{
  "id": 198260,
  "title": "Keras pretrained models dataset",
  "url": "/competitions/cassava-leaf-disease-classification/discussion/198260",
  "author_name": "",
  "post_date": "2020-11-20T13:16:02.524521600Z",
  "votes": 5,
  "comment_count": 3,
  "views": 0,
  "content": "<p>Since code competitions do not allow Internet, I created and uploaded all Keras pretrained models in this dataset: <a href=\"https://www.kaggle.com/aeryss/keras-pretrained-models\" target=\"_blank\">https://www.kaggle.com/aeryss/keras-pretrained-models</a></p>\n<p>You can load the model using this line:</p>\n<pre><code>model = tf.keras.models.load_model(\"[Name]_[A]_[B].h5\")\n</code></pre>\n<p>where <code>Name</code> is the model name, <code>A</code> is to include or not include top (either <code>Top</code> or <code>NoTop</code>), and <code>[B]</code> is its weight (either <code>ImageNet</code> or <code>None</code>).</p>\n<p>For example, to load the EfficientNetB4 without top and ImageNet weights:</p>\n<pre><code>model = tf.keras.models.load_model(\"EfficientNetB4_NoTop_ImageNet.h5\")\n</code></pre>\n<p>Note that the <code>pooling</code> parameter is <code>None</code> for all files. </p>",
  "messages": [
    {
      "id": "1084858",
      "postDate": "11/20/2020 13:16:02",
      "content": "<p>Since code competitions do not allow Internet, I created and uploaded all Keras pretrained models in this dataset: <a href=\"https://www.kaggle.com/aeryss/keras-pretrained-models\" target=\"_blank\">https://www.kaggle.com/aeryss/keras-pretrained-models</a></p>\n<p>You can load the model using this line:</p>\n<pre><code>model = tf.keras.models.load_model(\"[Name]_[A]_[B].h5\")\n</code></pre>\n<p>where <code>Name</code> is the model name, <code>A</code> is to include or not include top (either <code>Top</code> or <code>NoTop</code>), and <code>[B]</code> is its weight (either <code>ImageNet</code> or <code>None</code>).</p>\n<p>For example, to load the EfficientNetB4 without top and ImageNet weights:</p>\n<pre><code>model = tf.keras.models.load_model(\"EfficientNetB4_NoTop_ImageNet.h5\")\n</code></pre>\n<p>Note that the <code>pooling</code> parameter is <code>None</code> for all files. </p>",
      "rawMarkdown": "Since code competitions do not allow Internet, I created and uploaded all Keras pretrained models in this dataset: https://www.kaggle.com/aeryss/keras-pretrained-models\n\nYou can load the model using this line:\n```\nmodel = tf.keras.models.load_model(\"[Name]_[A]_[B].h5\")\n```\nwhere `Name` is the model name, `A` is to include or not include top (either `Top` or `NoTop`), and `[B]` is its weight (either `ImageNet` or `None`).\n\nFor example, to load the EfficientNetB4 without top and ImageNet weights:\n```\nmodel = tf.keras.models.load_model(\"EfficientNetB4_NoTop_ImageNet.h5\")\n```\nNote that the `pooling` parameter is `None` for all files.",
      "votes": null
    },
    {
      "id": "1084951",
      "postDate": "11/20/2020 14:43:05",
      "content": "<p>thx - the internet can be used for training - you can load your trained models into a dataset and use them for inference</p>",
      "rawMarkdown": "thx - the internet can be used for training - you can load your trained models into a dataset and use them for inference",
      "votes": null
    },
    {
      "id": "1085657",
      "postDate": "11/21/2020 04:18:06",
      "content": "<p>TPU does not work with TF greater than 2.2, while EfficientNet was introduced in 2.3. This is a short workaround. From: <a href=\"https://github.com/tensorflow/tensorflow/issues/41075\" target=\"_blank\">https://github.com/tensorflow/tensorflow/issues/41075</a></p>",
      "rawMarkdown": "TPU does not work with TF greater than 2.2, while EfficientNet was introduced in 2.3. This is a short workaround. From: https://github.com/tensorflow/tensorflow/issues/41075",
      "votes": null
    },
    {
      "id": "1093151",
      "postDate": "11/27/2020 13:46:30",
      "content": "<p>Thanks a Lot</p>",
      "rawMarkdown": "Thanks a Lot",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1084951,
      "author_name": "romanweilguny",
      "author_url": "",
      "post_date": "11/20/2020 14:43:05",
      "content": "<p>thx - the internet can be used for training - you can load your trained models into a dataset and use them for inference</p>",
      "votes": null,
      "replies": [
        {
          "id": 1085657,
          "author_name": "aeryss",
          "author_url": "",
          "post_date": "11/21/2020 04:18:06",
          "content": "<p>TPU does not work with TF greater than 2.2, while EfficientNet was introduced in 2.3. This is a short workaround. From: <a href=\"https://github.com/tensorflow/tensorflow/issues/41075\" target=\"_blank\">https://github.com/tensorflow/tensorflow/issues/41075</a></p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1093151,
      "author_name": "jamesbond00700",
      "author_url": "",
      "post_date": "11/27/2020 13:46:30",
      "content": "<p>Thanks a Lot</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1084858": "Since code competitions do not allow Internet, I created and uploaded all Keras pretrained models in this dataset: https://www.kaggle.com/aeryss/keras-pretrained-models\n\nYou can load the model using this line:\n```\nmodel = tf.keras.models.load_model(\"[Name]_[A]_[B].h5\")\n```\nwhere `Name` is the model name, `A` is to include or not include top (either `Top` or `NoTop`), and `[B]` is its weight (either `ImageNet` or `None`).\n\nFor example, to load the EfficientNetB4 without top and ImageNet weights:\n```\nmodel = tf.keras.models.load_model(\"EfficientNetB4_NoTop_ImageNet.h5\")\n```\nNote that the `pooling` parameter is `None` for all files.",
    "1084951": "thx - the internet can be used for training - you can load your trained models into a dataset and use them for inference",
    "1085657": "TPU does not work with TF greater than 2.2, while EfficientNet was introduced in 2.3. This is a short workaround. From: https://github.com/tensorflow/tensorflow/issues/41075",
    "1093151": "Thanks a Lot"
  },
  "source": "meta"
}