkuN1Ias</a></p>\n<p>Link to GitHub Repo : <a href=\"https://github.com/facebookresearch/deit?fbclid=IwAR1vZR8hdx4DHJ4eYZFEEXkwuwbR8gYueehiVRiYTzAuXpabKnRFt9VRKNU\" target=\"_blank\">https://github.com/facebookresearch/deit?fbclid=IwAR1vZR8hdx4DHJ4eYZFEEXkwuwbR8gYueehiVRiYTzAuXpabKnRFt9VRKNU</a></p>\n<p>PFA the image showing the improvement over past models. </p>",
  "messages": [
    {
      "id": "1124733",
      "postDate": "12/24/2020 07:04:42",
      "content": "<p>Hi everyone. Facebook just released Data-efficient image Transformers yesterday which beats Efficient Net.  </p>\n<p>Link to Paper : <a href=\"https://arxiv.org/abs/2012.12877?fbclid=IwAR0_TgUnnUE08lX1fMjVDQXYW-OQfH_wPy5g_hpB3jSlZwGcnnWUkuN1Ias\" target=\"_blank\">https://arxiv.org/abs/2012.12877?fbclid=IwAR0_TgUnnUE08lX1fMjVDQXYW-OQfH_wPy5g_hpB3jSlZwGcnnWUkuN1Ias</a></p>\n<p>Link to GitHub Repo : <a href=\"https://github.com/facebookresearch/deit?fbclid=IwAR1vZR8hdx4DHJ4eYZFEEXkwuwbR8gYueehiVRiYTzAuXpabKnRFt9VRKNU\" target=\"_blank\">https://github.com/facebookresearch/deit?fbclid=IwAR1vZR8hdx4DHJ4eYZFEEXkwuwbR8gYueehiVRiYTzAuXpabKnRFt9VRKNU</a></p>\n<p>PFA the image showing the improvement over past models. </p>",
      "rawMarkdown": "Hi everyone. Facebook just released Data-efficient image Transformers yesterday which beats Efficient Net.  \n\nLink to Paper : https://arxiv.org/abs/2012.12877?fbclid=IwAR0_TgUnnUE08lX1fMjVDQXYW-OQfH_wPy5g_hpB3jSlZwGcnnWUkuN1Ias\n\nLink to GitHub Repo : https://github.com/facebookresearch/deit?fbclid=IwAR1vZR8hdx4DHJ4eYZFEEXkwuwbR8gYueehiVRiYTzAuXpabKnRFt9VRKNU\n\n\nPFA the image showing the improvement over past models.",
      "votes": null
    },
    {
      "id": "1124900",
      "postDate": "12/24/2020 09:06:49",
      "content": "<p>Be Carefull when they say they beat EfficientNet . In general, they don't really respect the scaling and training processes in the original EfficientNet paper when doing the comparison. </p>\n<p>At Facebook, they have already claimed beating Effnet With the Regnet Paper.  But all those who retrained the models (like Ross Wightman with timm) show that the latter is way far from beating Effnet. </p>\n<p>And if you use pretrained regnet for fine-tuning , you'll see it performs in general poorly than efficientnet, particularly with higher resolutions than 224x224</p>",
      "rawMarkdown": "Be Carefull when they say they beat EfficientNet . In general, they don't really respect the scaling and training processes in the original EfficientNet paper when doing the comparison. \n\nAt Facebook, they have already claimed beating Effnet With the Regnet Paper.  But all those who retrained the models (like Ross Wightman with timm) show that the latter is way far from beating Effnet. \n\nAnd if you use pretrained regnet for fine-tuning , you'll see it performs in general poorly than efficientnet, particularly with higher resolutions than 224x224",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1124900,
      "author_name": "serigne",
      "author_url": "",
      "post_date": "12/24/2020 09:06:49",
      "content": "<p>Be Carefull when they say they beat EfficientNet . In general, they don't really respect the scaling and training processes in the original EfficientNet paper when doing the comparison. </p>\n<p>At Facebook, they have already claimed beating Effnet With the Regnet Paper.  But all those who retrained the models (like Ross Wightman with timm) show that the latter is way far from beating Effnet. </p>\n<p>And if you use pretrained regnet for fine-tuning , you'll see it performs in general poorly than efficientnet, particularly with higher resolutions than 224x224</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1124733": "Hi everyone. Facebook just released Data-efficient image Transformers yesterday which beats Efficient Net.  \n\nLink to Paper : https://arxiv.org/abs/2012.12877?fbclid=IwAR0_TgUnnUE08lX1fMjVDQXYW-OQfH_wPy5g_hpB3jSlZwGcnnWUkuN1Ias\n\nLink to GitHub Repo : https://github.com/facebookresearch/deit?fbclid=IwAR1vZR8hdx4DHJ4eYZFEEXkwuwbR8gYueehiVRiYTzAuXpabKnRFt9VRKNU\n\n\nPFA the image showing the improvement over past models.",
    "1124900": "Be Carefull when they say they beat EfficientNet . In general, they don't really respect the scaling and training processes in the original EfficientNet paper when doing the comparison. \n\nAt Facebook, they have already claimed beating Effnet With the Regnet Paper.  But all those who retrained the models (like Ross Wightman with timm) show that the latter is way far from beating Effnet. \n\nAnd if you use pretrained regnet for fine-tuning , you'll see it performs in general poorly than efficientnet, particularly with higher resolutions than 224x224"
  },
  "source": "meta"
}