{
  "id": 40522,
  "title": "Pretrained keras models available for Kernels",
  "url": "/competitions/dog-breed-identification/discussion/40522",
  "author_name": "beluga",
  "post_date": "2017-10-03T22:27:46.228000",
  "votes": 11,
  "comment_count": 5,
  "views": 0,
  "content": "<p>I uploaded a few pretrained keras models into a kaggle dataset. You could load pretrained weights from here: <a href=\"https://www.kaggle.com/gaborfodor/keras-pretrained-models\">https://www.kaggle.com/gaborfodor/keras-pretrained-models</a></p>\n\n<p>Example usage: <a href=\"https://www.kaggle.com/gaborfodor/example-usage\">https://www.kaggle.com/gaborfodor/example-usage</a></p>\n\n<p>Example usage for this competition: <a href=\"https://www.kaggle.com/gaborfodor/use-pretrained-keras-models-lb-0-3\">https://www.kaggle.com/gaborfodor/use-pretrained-keras-models-lb-0-3</a></p>",
  "messages": [
    {
      "id": 227193,
      "postDate": "2017-10-03T22:27:46.227Z",
      "content": "<p>I uploaded a few pretrained keras models into a kaggle dataset. You could load pretrained weights from here: <a href=\"https://www.kaggle.com/gaborfodor/keras-pretrained-models\">https://www.kaggle.com/gaborfodor/keras-pretrained-models</a></p>\n\n<p>Example usage: <a href=\"https://www.kaggle.com/gaborfodor/example-usage\">https://www.kaggle.com/gaborfodor/example-usage</a></p>\n\n<p>Example usage for this competition: <a href=\"https://www.kaggle.com/gaborfodor/use-pretrained-keras-models-lb-0-3\">https://www.kaggle.com/gaborfodor/use-pretrained-keras-models-lb-0-3</a></p>",
      "rawMarkdown": "I uploaded a few pretrained keras models into a kaggle dataset. You could load pretrained weights from here: https://www.kaggle.com/gaborfodor/keras-pretrained-models\n\nExample usage: https://www.kaggle.com/gaborfodor/example-usage\n\nExample usage for this competition: https://www.kaggle.com/gaborfodor/use-pretrained-keras-models-lb-0-3",
      "votes": 11
    },
    {
      "id": 230892,
      "postDate": "2017-10-13T03:12:50.817Z",
      "content": "<p>thanks, saved me a lot of time struggling with models downloading limits</p>",
      "rawMarkdown": "thanks, saved me a lot of time struggling with models downloading limits"
    },
    {
      "id": 234331,
      "postDate": "2017-10-23T01:18:53.237Z",
      "rawMarkdown": "",
      "isDeleted": true,
      "replies": [
        {
          "id": 234428,
          "postDate": "2017-10-23T09:28:05.490Z",
          "content": "<p>Same for me. Fine tuning convolutional layers decrease performance. But maybe thats because I do not know how to train them properly. Jerry Howard said on one of fast.ai lectures that tuning convolutional layers are rather rare case. And here our train/test datasets are from the same set on which all imagenet models were build. For some reason for me resnet50 does not perform even close to inception/xception, but I suspect some errors with resnet images preprocessing type. </p>",
          "rawMarkdown": "Same for me. Fine tuning convolutional layers decrease performance. But maybe thats because I do not know how to train them properly. Jerry Howard said on one of fast.ai lectures that tuning convolutional layers are rather rare case. And here our train/test datasets are from the same set on which all imagenet models were build. For some reason for me resnet50 does not perform even close to inception/xception, but I suspect some errors with resnet images preprocessing type. ",
          "votes": 2
        },
        {
          "id": 234540,
          "postDate": "2017-10-23T15:06:01.103Z",
          "rawMarkdown": "",
          "isDeleted": true
        }
      ]
    },
    {
      "id": 227294,
      "postDate": "2017-10-04T05:49:29.080Z",
      "content": "<p>Nice, Thank beluga</p>",
      "rawMarkdown": "Nice, Thank beluga"
    }
  ],
  "comments": [
    {
      "id": 230892,
      "author_name": "Nico Belov",
      "author_url": "",
      "post_date": "2017-10-13T03:12:50.817000",
      "content": "<p>thanks, saved me a lot of time struggling with models downloading limits</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 234331,
      "author_name": "",
      "author_url": "",
      "post_date": "2017-10-23T01:18:53.237000",
      "content": "",
      "votes": 0,
      "replies": [
        {
          "id": 234428,
          "author_name": "SM",
          "author_url": "",
          "post_date": "2017-10-23T09:28:05.490000",
          "content": "<p>Same for me. Fine tuning convolutional layers decrease performance. But maybe thats because I do not know how to train them properly. Jerry Howard said on one of fast.ai lectures that tuning convolutional layers are rather rare case. And here our train/test datasets are from the same set on which all imagenet models were build. For some reason for me resnet50 does not perform even close to inception/xception, but I suspect some errors with resnet images preprocessing type. </p>",
          "votes": 2,
          "replies": []
        },
        {
          "id": 234540,
          "author_name": "",
          "author_url": "",
          "post_date": "2017-10-23T15:06:01.103000",
          "content": "",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 227294,
      "author_name": "SM",
      "author_url": "",
      "post_date": "2017-10-04T05:49:29.080000",
      "content": "<p>Nice, Thank beluga</p>",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "227193": "I uploaded a few pretrained keras models into a kaggle dataset. You could load pretrained weights from here: https://www.kaggle.com/gaborfodor/keras-pretrained-models\n\nExample usage: https://www.kaggle.com/gaborfodor/example-usage\n\nExample usage for this competition: https://www.kaggle.com/gaborfodor/use-pretrained-keras-models-lb-0-3",
    "230892": "thanks, saved me a lot of time struggling with models downloading limits",
    "234331": "",
    "227294": "Nice, Thank beluga"
  }
}