{
  "id": 165039,
  "title": "great ways of using Pre-trained models ",
  "url": "/competitions/siim-isic-melanoma-classification/discussion/165039",
  "author_name": "",
  "post_date": "2020-07-08T10:29:29.196108500Z",
  "votes": 4,
  "comment_count": 4,
  "views": 0,
  "content": "<p>please help me any great way we can make use of pre-trained models or even if you gone through any articles appropriated if you can share here.</p>\n\n<p>Thank you. </p>\n\n<p>base_model = effnet.EfficientNetB6(weights='imagenet',include_top=False, input_shape=(256, 256, 3))\n    for layer in base_model.layers:\n        layer.trainble = False\n    layer = base_model.output\n    layer = L.GlobalAveragePooling2D()(layer)\n    layer = L.Dense(1024,activation='relu')(layer)\n    layer = L.Dropout(0.4)(layer)\n    predictions = L.Dense(1,activation='sigmoid')(layer)\n    model = keras.models.Model(inputs = base_model.input, outputs=predictions)</p>",
  "messages": [
    {
      "id": "920103",
      "postDate": "07/08/2020 10:29:29",
      "content": "<p>please help me any great way we can make use of pre-trained models or even if you gone through any articles appropriated if you can share here.</p>\n\n<p>Thank you. </p>\n\n<p>base_model = effnet.EfficientNetB6(weights='imagenet',include_top=False, input_shape=(256, 256, 3))\n    for layer in base_model.layers:\n        layer.trainble = False\n    layer = base_model.output\n    layer = L.GlobalAveragePooling2D()(layer)\n    layer = L.Dense(1024,activation='relu')(layer)\n    layer = L.Dropout(0.4)(layer)\n    predictions = L.Dense(1,activation='sigmoid')(layer)\n    model = keras.models.Model(inputs = base_model.input, outputs=predictions)</p>",
      "rawMarkdown": "please help me any great way we can make use of pre-trained models or even if you gone through any articles appropriated if you can share here.\n\nThank you. \n\nbase_model = effnet.EfficientNetB6(weights='imagenet',include_top=False, input_shape=(256, 256, 3))\n    for layer in base_model.layers:\n        layer.trainble = False\n    layer = base_model.output\n    layer = L.GlobalAveragePooling2D()(layer)\n    layer = L.Dense(1024,activation='relu')(layer)\n    layer = L.Dropout(0.4)(layer)\n    predictions = L.Dense(1,activation='sigmoid')(layer)\n    model = keras.models.Model(inputs = base_model.input, outputs=predictions)",
      "votes": null
    },
    {
      "id": "920714",
      "postDate": "07/08/2020 19:01:35",
      "content": "<p>You could try introducing multiple models, trained on different image sizes and try ensembling them\nOr you could try to include one more input layer after the efficientnet and introduce the textual data\nEnsembles can get you to a score of 0.93+ if you weigh them correctly.\nDo try it out!!</p>",
      "rawMarkdown": "You could try introducing multiple models, trained on different image sizes and try ensembling them\nOr you could try to include one more input layer after the efficientnet and introduce the textual data\nEnsembles can get you to a score of 0.93+ if you weigh them correctly.\nDo try it out!!",
      "votes": null
    },
    {
      "id": "921499",
      "postDate": "07/09/2020 10:38:32",
      "content": "<p>Here is a previous year top solution - <a href=\"https://www.kaggle.com/c/siim-isic-melanoma-classification/discussion/154683\">https://www.kaggle.com/c/siim-isic-melanoma-classification/discussion/154683</a>\nIf you explore the paper, you can find some cool architecture insights</p>",
      "rawMarkdown": "Here is a previous year top solution - https://www.kaggle.com/c/siim-isic-melanoma-classification/discussion/154683\nIf you explore the paper, you can find some cool architecture insights",
      "votes": null
    },
    {
      "id": "927559",
      "postDate": "07/13/2020 13:34:46",
      "content": "<p>use noisy-student can improve your Model</p>",
      "rawMarkdown": "use noisy-student can improve your Model",
      "votes": null
    },
    {
      "id": "945047",
      "postDate": "07/25/2020 14:33:19",
      "content": "<p>Any kernal we have for example?</p>",
      "rawMarkdown": "Any kernal we have for example?",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 920714,
      "author_name": "aryanpandey1109",
      "author_url": "",
      "post_date": "07/08/2020 19:01:35",
      "content": "<p>You could try introducing multiple models, trained on different image sizes and try ensembling them\nOr you could try to include one more input layer after the efficientnet and introduce the textual data\nEnsembles can get you to a score of 0.93+ if you weigh them correctly.\nDo try it out!!</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 921499,
      "author_name": "vladimirsydor",
      "author_url": "",
      "post_date": "07/09/2020 10:38:32",
      "content": "<p>Here is a previous year top solution - <a href=\"https://www.kaggle.com/c/siim-isic-melanoma-classification/discussion/154683\">https://www.kaggle.com/c/siim-isic-melanoma-classification/discussion/154683</a>\nIf you explore the paper, you can find some cool architecture insights</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 927559,
      "author_name": "doanquanvietnamca",
      "author_url": "",
      "post_date": "07/13/2020 13:34:46",
      "content": "<p>use noisy-student can improve your Model</p>",
      "votes": null,
      "replies": [
        {
          "id": 945047,
          "author_name": "kunduruanil",
          "author_url": "",
          "post_date": "07/25/2020 14:33:19",
          "content": "<p>Any kernal we have for example?</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "920103": "please help me any great way we can make use of pre-trained models or even if you gone through any articles appropriated if you can share here.\n\nThank you. \n\nbase_model = effnet.EfficientNetB6(weights='imagenet',include_top=False, input_shape=(256, 256, 3))\n    for layer in base_model.layers:\n        layer.trainble = False\n    layer = base_model.output\n    layer = L.GlobalAveragePooling2D()(layer)\n    layer = L.Dense(1024,activation='relu')(layer)\n    layer = L.Dropout(0.4)(layer)\n    predictions = L.Dense(1,activation='sigmoid')(layer)\n    model = keras.models.Model(inputs = base_model.input, outputs=predictions)",
    "920714": "You could try introducing multiple models, trained on different image sizes and try ensembling them\nOr you could try to include one more input layer after the efficientnet and introduce the textual data\nEnsembles can get you to a score of 0.93+ if you weigh them correctly.\nDo try it out!!",
    "921499": "Here is a previous year top solution - https://www.kaggle.com/c/siim-isic-melanoma-classification/discussion/154683\nIf you explore the paper, you can find some cool architecture insights",
    "927559": "use noisy-student can improve your Model",
    "945047": "Any kernal we have for example?"
  },
  "source": "meta"
}