ImageClassifierCNN-EfficientNet.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n",
      "/home/greg/.local/lib/python3.6/site-packages/tensorflow/python/framework/dtypes.py:516: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_qint8 = np.dtype([(\"qint8\", np.int8, 1)])\n",
      "/home/greg/.local/lib/python3.6/site-packages/tensorflow/python/framework/dtypes.py:517: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_quint8 = np.dtype([(\"quint8\", np.uint8, 1)])\n",
      "/home/greg/.local/lib/python3.6/site-packages/tensorflow/python/framework/dtypes.py:518: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_qint16 = np.dtype([(\"qint16\", np.int16, 1)])\n",
      "/home/greg/.local/lib/python3.6/site-packages/tensorflow/python/framework/dtypes.py:519: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_quint16 = np.dtype([(\"quint16\", np.uint16, 1)])\n",
      "/home/greg/.local/lib/python3.6/site-packages/tensorflow/python/framework/dtypes.py:520: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_qint32 = np.dtype([(\"qint32\", np.int32, 1)])\n",
      "/home/greg/.local/lib/python3.6/site-packages/tensorflow/python/framework/dtypes.py:525: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  np_resource = np.dtype([(\"resource\", np.ubyte, 1)])\n",
      "/home/greg/.local/lib/python3.6/site-packages/tensorboard/compat/tensorflow_stub/dtypes.py:541: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_qint8 = np.dtype([(\"qint8\", np.int8, 1)])\n",
      "/home/greg/.local/lib/python3.6/site-packages/tensorboard/compat/tensorflow_stub/dtypes.py:542: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_quint8 = np.dtype([(\"quint8\", np.uint8, 1)])\n",
      "/home/greg/.local/lib/python3.6/site-packages/tensorboard/compat/tensorflow_stub/dtypes.py:543: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_qint16 = np.dtype([(\"qint16\", np.int16, 1)])\n",
      "/home/greg/.local/lib/python3.6/site-packages/tensorboard/compat/tensorflow_stub/dtypes.py:544: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_quint16 = np.dtype([(\"quint16\", np.uint16, 1)])\n",
      "/home/greg/.local/lib/python3.6/site-packages/tensorboard/compat/tensorflow_stub/dtypes.py:545: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_qint32 = np.dtype([(\"qint32\", np.int32, 1)])\n",
      "/home/greg/.local/lib/python3.6/site-packages/tensorboard/compat/tensorflow_stub/dtypes.py:550: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  np_resource = np.dtype([(\"resource\", np.ubyte, 1)])\n"
     ]
    }
   ],
   "source": [
    "%matplotlib inline\n",
    "import matplotlib.pyplot as plt\n",
    "import matplotlib.image as mpimg\n",
    "from keras.preprocessing.image import ImageDataGenerator\n",
    "from keras.models import Sequential\n",
    "from keras.optimizers import RMSprop\n",
    "from keras.layers import Dense, Flatten, GlobalMaxPooling2D\n",
    "from keras.callbacks import CSVLogger\n",
    "from livelossplot.keras import PlotLossesCallback\n",
    "import efficientnet.keras as efn"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "TRAINING_LOGS_FILE = \"training_logs.csv\"\n",
    "MODEL_SUMMARY_FILE = \"model_summary.txt\"\n",
    "MODEL_FILE = \"cats_vs_dogs.h5\"\n",
    "\n",
    "# Data\n",
    "path = \"/home/greg/datasets/cats_and_dogs/\"\n",
    "training_data_dir = path + \"training\" # 10 000 * 2\n",
    "validation_data_dir = path + \"validation\" # 2 500 * 2\n",
    "test_data_dir = path + \"test\" # 12 500"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Hyperparams\n",
    "IMAGE_SIZE = 200\n",
    "IMAGE_WIDTH, IMAGE_HEIGHT = IMAGE_SIZE, IMAGE_SIZE\n",
    "EPOCHS = 20\n",
    "BATCH_SIZE = 32\n",
    "TEST_SIZE = 30\n",
    "\n",
    "input_shape = (IMAGE_WIDTH, IMAGE_HEIGHT, 3)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:Large dropout rate: 0.5125 (>0.5). In TensorFlow 2.x, dropout() uses dropout rate instead of keep_prob. Please ensure that this is intended.\n",
      "WARNING:tensorflow:Large dropout rate: 0.525 (>0.5). In TensorFlow 2.x, dropout() uses dropout rate instead of keep_prob. Please ensure that this is intended.\n",
      "WARNING:tensorflow:Large dropout rate: 0.5375 (>0.5). In TensorFlow 2.x, dropout() uses dropout rate instead of keep_prob. Please ensure that this is intended.\n",
      "WARNING:tensorflow:Large dropout rate: 0.55 (>0.5). In TensorFlow 2.x, dropout() uses dropout rate instead of keep_prob. Please ensure that this is intended.\n",
      "WARNING:tensorflow:Large dropout rate: 0.5625 (>0.5). In TensorFlow 2.x, dropout() uses dropout rate instead of keep_prob. Please ensure that this is intended.\n",
      "0\n",
      "<keras.engine.input_layer.InputLayer object at 0x7f18a3542cf8>\n",
      "1\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f18a3542fd0>\n",
      "2\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f18a354a240>\n",
      "3\n",
      "<keras.layers.core.Activation object at 0x7f18a354a6a0>\n",
      "4\n",
      "<keras.layers.convolutional.DepthwiseConv2D object at 0x7f18a354a668>\n",
      "5\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f18a0174d68>\n",
      "6\n",
      "<keras.layers.core.Activation object at 0x7f18a0115668>\n",
      "7\n",
      "<keras.layers.pooling.GlobalAveragePooling2D object at 0x7f18a009ef60>\n",
      "8\n",
      "<keras.layers.core.Reshape object at 0x7f18a01343c8>\n",
      "9\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f18a0134668>\n",
      "10\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f18907c9240>\n",
      "11\n",
      "<keras.layers.merge.Multiply object at 0x7f189078cc50>\n",
      "12\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f189078cb38>\n",
      "13\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f18907468d0>\n",
      "14\n",
      "<keras.layers.convolutional.DepthwiseConv2D object at 0x7f189078cfd0>\n",
      "15\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f1890704eb8>\n",
      "16\n",
      "<keras.layers.core.Activation object at 0x7f18906f54a8>\n",
      "17\n",
      "<keras.layers.pooling.GlobalAveragePooling2D object at 0x7f1890614c18>\n",
      "18\n",
      "<keras.layers.core.Reshape object at 0x7f1890697f98>\n",
      "19\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f18906b0048>\n",
      "20\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1890584f28>\n",
      "21\n",
      "<keras.layers.merge.Multiply object at 0x7f189059be10>\n",
      "22\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1890541780>\n",
      "23\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f18904fe860>\n",
      "24\n",
      "<efficientnet.model.get_dropout.<locals>.FixedDropout object at 0x7f1890541c88>\n",
      "25\n",
      "<keras.layers.merge.Add object at 0x7f18904d9ef0>\n",
      "26\n",
      "<keras.layers.convolutional.DepthwiseConv2D object at 0x7f18904aa438>\n",
      "27\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f1890463b00>\n",
      "28\n",
      "<keras.layers.core.Activation object at 0x7f189051f2e8>\n",
      "29\n",
      "<keras.layers.pooling.GlobalAveragePooling2D object at 0x7f1890396e48>\n",
      "30\n",
      "<keras.layers.core.Reshape object at 0x7f18903bb9e8>\n",
      "31\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f18903bb710>\n",
      "32\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f189031f828>\n",
      "33\n",
      "<keras.layers.merge.Multiply object at 0x7f18902c6e10>\n",
      "34\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f18902c69e8>\n",
      "35\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f1890284978>\n",
      "36\n",
      "<efficientnet.model.get_dropout.<locals>.FixedDropout object at 0x7f18902c6f28>\n",
      "37\n",
      "<keras.layers.merge.Add object at 0x7f189025ef98>\n",
      "38\n",
      "<keras.layers.convolutional.DepthwiseConv2D object at 0x7f18902a2358>\n",
      "39\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f18901d17b8>\n",
      "40\n",
      "<keras.layers.core.Activation object at 0x7f189022f898>\n",
      "41\n",
      "<keras.layers.pooling.GlobalAveragePooling2D object at 0x7f18900c43c8>\n",
      "42\n",
      "<keras.layers.core.Reshape object at 0x7f18901adb00>\n",
      "43\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f18900884a8>\n",
      "44\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f18900a4a58>\n",
      "45\n",
      "<keras.layers.merge.Multiply object at 0x7f189004a278>\n",
      "46\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f189004a2b0>\n",
      "47\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f1890066b38>\n",
      "48\n",
      "<efficientnet.model.get_dropout.<locals>.FixedDropout object at 0x7f1890066dd8>\n",
      "49\n",
      "<keras.layers.merge.Add object at 0x7f18885e4a58>\n",
      "50\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f18885bcef0>\n",
      "51\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f188852dac8>\n",
      "52\n",
      "<keras.layers.core.Activation object at 0x7f188859d7b8>\n",
      "53\n",
      "<keras.layers.convolutional.DepthwiseConv2D object at 0x7f188850d9e8>\n",
      "54\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f1888423cf8>\n",
      "55\n",
      "<keras.layers.core.Activation object at 0x7f188844af28>\n",
      "56\n",
      "<keras.layers.pooling.GlobalAveragePooling2D object at 0x7f1888403390>\n",
      "57\n",
      "<keras.layers.core.Reshape object at 0x7f18883eb240>\n",
      "58\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1888336c18>\n",
      "59\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f18882d2908>\n",
      "60\n",
      "<keras.layers.merge.Multiply object at 0x7f18882f8908>\n",
      "61\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f18882f8eb8>\n",
      "62\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f18882b2dd8>\n",
      "63\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f18882b24a8>\n",
      "64\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f1888290400>\n",
      "65\n",
      "<keras.layers.core.Activation object at 0x7f18881dd278>\n",
      "66\n",
      "<keras.layers.convolutional.DepthwiseConv2D object at 0x7f1888182dd8>\n",
      "67\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f18880f1518>\n",
      "68\n",
      "<keras.layers.core.Activation object at 0x7f1888148cc0>\n",
      "69\n",
      "<keras.layers.pooling.GlobalAveragePooling2D object at 0x7f188806ec18>\n",
      "70\n",
      "<keras.layers.core.Reshape object at 0x7f188810b048>\n",
      "71\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f188810b320>\n",
      "72\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f184c7c7f28>\n",
      "73\n",
      "<keras.layers.merge.Multiply object at 0x7f184c7dbeb8>\n",
      "74\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f184c7dbef0>\n",
      "75\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f184c741828>\n",
      "76\n",
      "<efficientnet.model.get_dropout.<locals>.FixedDropout object at 0x7f184c787dd8>\n",
      "77\n",
      "<keras.layers.merge.Add object at 0x7f184c6c87f0>\n",
      "78\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f184c76a208>\n",
      "79\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f184c693f98>\n",
      "80\n",
      "<keras.layers.core.Activation object at 0x7f184c6ffe48>\n",
      "81\n",
      "<keras.layers.convolutional.DepthwiseConv2D object at 0x7f184c5e1f28>\n",
      "82\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f184c5bd0f0>\n",
      "83\n",
      "<keras.layers.core.Activation object at 0x7f184c5afd30>\n",
      "84\n",
      "<keras.layers.pooling.GlobalAveragePooling2D object at 0x7f184c4d2860>\n",
      "85\n",
      "<keras.layers.core.Reshape object at 0x7f184c54a438>\n",
      "86\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f184c497978>\n",
      "87\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f184c4b52e8>\n",
      "88\n",
      "<keras.layers.merge.Multiply object at 0x7f184c45e668>\n",
      "89\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f184c45ea58>\n",
      "90\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f184c41ac50>\n",
      "91\n",
      "<efficientnet.model.get_dropout.<locals>.FixedDropout object at 0x7f184c45e710>\n",
      "92\n",
      "<keras.layers.merge.Add object at 0x7f184c3d3e10>\n",
      "93\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f184c340ef0>\n",
      "94\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f184c364f60>\n",
      "95\n",
      "<keras.layers.core.Activation object at 0x7f184c3f2630>\n",
      "96\n",
      "<keras.layers.convolutional.DepthwiseConv2D object at 0x7f184c2b3e10>\n",
      "97\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f184c220550>\n",
      "98\n",
      "<keras.layers.core.Activation object at 0x7f184c27cb38>\n",
      "99\n",
      "<keras.layers.pooling.GlobalAveragePooling2D object at 0x7f184c19ec18>\n",
      "100\n",
      "<keras.layers.core.Reshape object at 0x7f184c23c0b8>\n",
      "101\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f184c23c358>\n",
      "102\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f184c110f28>\n",
      "103\n",
      "<keras.layers.merge.Multiply object at 0x7f184c122eb8>\n",
      "104\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f184c122ef0>\n",
      "105\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f184c089828>\n",
      "106\n",
      "<efficientnet.model.get_dropout.<locals>.FixedDropout object at 0x7f184c0d1dd8>\n",
      "107\n",
      "<keras.layers.merge.Add object at 0x7f184c0b0208>\n",
      "108\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f184c0b02b0>\n",
      "109\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f182c795f60>\n",
      "110\n",
      "<keras.layers.core.Activation object at 0x7f184c047e48>\n",
      "111\n",
      "<keras.layers.convolutional.DepthwiseConv2D object at 0x7f182c6e3f28>\n",
      "112\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f182c7028d0>\n",
      "113\n",
      "<keras.layers.core.Activation object at 0x7f182c6b1d68>\n",
      "114\n",
      "<keras.layers.pooling.GlobalAveragePooling2D object at 0x7f182c629f98>\n",
      "115\n",
      "<keras.layers.core.Reshape object at 0x7f182c64f470>\n",
      "116\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f182c59a978>\n",
      "117\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f182c5b22e8>\n",
      "118\n",
      "<keras.layers.merge.Multiply object at 0x7f182c564668>\n",
      "119\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f182c564a58>\n",
      "120\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f182c520c50>\n",
      "121\n",
      "<efficientnet.model.get_dropout.<locals>.FixedDropout object at 0x7f182c564710>\n",
      "122\n",
      "<keras.layers.merge.Add object at 0x7f182c496ef0>\n",
      "123\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f182c532b00>\n",
      "124\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f182c464fd0>\n",
      "125\n",
      "<keras.layers.core.Activation object at 0x7f182c4f5cc0>\n",
      "126\n",
      "<keras.layers.convolutional.DepthwiseConv2D object at 0x7f182c3b5e10>\n",
      "127\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f182c325550>\n",
      "128\n",
      "<keras.layers.core.Activation object at 0x7f182c300b38>\n",
      "129\n",
      "<keras.layers.pooling.GlobalAveragePooling2D object at 0x7f182c2a1c18>\n",
      "130\n",
      "<keras.layers.core.Reshape object at 0x7f182c2c10b8>\n",
      "131\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f182c2c1358>\n",
      "132\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f182c20df28>\n",
      "133\n",
      "<keras.layers.merge.Multiply object at 0x7f182c226eb8>\n",
      "134\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f182c226ef0>\n",
      "135\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f182c1e8b70>\n",
      "136\n",
      "<efficientnet.model.get_dropout.<locals>.FixedDropout object at 0x7f182c1e8f60>\n",
      "137\n",
      "<keras.layers.merge.Add object at 0x7f182c18a7f0>\n",
      "138\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f182c1147f0>\n",
      "139\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f182c0dcf98>\n",
      "140\n",
      "<keras.layers.core.Activation object at 0x7f182c149e48>\n",
      "141\n",
      "<keras.layers.convolutional.DepthwiseConv2D object at 0x7f18247eaf28>\n",
      "142\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f18247c60f0>\n",
      "143\n",
      "<keras.layers.core.Activation object at 0x7f18247b7d30>\n",
      "144\n",
      "<keras.layers.pooling.GlobalAveragePooling2D object at 0x7f1824733f98>\n",
      "145\n",
      "<keras.layers.core.Reshape object at 0x7f1824757438>\n",
      "146\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f18246a39b0>\n",
      "147\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f182463d6a0>\n",
      "148\n",
      "<keras.layers.merge.Multiply object at 0x7f182466a6d8>\n",
      "149\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f182466a710>\n",
      "150\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f182462ac50>\n",
      "151\n",
      "<efficientnet.model.get_dropout.<locals>.FixedDropout object at 0x7f182466aa58>\n",
      "152\n",
      "<keras.layers.merge.Add object at 0x7f182459fef0>\n",
      "153\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f18245be978>\n",
      "154\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f182456cfd0>\n",
      "155\n",
      "<keras.layers.core.Activation object at 0x7f182457fcc0>\n",
      "156\n",
      "<keras.layers.convolutional.DepthwiseConv2D object at 0x7f182443ee10>\n",
      "157\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f182442c550>\n",
      "158\n",
      "<keras.layers.core.Activation object at 0x7f1824409b38>\n",
      "159\n",
      "<keras.layers.pooling.GlobalAveragePooling2D object at 0x7f18243a9c18>\n",
      "160\n",
      "<keras.layers.core.Reshape object at 0x7f18243c80b8>\n",
      "161\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f18243c8358>\n",
      "162\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f182431af28>\n",
      "163\n",
      "<keras.layers.merge.Multiply object at 0x7f1824331eb8>\n",
      "164\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1824331ef0>\n",
      "165\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f1824295828>\n",
      "166\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f18242d9dd8>\n",
      "167\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f1824252550>\n",
      "168\n",
      "<keras.layers.core.Activation object at 0x7f18241c0c88>\n",
      "169\n",
      "<keras.layers.convolutional.DepthwiseConv2D object at 0x7f182413d8d0>\n",
      "170\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f18241a2ef0>\n",
      "171\n",
      "<keras.layers.core.Activation object at 0x7f1824132ba8>\n",
      "172\n",
      "<keras.layers.pooling.GlobalAveragePooling2D object at 0x7f18240aafd0>\n",
      "173\n",
      "<keras.layers.core.Reshape object at 0x7f18240ce2b0>\n",
      "174\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f18147de828>\n",
      "175\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f18147f6e10>\n",
      "176\n",
      "<keras.layers.merge.Multiply object at 0x7f18147a10b8>\n",
      "177\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f18147b95c0>\n",
      "178\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f181475d7b8>\n",
      "179\n",
      "<efficientnet.model.get_dropout.<locals>.FixedDropout object at 0x7f18147a18d0>\n",
      "180\n",
      "<keras.layers.merge.Add object at 0x7f1814716e48>\n",
      "181\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1814683d68>\n",
      "182\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f18146a7e10>\n",
      "183\n",
      "<keras.layers.core.Activation object at 0x7f1814732320>\n",
      "184\n",
      "<keras.layers.convolutional.DepthwiseConv2D object at 0x7f18145a2470>\n",
      "185\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f1814564208>\n",
      "186\n",
      "<keras.layers.core.Activation object at 0x7f181460af28>\n",
      "187\n",
      "<keras.layers.pooling.GlobalAveragePooling2D object at 0x7f1814521c18>\n",
      "188\n",
      "<keras.layers.core.Reshape object at 0x7f18144bdc88>\n",
      "189\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f18144b0da0>\n",
      "190\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f181444ee10>\n",
      "191\n",
      "<keras.layers.merge.Multiply object at 0x7f181446b710>\n",
      "192\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f181446bcc0>\n",
      "193\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f181442b940>\n",
      "194\n",
      "<efficientnet.model.get_dropout.<locals>.FixedDropout object at 0x7f181442b908>\n",
      "195\n",
      "<keras.layers.merge.Add object at 0x7f18143f20b8>\n",
      "196\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f18143afd68>\n",
      "197\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f181431c940>\n",
      "198\n",
      "<keras.layers.core.Activation object at 0x7f1814335dd8>\n",
      "199\n",
      "<keras.layers.convolutional.DepthwiseConv2D object at 0x7f181438ac50>\n",
      "200\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f18142aae80>\n",
      "201\n",
      "<keras.layers.core.Activation object at 0x7f181423ab38>\n",
      "202\n",
      "<keras.layers.pooling.GlobalAveragePooling2D object at 0x7f18141b1e48>\n",
      "203\n",
      "<keras.layers.core.Reshape object at 0x7f18141d8240>\n",
      "204\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f18141227b8>\n",
      "205\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f181413ccc0>\n",
      "206\n",
      "<keras.layers.merge.Multiply object at 0x7f18140e5080>\n",
      "207\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1814082ef0>\n",
      "208\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f18140a6b38>\n",
      "209\n",
      "<efficientnet.model.get_dropout.<locals>.FixedDropout object at 0x7f18140e5860>\n",
      "210\n",
      "<keras.layers.merge.Add object at 0x7f17ec7dbdd8>\n",
      "211\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1814060d30>\n",
      "212\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f17ec7aada0>\n",
      "213\n",
      "<keras.layers.core.Activation object at 0x7f181407b128>\n",
      "214\n",
      "<keras.layers.convolutional.DepthwiseConv2D object at 0x7f17ec6a44a8>\n",
      "215\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f17ec667208>\n",
      "216\n",
      "<keras.layers.core.Activation object at 0x7f17ec70ceb8>\n",
      "217\n",
      "<keras.layers.pooling.GlobalAveragePooling2D object at 0x7f17ec623c18>\n",
      "218\n",
      "<keras.layers.core.Reshape object at 0x7f17ec5bfc88>\n",
      "219\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f17ec5b0da0>\n",
      "220\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f17ec54ee10>\n",
      "221\n",
      "<keras.layers.merge.Multiply object at 0x7f17ec569710>\n",
      "222\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f17ec569cc0>\n",
      "223\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f17ec52c940>\n",
      "224\n",
      "<efficientnet.model.get_dropout.<locals>.FixedDropout object at 0x7f17ec52c908>\n",
      "225\n",
      "<keras.layers.merge.Add object at 0x7f17ec4f30b8>\n",
      "226\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f17ec4add68>\n",
      "227\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f17ec41b940>\n",
      "228\n",
      "<keras.layers.core.Activation object at 0x7f17ec489c50>\n",
      "229\n",
      "<keras.layers.convolutional.DepthwiseConv2D object at 0x7f17ec347860>\n",
      "230\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f17ec3a9e80>\n",
      "231\n",
      "<keras.layers.core.Activation object at 0x7f17ec2b9b00>\n",
      "232\n",
      "<keras.layers.pooling.GlobalAveragePooling2D object at 0x7f17ec2b3e48>\n",
      "233\n",
      "<keras.layers.core.Reshape object at 0x7f17ec2d9208>\n",
      "234\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f17ec2237b8>\n",
      "235\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f17ec1becc0>\n",
      "236\n",
      "<keras.layers.merge.Multiply object at 0x7f17ec1e8080>\n",
      "237\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f17ec183ef0>\n",
      "238\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f17ec1a9b38>\n",
      "239\n",
      "<efficientnet.model.get_dropout.<locals>.FixedDropout object at 0x7f17ec1e8860>\n",
      "240\n",
      "<keras.layers.merge.Add object at 0x7f17ec125dd8>\n",
      "241\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f17ec162d30>\n",
      "242\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f17ec0f4dd8>\n",
      "243\n",
      "<keras.layers.core.Activation object at 0x7f17ec0fd128>\n",
      "244\n",
      "<keras.layers.convolutional.DepthwiseConv2D object at 0x7f17d07aa4a8>\n",
      "245\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f17d076d240>\n",
      "246\n",
      "<keras.layers.core.Activation object at 0x7f17ec053ef0>\n",
      "247\n",
      "<keras.layers.pooling.GlobalAveragePooling2D object at 0x7f17d07071d0>\n",
      "248\n",
      "<keras.layers.core.Reshape object at 0x7f17d06c8b70>\n",
      "249\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f17d0639f28>\n",
      "250\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f17d0657dd8>\n",
      "251\n",
      "<keras.layers.merge.Multiply object at 0x7f17d06746d8>\n",
      "252\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f17d0674c88>\n",
      "253\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f17d0630940>\n",
      "254\n",
      "<efficientnet.model.get_dropout.<locals>.FixedDropout object at 0x7f17d0630908>\n",
      "255\n",
      "<keras.layers.merge.Add object at 0x7f17d057c0b8>\n",
      "256\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f17d05b5d68>\n",
      "257\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f17d0524fd0>\n",
      "258\n",
      "<keras.layers.core.Activation object at 0x7f17d0592c50>\n",
      "259\n",
      "<keras.layers.convolutional.DepthwiseConv2D object at 0x7f17d044d860>\n",
      "260\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f17d04b1e80>\n",
      "261\n",
      "<keras.layers.core.Activation object at 0x7f17d03bdb38>\n",
      "262\n",
      "<keras.layers.pooling.GlobalAveragePooling2D object at 0x7f17d0363748>\n",
      "263\n",
      "<keras.layers.core.Reshape object at 0x7f17d03e0240>\n",
      "264\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f17d032b7b8>\n",
      "265\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f17d02c4cc0>\n",
      "266\n",
      "<keras.layers.merge.Multiply object at 0x7f17d02ef080>\n",
      "267\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f17d0289ef0>\n",
      "268\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f17d02abb38>\n",
      "269\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f17d02ef860>\n",
      "270\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f17d0267fd0>\n",
      "271\n",
      "<keras.layers.core.Activation object at 0x7f17d0245780>\n",
      "272\n",
      "<keras.layers.convolutional.DepthwiseConv2D object at 0x7f17d01975c0>\n",
      "273\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f17d00e8080>\n",
      "274\n",
      "<keras.layers.core.Activation object at 0x7f17d0197550>\n",
      "275\n",
      "<keras.layers.pooling.GlobalAveragePooling2D object at 0x7f17d0062da0>\n",
      "276\n",
      "<keras.layers.core.Reshape object at 0x7f17d007f550>\n",
      "277\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f17d007f1d0>\n",
      "278\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f17c452de10>\n",
      "279\n",
      "<keras.layers.merge.Multiply object at 0x7f17c4552550>\n",
      "280\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f17c4552eb8>\n",
      "281\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f17c450ba90>\n",
      "282\n",
      "<efficientnet.model.get_dropout.<locals>.FixedDropout object at 0x7f17c450b710>\n",
      "283\n",
      "<keras.layers.merge.Add object at 0x7f17c448aef0>\n",
      "284\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f17c43db9e8>\n",
      "285\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f17c43faeb8>\n",
      "286\n",
      "<keras.layers.core.Activation object at 0x7f17c4415c50>\n",
      "287\n",
      "<keras.layers.convolutional.DepthwiseConv2D object at 0x7f17c4325ef0>\n",
      "288\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f17c4383cf8>\n",
      "289\n",
      "<keras.layers.core.Activation object at 0x7f17c43169b0>\n",
      "290\n",
      "<keras.layers.pooling.GlobalAveragePooling2D object at 0x7f17c4294d30>\n",
      "291\n",
      "<keras.layers.core.Reshape object at 0x7f17c42b20b8>\n",
      "292\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f17c42026a0>\n",
      "293\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f17c419bc50>\n",
      "294\n",
      "<keras.layers.merge.Multiply object at 0x7f17c41c3f60>\n",
      "295\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f17c41c36d8>\n",
      "296\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f17c415ad30>\n",
      "297\n",
      "<efficientnet.model.get_dropout.<locals>.FixedDropout object at 0x7f17c415afd0>\n",
      "298\n",
      "<keras.layers.merge.Add object at 0x7f17c40da828>\n",
      "299\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f17c40a8ba8>\n",
      "300\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f17c40cbcc0>\n",
      "301\n",
      "<keras.layers.core.Activation object at 0x7f17c406bb38>\n",
      "302\n",
      "<keras.layers.convolutional.DepthwiseConv2D object at 0x7f17c413bba8>\n",
      "303\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f1790582080>\n",
      "304\n",
      "<keras.layers.core.Activation object at 0x7f17c402cb38>\n",
      "305\n",
      "<keras.layers.pooling.GlobalAveragePooling2D object at 0x7f179053ee10>\n",
      "306\n",
      "<keras.layers.core.Reshape object at 0x7f179051a550>\n",
      "307\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f17904cdda0>\n",
      "308\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f179046bc88>\n",
      "309\n",
      "<keras.layers.merge.Multiply object at 0x7f1790411588>\n",
      "310\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1790411ef0>\n",
      "311\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f179044ba90>\n",
      "312\n",
      "<efficientnet.model.get_dropout.<locals>.FixedDropout object at 0x7f179044b710>\n",
      "313\n",
      "<keras.layers.merge.Add object at 0x7f17903c7ef0>\n",
      "314\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f17903199e8>\n",
      "315\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f179033beb8>\n",
      "316\n",
      "<keras.layers.core.Activation object at 0x7f17902d7c50>\n",
      "317\n",
      "<keras.layers.convolutional.DepthwiseConv2D object at 0x7f1790266ef0>\n",
      "318\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f17902c7cf8>\n",
      "319\n",
      "<keras.layers.core.Activation object at 0x7f17901d49b0>\n",
      "320\n",
      "<keras.layers.pooling.GlobalAveragePooling2D object at 0x7f1790153d30>\n",
      "321\n",
      "<keras.layers.core.Reshape object at 0x7f17901f00b8>\n",
      "322\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1790140668>\n",
      "323\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f17900dbc18>\n",
      "324\n",
      "<keras.layers.merge.Multiply object at 0x7f1790101fd0>\n",
      "325\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f17901016a0>\n",
      "326\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f1790099cf8>\n",
      "327\n",
      "<efficientnet.model.get_dropout.<locals>.FixedDropout object at 0x7f17901012e8>\n",
      "328\n",
      "<keras.layers.merge.Add object at 0x7f179001a828>\n",
      "329\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f17787d0ba8>\n",
      "330\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f17787f4cc0>\n",
      "331\n",
      "<keras.layers.core.Activation object at 0x7f179007dc18>\n",
      "332\n",
      "<keras.layers.convolutional.DepthwiseConv2D object at 0x7f17786ea320>\n",
      "333\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f17786af0b8>\n",
      "334\n",
      "<keras.layers.core.Activation object at 0x7f1778754d68>\n",
      "335\n",
      "<keras.layers.pooling.GlobalAveragePooling2D object at 0x7f1778721e10>\n",
      "336\n",
      "<keras.layers.core.Reshape object at 0x7f177864a048>\n",
      "337\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1778579da0>\n",
      "338\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1778598c88>\n",
      "339\n",
      "<keras.layers.merge.Multiply object at 0x7f177853f588>\n",
      "340\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f177853fef0>\n",
      "341\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f17784f9a90>\n",
      "342\n",
      "<efficientnet.model.get_dropout.<locals>.FixedDropout object at 0x7f17784f9710>\n",
      "343\n",
      "<keras.layers.merge.Add object at 0x7f17784f5f60>\n",
      "344\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f17784479b0>\n",
      "345\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f1778465e48>\n",
      "346\n",
      "<keras.layers.core.Activation object at 0x7f17784d3fd0>\n",
      "347\n",
      "<keras.layers.convolutional.DepthwiseConv2D object at 0x7f1778393ef0>\n",
      "348\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f1778304978>\n",
      "349\n",
      "<keras.layers.core.Activation object at 0x7f177835eb00>\n",
      "350\n",
      "<keras.layers.pooling.GlobalAveragePooling2D object at 0x7f17782a5588>\n",
      "351\n",
      "<keras.layers.core.Reshape object at 0x7f177831f080>\n",
      "352\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1778271668>\n",
      "353\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f177820cc18>\n",
      "354\n",
      "<keras.layers.merge.Multiply object at 0x7f1778231fd0>\n",
      "355\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f17782316a0>\n",
      "356\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f17781cdd30>\n",
      "357\n",
      "<efficientnet.model.get_dropout.<locals>.FixedDropout object at 0x7f17781cdf28>\n",
      "358\n",
      "<keras.layers.merge.Add object at 0x7f1778146828>\n",
      "359\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1778118ba8>\n",
      "360\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f1778138cc0>\n",
      "361\n",
      "<keras.layers.core.Activation object at 0x7f17781acc18>\n",
      "362\n",
      "<keras.layers.convolutional.DepthwiseConv2D object at 0x7f1756fcce48>\n",
      "363\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f1756f40048>\n",
      "364\n",
      "<keras.layers.core.Activation object at 0x7f1778097d68>\n",
      "365\n",
      "<keras.layers.pooling.GlobalAveragePooling2D object at 0x7f1778068e80>\n",
      "366\n",
      "<keras.layers.core.Reshape object at 0x7f1756f14a90>\n",
      "367\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1756e84c18>\n",
      "368\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1756ea4a58>\n",
      "369\n",
      "<keras.layers.merge.Multiply object at 0x7f1756e4c4e0>\n",
      "370\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1756e4ce48>\n",
      "371\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f1756e059e8>\n",
      "372\n",
      "<efficientnet.model.get_dropout.<locals>.FixedDropout object at 0x7f1756e05748>\n",
      "373\n",
      "<keras.layers.merge.Add object at 0x7f1756e26f98>\n",
      "374\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1756e264a8>\n",
      "375\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f1756d74f60>\n",
      "376\n",
      "<keras.layers.core.Activation object at 0x7f1756de5a58>\n",
      "377\n",
      "<keras.layers.convolutional.DepthwiseConv2D object at 0x7f1756ca0f28>\n",
      "378\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f1756c11940>\n",
      "379\n",
      "<keras.layers.core.Activation object at 0x7f1756c726a0>\n",
      "380\n",
      "<keras.layers.pooling.GlobalAveragePooling2D object at 0x7f1756b8cf98>\n",
      "381\n",
      "<keras.layers.core.Reshape object at 0x7f1756c2d048>\n",
      "382\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1756b7d630>\n",
      "383\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1756b16be0>\n",
      "384\n",
      "<keras.layers.merge.Multiply object at 0x7f1756b3df98>\n",
      "385\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1756b3d668>\n",
      "386\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f1756ad7cc0>\n",
      "387\n",
      "<efficientnet.model.get_dropout.<locals>.FixedDropout object at 0x7f1756ad7f60>\n",
      "388\n",
      "<keras.layers.merge.Add object at 0x7f1756a53fd0>\n",
      "389\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1756ab5b38>\n",
      "390\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f17569c7c50>\n",
      "391\n",
      "<keras.layers.core.Activation object at 0x7f1756a24be0>\n",
      "392\n",
      "<keras.layers.convolutional.DepthwiseConv2D object at 0x7f1756912e48>\n",
      "393\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f1756884048>\n",
      "394\n",
      "<keras.layers.core.Activation object at 0x7f17569a5cc0>\n",
      "395\n",
      "<keras.layers.pooling.GlobalAveragePooling2D object at 0x7f1756971da0>\n",
      "396\n",
      "<keras.layers.core.Reshape object at 0x7f175689b160>\n",
      "397\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f17567ced68>\n",
      "398\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f17567edb70>\n",
      "399\n",
      "<keras.layers.merge.Multiply object at 0x7f17567974e0>\n",
      "400\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1756797e48>\n",
      "401\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f175674cf60>\n",
      "402\n",
      "<efficientnet.model.get_dropout.<locals>.FixedDropout object at 0x7f175674c710>\n",
      "403\n",
      "<keras.layers.merge.Add object at 0x7f17566f6400>\n",
      "404\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f175676bd68>\n",
      "405\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f17566b9fd0>\n",
      "406\n",
      "<keras.layers.core.Activation object at 0x7f1756655be0>\n",
      "407\n",
      "<keras.layers.convolutional.DepthwiseConv2D object at 0x7f17565e6f28>\n",
      "408\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f1756556940>\n",
      "409\n",
      "<keras.layers.core.Activation object at 0x7f17565b56a0>\n",
      "410\n",
      "<keras.layers.pooling.GlobalAveragePooling2D object at 0x7f17564d0f98>\n",
      "411\n",
      "<keras.layers.core.Reshape object at 0x7f1756572048>\n",
      "412\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1756440630>\n",
      "413\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1756459be0>\n",
      "414\n",
      "<keras.layers.merge.Multiply object at 0x7f175641cd30>\n",
      "415\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f175641cda0>\n",
      "416\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f1756402438>\n",
      "417\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1756402358>\n",
      "418\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f17563fc9b0>\n",
      "419\n",
      "<keras.layers.core.Activation object at 0x7f1756346eb8>\n",
      "420\n",
      "<keras.layers.convolutional.DepthwiseConv2D object at 0x7f175632a780>\n",
      "421\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f17562b2cf8>\n",
      "422\n",
      "<keras.layers.core.Activation object at 0x7f175625af28>\n",
      "423\n",
      "<keras.layers.pooling.GlobalAveragePooling2D object at 0x7f1756211390>\n",
      "424\n",
      "<keras.layers.core.Reshape object at 0x7f1756278240>\n",
      "425\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1756145c18>\n",
      "426\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1756160908>\n",
      "427\n",
      "<keras.layers.merge.Multiply object at 0x7f1756106908>\n",
      "428\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1756106eb8>\n",
      "429\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f17560c0eb8>\n",
      "430\n",
      "<efficientnet.model.get_dropout.<locals>.FixedDropout object at 0x7f17560c0438>\n",
      "431\n",
      "<keras.layers.merge.Add object at 0x7f17560bce10>\n",
      "432\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f175600f160>\n",
      "433\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f1756032e10>\n",
      "434\n",
      "<keras.layers.core.Activation object at 0x7f175609d710>\n",
      "435\n",
      "<keras.layers.convolutional.DepthwiseConv2D object at 0x7f1755ff7080>\n",
      "436\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f1755ecb780>\n",
      "437\n",
      "<keras.layers.core.Activation object at 0x7f1755f2d7b8>\n",
      "438\n",
      "<keras.layers.pooling.GlobalAveragePooling2D object at 0x7f1755e733c8>\n",
      "439\n",
      "<keras.layers.core.Reshape object at 0x7f1755ee7048>\n",
      "440\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1755e374a8>\n",
      "441\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1755dd1a58>\n",
      "442\n",
      "<keras.layers.merge.Multiply object at 0x7f1755dfab00>\n",
      "443\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1755dfa2b0>\n",
      "444\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f1755d93b38>\n",
      "445\n",
      "<efficientnet.model.get_dropout.<locals>.FixedDropout object at 0x7f1755d93dd8>\n",
      "446\n",
      "<keras.layers.merge.Add object at 0x7f1755db5a58>\n",
      "447\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1755d13ef0>\n",
      "448\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f1755c81ac8>\n",
      "449\n",
      "<keras.layers.core.Activation object at 0x7f1755d727b8>\n",
      "450\n",
      "<keras.layers.convolutional.DepthwiseConv2D object at 0x7f1755c5f9e8>\n",
      "451\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f1755bf7cf8>\n",
      "452\n",
      "<keras.layers.core.Activation object at 0x7f1755ba2f28>\n",
      "453\n",
      "<keras.layers.pooling.GlobalAveragePooling2D object at 0x7f1755b58390>\n",
      "454\n",
      "<keras.layers.core.Reshape object at 0x7f1755b3f240>\n",
      "455\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1755a8bc18>\n",
      "456\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1755aa8908>\n",
      "457\n",
      "<keras.layers.merge.Multiply object at 0x7f1755a4f908>\n",
      "458\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1755a4feb8>\n",
      "459\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f1755a07eb8>\n",
      "460\n",
      "<efficientnet.model.get_dropout.<locals>.FixedDropout object at 0x7f1755a07438>\n",
      "461\n",
      "<keras.layers.merge.Add object at 0x7f1755988e10>\n",
      "462\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1755958128>\n",
      "463\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f175597bdd8>\n",
      "464\n",
      "<keras.layers.core.Activation object at 0x7f17559b2e48>\n",
      "465\n",
      "<keras.layers.convolutional.DepthwiseConv2D object at 0x7f17558be320>\n",
      "466\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f17558137b8>\n",
      "467\n",
      "<keras.layers.core.Activation object at 0x7f17558737f0>\n",
      "468\n",
      "<keras.layers.pooling.GlobalAveragePooling2D object at 0x7f17557bb3c8>\n",
      "469\n",
      "<keras.layers.core.Reshape object at 0x7f175582d320>\n",
      "470\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f175577d4a8>\n",
      "471\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1755719a58>\n",
      "472\n",
      "<keras.layers.merge.Multiply object at 0x7f17556bf278>\n",
      "473\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f17556bf2b0>\n",
      "474\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f17556d9b38>\n",
      "475\n",
      "<efficientnet.model.get_dropout.<locals>.FixedDropout object at 0x7f17556d9dd8>\n",
      "476\n",
      "<keras.layers.merge.Add object at 0x7f175567fa58>\n",
      "477\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1755656ef0>\n",
      "478\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f17555c9ac8>\n",
      "479\n",
      "<keras.layers.core.Activation object at 0x7f17556b47b8>\n",
      "480\n",
      "<keras.layers.convolutional.DepthwiseConv2D object at 0x7f17555aa9e8>\n",
      "481\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f17554c1cf8>\n",
      "482\n",
      "<keras.layers.core.Activation object at 0x7f17554e6f28>\n",
      "483\n",
      "<keras.layers.pooling.GlobalAveragePooling2D object at 0x7f175549e390>\n",
      "484\n",
      "<keras.layers.core.Reshape object at 0x7f175548a240>\n",
      "485\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f17553d2ba8>\n",
      "486\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f17553ed898>\n",
      "487\n",
      "<keras.layers.merge.Multiply object at 0x7f1755395898>\n",
      "488\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1755395f98>\n",
      "489\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f175534ff60>\n",
      "490\n",
      "<efficientnet.model.get_dropout.<locals>.FixedDropout object at 0x7f175534f3c8>\n",
      "491\n",
      "<keras.layers.merge.Add object at 0x7f17552d1da0>\n",
      "492\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f17552a06d8>\n",
      "493\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f1755257e48>\n",
      "494\n",
      "<keras.layers.core.Activation object at 0x7f175532c390>\n",
      "495\n",
      "<keras.layers.convolutional.DepthwiseConv2D object at 0x7f17551e9f28>\n",
      "496\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f1755158748>\n",
      "497\n",
      "<keras.layers.core.Activation object at 0x7f17551b7780>\n",
      "498\n",
      "<keras.layers.pooling.GlobalAveragePooling2D object at 0x7f175507e358>\n",
      "499\n",
      "<keras.layers.core.Reshape object at 0x7f1755174048>\n",
      "500\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1755043438>\n",
      "501\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f175505e9e8>\n",
      "502\n",
      "<keras.layers.merge.Multiply object at 0x7f1755009c88>\n",
      "503\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1755009240>\n",
      "504\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f1755021f98>\n",
      "505\n",
      "<efficientnet.model.get_dropout.<locals>.FixedDropout object at 0x7f1755021d68>\n",
      "506\n",
      "<keras.layers.merge.Add object at 0x7f1754f9ff98>\n",
      "507\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1754f705f8>\n",
      "508\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f1754f11eb8>\n",
      "509\n",
      "<keras.layers.core.Activation object at 0x7f1754f2ecc0>\n",
      "510\n",
      "<keras.layers.convolutional.DepthwiseConv2D object at 0x7f1754ffde80>\n",
      "511\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f1754eba9e8>\n",
      "512\n",
      "<keras.layers.core.Activation object at 0x7f1754e2bef0>\n",
      "513\n",
      "<keras.layers.pooling.GlobalAveragePooling2D object at 0x7f1754da6fd0>\n",
      "514\n",
      "<keras.layers.core.Reshape object at 0x7f1754dcc5f8>\n",
      "515\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1754d18b70>\n",
      "516\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1754d33860>\n",
      "517\n",
      "<keras.layers.merge.Multiply object at 0x7f1754cd9860>\n",
      "518\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1754cd9f60>\n",
      "519\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f1754c92f60>\n",
      "520\n",
      "<efficientnet.model.get_dropout.<locals>.FixedDropout object at 0x7f1754c923c8>\n",
      "521\n",
      "<keras.layers.merge.Add object at 0x7f1754c14da0>\n",
      "522\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1754be5080>\n",
      "523\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f1754b85780>\n",
      "524\n",
      "<keras.layers.core.Activation object at 0x7f1754c6eeb8>\n",
      "525\n",
      "<keras.layers.convolutional.DepthwiseConv2D object at 0x7f1754b4a160>\n",
      "526\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f1754aa0748>\n",
      "527\n",
      "<keras.layers.core.Activation object at 0x7f1754a80780>\n",
      "528\n",
      "<keras.layers.pooling.GlobalAveragePooling2D object at 0x7f17549c7358>\n",
      "529\n",
      "<keras.layers.core.Reshape object at 0x7f1754ab9048>\n",
      "530\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1754ab9550>\n",
      "531\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f17549a7898>\n",
      "532\n",
      "<keras.layers.merge.Multiply object at 0x7f175494fc88>\n",
      "533\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f175494f208>\n",
      "534\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f175496bf98>\n",
      "535\n",
      "<efficientnet.model.get_dropout.<locals>.FixedDropout object at 0x7f175496bd68>\n",
      "536\n",
      "<keras.layers.merge.Add object at 0x7f17548d7f98>\n",
      "537\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f17548c19b0>\n",
      "538\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f1754855fd0>\n",
      "539\n",
      "<keras.layers.core.Activation object at 0x7f17548b3eb8>\n",
      "540\n",
      "<keras.layers.convolutional.DepthwiseConv2D object at 0x7f17547a6f60>\n",
      "541\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f1754781d68>\n",
      "542\n",
      "<keras.layers.core.Activation object at 0x7f1754771f28>\n",
      "543\n",
      "<keras.layers.pooling.GlobalAveragePooling2D object at 0x7f17546eeef0>\n",
      "544\n",
      "<keras.layers.core.Reshape object at 0x7f1754714630>\n",
      "545\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1754660ba8>\n",
      "546\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f175467a898>\n",
      "547\n",
      "<keras.layers.merge.Multiply object at 0x7f175461f898>\n",
      "548\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f175461ff98>\n",
      "549\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f17545dbf60>\n",
      "550\n",
      "<efficientnet.model.get_dropout.<locals>.FixedDropout object at 0x7f17545db3c8>\n",
      "551\n",
      "<keras.layers.merge.Add object at 0x7f175455dda0>\n",
      "552\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f175452b080>\n",
      "553\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f17544cd780>\n",
      "554\n",
      "<keras.layers.core.Activation object at 0x7f17545b8eb8>\n",
      "555\n",
      "<keras.layers.convolutional.DepthwiseConv2D object at 0x7f1754490160>\n",
      "556\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f17543e7710>\n",
      "557\n",
      "<keras.layers.core.Activation object at 0x7f17543c3780>\n",
      "558\n",
      "<keras.layers.pooling.GlobalAveragePooling2D object at 0x7f175430b358>\n",
      "559\n",
      "<keras.layers.core.Reshape object at 0x7f175437e048>\n",
      "560\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f17542cf438>\n",
      "561\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f17542eb9e8>\n",
      "562\n",
      "<keras.layers.merge.Multiply object at 0x7f1754294c88>\n",
      "563\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f1754294240>\n",
      "564\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f17542adf98>\n",
      "565\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f17542add68>\n",
      "566\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f175420b7b8>\n",
      "567\n",
      "<keras.layers.core.Activation object at 0x7f17541faeb8>\n",
      "568\n",
      "<keras.layers.convolutional.DepthwiseConv2D object at 0x7f175411af98>\n",
      "569\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f1754175860>\n",
      "570\n",
      "<keras.layers.core.Activation object at 0x7f17540eada0>\n",
      "571\n",
      "<keras.layers.pooling.GlobalAveragePooling2D object at 0x7f1754065fd0>\n",
      "572\n",
      "<keras.layers.core.Reshape object at 0x7f17540894a8>\n",
      "573\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174f795a20>\n",
      "574\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174f7b0710>\n",
      "575\n",
      "<keras.layers.merge.Multiply object at 0x7f174f75e748>\n",
      "576\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174f7757b8>\n",
      "577\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f174f71eda0>\n",
      "578\n",
      "<efficientnet.model.get_dropout.<locals>.FixedDropout object at 0x7f174f71e3c8>\n",
      "579\n",
      "<keras.layers.merge.Add object at 0x7f174f71ee80>\n",
      "580\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174f694fd0>\n",
      "581\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f174f660f60>\n",
      "582\n",
      "<keras.layers.core.Activation object at 0x7f174f6eed30>\n",
      "583\n",
      "<keras.layers.convolutional.DepthwiseConv2D object at 0x7f174f5ea198>\n",
      "584\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f174f51c588>\n",
      "585\n",
      "<keras.layers.core.Activation object at 0x7f174f57ab70>\n",
      "586\n",
      "<keras.layers.pooling.GlobalAveragePooling2D object at 0x7f174f498d68>\n",
      "587\n",
      "<keras.layers.core.Reshape object at 0x7f174f5370f0>\n",
      "588\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174f537390>\n",
      "589\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174f40a2b0>\n",
      "590\n",
      "<keras.layers.merge.Multiply object at 0x7f174f3cc908>\n",
      "591\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174f3ccb00>\n",
      "592\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f174f388898>\n",
      "593\n",
      "<efficientnet.model.get_dropout.<locals>.FixedDropout object at 0x7f174f3cce10>\n",
      "594\n",
      "<keras.layers.merge.Add object at 0x7f174f362f60>\n",
      "595\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174f332470>\n",
      "596\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f174f2d4fd0>\n",
      "597\n",
      "<keras.layers.core.Activation object at 0x7f174f3415c0>\n",
      "598\n",
      "<keras.layers.convolutional.DepthwiseConv2D object at 0x7f174f220f98>\n",
      "599\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f174f27e860>\n",
      "600\n",
      "<keras.layers.core.Activation object at 0x7f174f1f2da0>\n",
      "601\n",
      "<keras.layers.pooling.GlobalAveragePooling2D object at 0x7f174f167fd0>\n",
      "602\n",
      "<keras.layers.core.Reshape object at 0x7f174f18e4a8>\n",
      "603\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174f0dc9e8>\n",
      "604\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174f0f66d8>\n",
      "605\n",
      "<keras.layers.merge.Multiply object at 0x7f174f0a36d8>\n",
      "606\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174f0a3d30>\n",
      "607\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f174f063cc0>\n",
      "608\n",
      "<efficientnet.model.get_dropout.<locals>.FixedDropout object at 0x7f174f0a3780>\n",
      "609\n",
      "<keras.layers.merge.Add object at 0x7f174f019e80>\n",
      "610\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174ef87ef0>\n",
      "611\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f174efa8f60>\n",
      "612\n",
      "<keras.layers.core.Activation object at 0x7f174f0356d8>\n",
      "613\n",
      "<keras.layers.convolutional.DepthwiseConv2D object at 0x7f174eef3e80>\n",
      "614\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f174ee67630>\n",
      "615\n",
      "<keras.layers.core.Activation object at 0x7f174ee44cc0>\n",
      "616\n",
      "<keras.layers.pooling.GlobalAveragePooling2D object at 0x7f174ede0dd8>\n",
      "617\n",
      "<keras.layers.core.Reshape object at 0x7f174edff198>\n",
      "618\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174edff438>\n",
      "619\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174ed56f98>\n",
      "620\n",
      "<keras.layers.merge.Multiply object at 0x7f174ed138d0>\n",
      "621\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174ed13978>\n",
      "622\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f174ecd1908>\n",
      "623\n",
      "<efficientnet.model.get_dropout.<locals>.FixedDropout object at 0x7f174ed13668>\n",
      "624\n",
      "<keras.layers.merge.Add object at 0x7f174ecacf60>\n",
      "625\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174ec7c4e0>\n",
      "626\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f174ec1ce10>\n",
      "627\n",
      "<keras.layers.core.Activation object at 0x7f174ec8b630>\n",
      "628\n",
      "<keras.layers.convolutional.DepthwiseConv2D object at 0x7f174eb68e80>\n",
      "629\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f174eb48470>\n",
      "630\n",
      "<keras.layers.core.Activation object at 0x7f174eb38dd8>\n",
      "631\n",
      "<keras.layers.pooling.GlobalAveragePooling2D object at 0x7f174eab2eb8>\n",
      "632\n",
      "<keras.layers.core.Reshape object at 0x7f174ead5908>\n",
      "633\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174ea24a20>\n",
      "634\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174e9bf710>\n",
      "635\n",
      "<keras.layers.merge.Multiply object at 0x7f174e9eb710>\n",
      "636\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174e9ebdd8>\n",
      "637\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f174e9a8dd8>\n",
      "638\n",
      "<efficientnet.model.get_dropout.<locals>.FixedDropout object at 0x7f174e9eb7b8>\n",
      "639\n",
      "<keras.layers.merge.Add object at 0x7f174e916ef0>\n",
      "640\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174e9beba8>\n",
      "641\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f174e8f0fd0>\n",
      "642\n",
      "<keras.layers.core.Activation object at 0x7f174e97ccc0>\n",
      "643\n",
      "<keras.layers.convolutional.DepthwiseConv2D object at 0x7f174e8aa860>\n",
      "644\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f174e7b0550>\n",
      "645\n",
      "<keras.layers.core.Activation object at 0x7f174e787b38>\n",
      "646\n",
      "<keras.layers.pooling.GlobalAveragePooling2D object at 0x7f174e72bd30>\n",
      "647\n",
      "<keras.layers.core.Reshape object at 0x7f174e7490b8>\n",
      "648\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174e749358>\n",
      "649\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174e69deb8>\n",
      "650\n",
      "<keras.layers.merge.Multiply object at 0x7f174e6b2f60>\n",
      "651\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174e65b6a0>\n",
      "652\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f174e675ba8>\n",
      "653\n",
      "<efficientnet.model.get_dropout.<locals>.FixedDropout object at 0x7f174e675f98>\n",
      "654\n",
      "<keras.layers.merge.Add object at 0x7f174e617828>\n",
      "655\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174e5f2ef0>\n",
      "656\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f174e565f98>\n",
      "657\n",
      "<keras.layers.core.Activation object at 0x7f174e5d5588>\n",
      "658\n",
      "<keras.layers.convolutional.DepthwiseConv2D object at 0x7f174e57dba8>\n",
      "659\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f174e491ac8>\n",
      "660\n",
      "<keras.layers.core.Activation object at 0x7f174e404d68>\n",
      "661\n",
      "<keras.layers.pooling.GlobalAveragePooling2D object at 0x7f174e3fcf98>\n",
      "662\n",
      "<keras.layers.core.Reshape object at 0x7f174e421470>\n",
      "663\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174e36b9e8>\n",
      "664\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174e3066d8>\n",
      "665\n",
      "<keras.layers.merge.Multiply object at 0x7f174e333710>\n",
      "666\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174e333748>\n",
      "667\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f174e2f0d68>\n",
      "668\n",
      "<efficientnet.model.get_dropout.<locals>.FixedDropout object at 0x7f174e333a90>\n",
      "669\n",
      "<keras.layers.merge.Add object at 0x7f174e26bf28>\n",
      "670\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174e287b38>\n",
      "671\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f174e236e80>\n",
      "672\n",
      "<keras.layers.core.Activation object at 0x7f174e245cf8>\n",
      "673\n",
      "<keras.layers.convolutional.DepthwiseConv2D object at 0x7f174e141278>\n",
      "674\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f174e160e48>\n",
      "675\n",
      "<keras.layers.core.Activation object at 0x7f174e0d0c88>\n",
      "676\n",
      "<keras.layers.pooling.GlobalAveragePooling2D object at 0x7f174e073d30>\n",
      "677\n",
      "<keras.layers.core.Reshape object at 0x7f174e08d0f0>\n",
      "678\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174e08d390>\n",
      "679\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174dfe1eb8>\n",
      "680\n",
      "<keras.layers.merge.Multiply object at 0x7f174e04f208>\n",
      "681\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174dfa18d0>\n",
      "682\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f174df5d860>\n",
      "683\n",
      "<efficientnet.model.get_dropout.<locals>.FixedDropout object at 0x7f174dfa1dd8>\n",
      "684\n",
      "<keras.layers.merge.Add object at 0x7f174df3aef0>\n",
      "685\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174de89438>\n",
      "686\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f174df1ae48>\n",
      "687\n",
      "<keras.layers.core.Activation object at 0x7f174deff1d0>\n",
      "688\n",
      "<keras.layers.convolutional.DepthwiseConv2D object at 0x7f174de38c88>\n",
      "689\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f174ddd45f8>\n",
      "690\n",
      "<keras.layers.core.Activation object at 0x7f174dd48d30>\n",
      "691\n",
      "<keras.layers.pooling.GlobalAveragePooling2D object at 0x7f174dcc1f98>\n",
      "692\n",
      "<keras.layers.core.Reshape object at 0x7f174dd66438>\n",
      "693\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174dcb29b0>\n",
      "694\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174dc4d6a0>\n",
      "695\n",
      "<keras.layers.merge.Multiply object at 0x7f174dc796d8>\n",
      "696\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174dc79710>\n",
      "697\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f174dc38c50>\n",
      "698\n",
      "<efficientnet.model.get_dropout.<locals>.FixedDropout object at 0x7f174dc79a58>\n",
      "699\n",
      "<keras.layers.merge.Add object at 0x7f174dbaeef0>\n",
      "700\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174dbccb00>\n",
      "701\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f174db7ef60>\n",
      "702\n",
      "<keras.layers.core.Activation object at 0x7f174db8ecc0>\n",
      "703\n",
      "<keras.layers.convolutional.DepthwiseConv2D object at 0x7f174da4de10>\n",
      "704\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f174da3b550>\n",
      "705\n",
      "<keras.layers.core.Activation object at 0x7f174da15b38>\n",
      "706\n",
      "<keras.layers.pooling.GlobalAveragePooling2D object at 0x7f174d9b8c18>\n",
      "707\n",
      "<keras.layers.core.Reshape object at 0x7f174d9d70b8>\n",
      "708\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174d9d7358>\n",
      "709\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174d924f28>\n",
      "710\n",
      "<keras.layers.merge.Multiply object at 0x7f174d93ceb8>\n",
      "711\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174d93cef0>\n",
      "712\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f174d8a1828>\n",
      "713\n",
      "<efficientnet.model.get_dropout.<locals>.FixedDropout object at 0x7f174d8e8dd8>\n",
      "714\n",
      "<keras.layers.merge.Add object at 0x7f174d82a7f0>\n",
      "715\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174d84a208>\n",
      "716\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f174d7f4f98>\n",
      "717\n",
      "<keras.layers.core.Activation object at 0x7f174d861e48>\n",
      "718\n",
      "<keras.layers.convolutional.DepthwiseConv2D object at 0x7f174d7190f0>\n",
      "719\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f174d77df98>\n",
      "720\n",
      "<keras.layers.core.Activation object at 0x7f174d68fcf8>\n",
      "721\n",
      "<keras.layers.pooling.GlobalAveragePooling2D object at 0x7f174d605f98>\n",
      "722\n",
      "<keras.layers.core.Reshape object at 0x7f174d6ac400>\n",
      "723\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174d5f89b0>\n",
      "724\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174d5926a0>\n",
      "725\n",
      "<keras.layers.merge.Multiply object at 0x7f174d5416d8>\n",
      "726\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174d541710>\n",
      "727\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f174d57ec50>\n",
      "728\n",
      "<efficientnet.model.get_dropout.<locals>.FixedDropout object at 0x7f174d541a58>\n",
      "729\n",
      "<keras.layers.merge.Add object at 0x7f174d535e10>\n",
      "730\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174d4a2f28>\n",
      "731\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f174d444fd0>\n",
      "732\n",
      "<keras.layers.core.Activation object at 0x7f174d4d5668>\n",
      "733\n",
      "<keras.layers.convolutional.DepthwiseConv2D object at 0x7f174d393e10>\n",
      "734\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f174d302550>\n",
      "735\n",
      "<keras.layers.core.Activation object at 0x7f174d35fb38>\n",
      "736\n",
      "<keras.layers.pooling.GlobalAveragePooling2D object at 0x7f174d2fdc18>\n",
      "737\n",
      "<keras.layers.core.Reshape object at 0x7f174d31e0b8>\n",
      "738\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174d31e358>\n",
      "739\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174d26ef28>\n",
      "740\n",
      "<keras.layers.merge.Multiply object at 0x7f174d203eb8>\n",
      "741\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174d203ef0>\n",
      "742\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f174d1e9828>\n",
      "743\n",
      "<efficientnet.model.get_dropout.<locals>.FixedDropout object at 0x7f174d230dd8>\n",
      "744\n",
      "<keras.layers.merge.Add object at 0x7f174d1727f0>\n",
      "745\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174d18f198>\n",
      "746\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f174d139f60>\n",
      "747\n",
      "<keras.layers.core.Activation object at 0x7f174d1a7e48>\n",
      "748\n",
      "<keras.layers.convolutional.DepthwiseConv2D object at 0x7f174d007f28>\n",
      "749\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f174d063a58>\n",
      "750\n",
      "<keras.layers.core.Activation object at 0x7f174cfd6cf8>\n",
      "751\n",
      "<keras.layers.pooling.GlobalAveragePooling2D object at 0x7f174cf4ff98>\n",
      "752\n",
      "<keras.layers.core.Reshape object at 0x7f174cff5400>\n",
      "753\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174cec09b0>\n",
      "754\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174ceda6a0>\n",
      "755\n",
      "<keras.layers.merge.Multiply object at 0x7f174ce866d8>\n",
      "756\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174ce86710>\n",
      "757\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f174ce46c50>\n",
      "758\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174ce86a58>\n",
      "759\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f174cd8af98>\n",
      "760\n",
      "<keras.layers.core.Activation object at 0x7f174ce5a978>\n",
      "761\n",
      "<keras.layers.convolutional.DepthwiseConv2D object at 0x7f174cd09fd0>\n",
      "762\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f174ccf72b0>\n",
      "763\n",
      "<keras.layers.core.Activation object at 0x7f174cdab588>\n",
      "764\n",
      "<keras.layers.pooling.GlobalAveragePooling2D object at 0x7f174cc76e80>\n",
      "765\n",
      "<keras.layers.core.Reshape object at 0x7f174cc92780>\n",
      "766\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174cbc2dd8>\n",
      "767\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174cbe2e48>\n",
      "768\n",
      "<keras.layers.merge.Multiply object at 0x7f174cbfd748>\n",
      "769\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174cbfdcf8>\n",
      "770\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f174cbbbc88>\n",
      "771\n",
      "<efficientnet.model.get_dropout.<locals>.FixedDropout object at 0x7f174cbbbf28>\n",
      "772\n",
      "<keras.layers.merge.Add object at 0x7f174cb60fd0>\n",
      "773\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174cac0dd8>\n",
      "774\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f174caafe10>\n",
      "775\n",
      "<keras.layers.core.Activation object at 0x7f174cb1ad30>\n",
      "776\n",
      "<keras.layers.convolutional.DepthwiseConv2D object at 0x7f174c9d98d0>\n",
      "777\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f174ca3eef0>\n",
      "778\n",
      "<keras.layers.core.Activation object at 0x7f174c94dba8>\n",
      "779\n",
      "<keras.layers.pooling.GlobalAveragePooling2D object at 0x7f174c8c2fd0>\n",
      "780\n",
      "<keras.layers.core.Reshape object at 0x7f174c9692b0>\n",
      "781\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174c8b9828>\n",
      "782\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174c851e10>\n",
      "783\n",
      "<keras.layers.merge.Multiply object at 0x7f174c87b0b8>\n",
      "784\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174c8155c0>\n",
      "785\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f174c83b7b8>\n",
      "786\n",
      "<efficientnet.model.get_dropout.<locals>.FixedDropout object at 0x7f174c87b8d0>\n",
      "787\n",
      "<keras.layers.merge.Add object at 0x7f174c7b4e48>\n",
      "788\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174c7f1dd8>\n",
      "789\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f174c703e48>\n",
      "790\n",
      "<keras.layers.core.Activation object at 0x7f174c7910b8>\n",
      "791\n",
      "<keras.layers.convolutional.DepthwiseConv2D object at 0x7f174c64ff98>\n",
      "792\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f174c63d278>\n",
      "793\n",
      "<keras.layers.core.Activation object at 0x7f174c6e4f60>\n",
      "794\n",
      "<keras.layers.pooling.GlobalAveragePooling2D object at 0x7f174c5bcf28>\n",
      "795\n",
      "<keras.layers.core.Reshape object at 0x7f174c5d8748>\n",
      "796\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174c50bdd8>\n",
      "797\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174c529e48>\n",
      "798\n",
      "<keras.layers.merge.Multiply object at 0x7f174c4c4748>\n",
      "799\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f174c4c4cf8>\n",
      "800\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f174c483c88>\n",
      "801\n",
      "<efficientnet.model.get_dropout.<locals>.FixedDropout object at 0x7f174c483f28>\n",
      "802\n",
      "<keras.layers.merge.Add object at 0x7f174c4a7fd0>\n",
      "803\n",
      "<keras.layers.convolutional.Conv2D object at 0x7f18a2652b00>\n",
      "804\n",
      "<keras.layers.normalization.BatchNormalization object at 0x7f174c3f5e10>\n",
      "805\n",
      "<keras.layers.core.Activation object at 0x7f174c3ba828>\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:From /home/greg/.local/lib/python3.6/site-packages/tensorflow/python/ops/nn_impl.py:180: add_dispatch_support.<locals>.wrapper (from tensorflow.python.ops.array_ops) is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Use tf.where in 2.0, which has the same broadcast rule as np.where\n"
     ]
    }
   ],
   "source": [
    "# CNN EfficientNet (https://arxiv.org/abs/1905.11946)\n",
    "\n",
    "model = Sequential()\n",
    "efficient_net = efn.EfficientNetB7(weights='imagenet', include_top=False, input_shape=input_shape)\n",
    "#efficient_net.trainable = False\n",
    "for index, layer in enumerate(efficient_net.layers):\n",
    "    if index < 761:\n",
    "        layer.trainable = False\n",
    "\n",
    "    print(index)\n",
    "    print(layer)\n",
    "model.add(efficient_net)\n",
    "#model.add(GlobalMaxPooling2D())\n",
    "model.add(Dense(1024, activation='relu'))\n",
    "model.add(Flatten())\n",
    "# if dropout_rate > 0:\n",
    "#     model.add(layers.Dropout(dropout_rate, name=\"dropout_out\"))\n",
    "# model.add(layers.Dense(256, activation='relu', name=\"fc1\"))\n",
    "model.add(Dense(1, activation='sigmoid')) #, name=\"output\"\n",
    "model.compile(loss='binary_crossentropy',\n",
    "            optimizer=RMSprop(lr=0.0001),\n",
    "            metrics=['accuracy'])\n",
    "\n",
    "with open(MODEL_SUMMARY_FILE,\"w\") as fh:\n",
    "    model.summary(print_fn=lambda line: fh.write(line + \"\\n\"))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Data augmentation\n",
    "training_data_generator = ImageDataGenerator(\n",
    "    rescale=1./255,\n",
    "    shear_range=0.1,\n",
    "    zoom_range=0.1,\n",
    "    horizontal_flip=True)\n",
    "validation_data_generator = ImageDataGenerator(rescale=1./255)\n",
    "test_data_generator = ImageDataGenerator(rescale=1./255)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Found 20000 images belonging to 2 classes.\n",
      "Found 5000 images belonging to 2 classes.\n",
      "Found 100 images belonging to 1 classes.\n"
     ]
    }
   ],
   "source": [
    "# Data preparation\n",
    "training_generator = training_data_generator.flow_from_directory(\n",
    "    training_data_dir,\n",
    "    target_size=(IMAGE_WIDTH, IMAGE_HEIGHT),\n",
    "    batch_size=BATCH_SIZE,\n",
    "    class_mode=\"binary\")\n",
    "validation_generator = validation_data_generator.flow_from_directory(\n",
    "    validation_data_dir,\n",
    "    target_size=(IMAGE_WIDTH, IMAGE_HEIGHT),\n",
    "    batch_size=BATCH_SIZE,\n",
    "    class_mode=\"binary\")\n",
    "test_generator = test_data_generator.flow_from_directory(\n",
    "    test_data_dir,\n",
    "    target_size=(IMAGE_WIDTH, IMAGE_HEIGHT),\n",
    "    batch_size=1,\n",
    "    class_mode=\"binary\", \n",
    "    shuffle=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 864x576 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Log-loss (cost function):\n",
      "training   (min:    0.133, max:    0.258, cur:    0.133)\n",
      "validation (min:    0.009, max:    0.134, cur:    0.040)\n",
      "\n",
      "accuracy:\n",
      "training   (min:    0.891, max:    0.951, cur:    0.951)\n",
      "validation (min:    0.966, max:    0.978, cur:    0.973)\n",
      "Epoch 8/20\n",
      "624/625 [============================>.] - ETA: 0s - loss: 0.1323 - accuracy: 0.9514"
     ]
    }
   ],
   "source": [
    "# Training\n",
    "model.fit_generator(\n",
    "    training_generator,\n",
    "    steps_per_epoch=len(training_generator.filenames) // BATCH_SIZE,\n",
    "    epochs=EPOCHS,\n",
    "    validation_data=validation_generator,\n",
    "    validation_steps=len(validation_generator.filenames) // BATCH_SIZE,\n",
    "    callbacks=[PlotLossesCallback(), CSVLogger(TRAINING_LOGS_FILE,\n",
    "                                            append=False,\n",
    "                                            separator=\";\")], \n",
    "    verbose=1)\n",
    "model.save_weights(MODEL_FILE)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "# Testing\n",
    "probabilities = model.predict_generator(test_generator, TEST_SIZE)\n",
    "for index, probability in enumerate(probabilities):\n",
    "    image_path = test_data_dir + \"/\" +test_generator.filenames[index]\n",
    "    img = mpimg.imread(image_path)\n",
    "    plt.imshow(img)\n",
    "    if probability > 0.5:\n",
    "        plt.title(\"%.2f\" % (probability[0]*100) + \"% dog\")\n",
    "    else:\n",
    "        plt.title(\"%.2f\" % ((1-probability[0])*100) + \"% cat\")\n",
    "    plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.9"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}