{
  "id": 159257,
  "title": "EfficientNet Size x Performance",
  "url": "/competitions/siim-isic-melanoma-classification/discussion/159257",
  "author_name": "Fred Dias",
  "post_date": "2020-06-17T00:22:20.763000",
  "votes": 3,
  "comment_count": 0,
  "views": 0,
  "content": "<p>After a looooong time running TPUs, I analyzed the influence of the model size on performance (image below). It is interesting to note that there really is an improvement with the increase in the model size. I hope the results can be useful to the competitors! 😊 </p>\n\n<p>The results are all compiled <a href=\"https://www.kaggle.com/fredericods/from-efficientb0-to-b7-melanoma-classification\">here</a>. The kernel used to train the model is <a href=\"https://www.kaggle.com/fredericods/efficientnetbi-melanoma-classification-with-tf\">here</a>.</p>\n\n<p>I achieved 0.916 on Public Leaderboard with EfficientNetB3 (CV Average).</p>\n\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2590760%2F83ad4c9c7c5759c068ca078e55133eb6%2FPerformance%20x%20Model.PNG?generation=1592352782970921&amp;alt=media\" alt=\"EfficientNet Size x Performance\"></p>",
  "messages": [
    {
      "id": 889399,
      "postDate": "2020-06-17T00:22:20.763Z",
      "content": "<p>After a looooong time running TPUs, I analyzed the influence of the model size on performance (image below). It is interesting to note that there really is an improvement with the increase in the model size. I hope the results can be useful to the competitors! 😊 </p>\n\n<p>The results are all compiled <a href=\"https://www.kaggle.com/fredericods/from-efficientb0-to-b7-melanoma-classification\">here</a>. The kernel used to train the model is <a href=\"https://www.kaggle.com/fredericods/efficientnetbi-melanoma-classification-with-tf\">here</a>.</p>\n\n<p>I achieved 0.916 on Public Leaderboard with EfficientNetB3 (CV Average).</p>\n\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2590760%2F83ad4c9c7c5759c068ca078e55133eb6%2FPerformance%20x%20Model.PNG?generation=1592352782970921&amp;alt=media\" alt=\"EfficientNet Size x Performance\"></p>",
      "rawMarkdown": "After a looooong time running TPUs, I analyzed the influence of the model size on performance (image below). It is interesting to note that there really is an improvement with the increase in the model size. I hope the results can be useful to the competitors! 😊 \n\nThe results are all compiled [here](https://www.kaggle.com/fredericods/from-efficientb0-to-b7-melanoma-classification). The kernel used to train the model is [here](https://www.kaggle.com/fredericods/efficientnetbi-melanoma-classification-with-tf).\n\nI achieved 0.916 on Public Leaderboard with EfficientNetB3 (CV Average).\n\n![EfficientNet Size x Performance](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2590760%2F83ad4c9c7c5759c068ca078e55133eb6%2FPerformance%20x%20Model.PNG?generation=1592352782970921&amp;alt=media)\n",
      "votes": 3
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "889399": "After a looooong time running TPUs, I analyzed the influence of the model size on performance (image below). It is interesting to note that there really is an improvement with the increase in the model size. I hope the results can be useful to the competitors! 😊 \n\nThe results are all compiled [here](https://www.kaggle.com/fredericods/from-efficientb0-to-b7-melanoma-classification). The kernel used to train the model is [here](https://www.kaggle.com/fredericods/efficientnetbi-melanoma-classification-with-tf).\n\nI achieved 0.916 on Public Leaderboard with EfficientNetB3 (CV Average).\n\n![EfficientNet Size x Performance](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2590760%2F83ad4c9c7c5759c068ca078e55133eb6%2FPerformance%20x%20Model.PNG?generation=1592352782970921&amp;alt=media)\n"
  }
}