Emotion Recognition - Neural Networks-part2.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "#dependencies\n",
    "import numpy as np\n",
    "import cv2\n",
    "import pandas as pd\n",
    "import matplotlib.pyplot as plt\n",
    "from keras.utils import np_utils\n",
    "from keras.models import Sequential, Model, model_from_json\n",
    "from keras.layers import Conv2D, Activation, MaxPool2D, Dropout, Dense, BatchNormalization, Flatten\n",
    "from keras.callbacks import ModelCheckpoint\n",
    "import tarfile"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style>\n",
       "    .dataframe thead tr:only-child th {\n",
       "        text-align: right;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: left;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>emotion</th>\n",
       "      <th>pixels</th>\n",
       "      <th>Usage</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>0</td>\n",
       "      <td>70 80 82 72 58 58 60 63 54 58 60 48 89 115 121...</td>\n",
       "      <td>Training</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>0</td>\n",
       "      <td>151 150 147 155 148 133 111 140 170 174 182 15...</td>\n",
       "      <td>Training</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>2</td>\n",
       "      <td>231 212 156 164 174 138 161 173 182 200 106 38...</td>\n",
       "      <td>Training</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>4</td>\n",
       "      <td>24 32 36 30 32 23 19 20 30 41 21 22 32 34 21 1...</td>\n",
       "      <td>Training</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>6</td>\n",
       "      <td>4 0 0 0 0 0 0 0 0 0 0 0 3 15 23 28 48 50 58 84...</td>\n",
       "      <td>Training</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   emotion                                             pixels     Usage\n",
       "0        0  70 80 82 72 58 58 60 63 54 58 60 48 89 115 121...  Training\n",
       "1        0  151 150 147 155 148 133 111 140 170 174 182 15...  Training\n",
       "2        2  231 212 156 164 174 138 161 173 182 200 106 38...  Training\n",
       "3        4  24 32 36 30 32 23 19 20 30 41 21 22 32 34 21 1...  Training\n",
       "4        6  4 0 0 0 0 0 0 0 0 0 0 0 3 15 23 28 48 50 58 84...  Training"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#dataset\n",
    "data_comp = tarfile.open(\"fer2013.tar\") \n",
    "ds = pd.read_csv(data_comp.extractfile(\"fer2013/fer2013.csv\"))\n",
    "ds.head()\n",
    "#0=Angry, 1=Disgust, 2=Fear, 3=Happy, 4=Sad, 5=Surprise, 6=Neutral"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Training       28709\n",
       "PrivateTest     3589\n",
       "PublicTest      3589\n",
       "Name: Usage, dtype: int64"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "ds[\"Usage\"].value_counts()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "((28709, 2304), (3589, 2304))"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train = ds[[\"emotion\", \"pixels\"]][ds[\"Usage\"] == \"Training\"]\n",
    "train['pixels'] = train['pixels'].apply(lambda x: np.fromstring(x, sep=' '))\n",
    "train_pix = np.vstack(train['pixels'].values)\n",
    "test = ds[[\"emotion\", \"pixels\"]][ds[\"Usage\"] == \"PublicTest\"]\n",
    "test['pixels'] = test['pixels'].apply(lambda x: np.fromstring(x, sep=' '))\n",
    "test_pix = np.vstack(test['pixels'].values)\n",
    "train_pix.shape, test_pix.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "((28709,), (3589,))"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#0=Angry, 1=Disgust, 2=Fear, 3=Happy, 4=Sad, 5=Surprise, 6=Neutral\n",
    "#0=angry, 1=disgust+fear+surprise, 3=Happy, 4=sad, 6= neutral\n",
    "#apending 2 and 5 to 1; 5 to 2; 6 to 4\n",
    "#:::: set append = false to avoid changes\n",
    "def e_ind(x):\n",
    "    if(x==2 or x==5):\n",
    "        return 1\n",
    "    elif(x==5):\n",
    "        return 2\n",
    "    elif(x==6):\n",
    "        return 4\n",
    "    else:\n",
    "        return x\n",
    "F_S_D = True\n",
    "if(F_S_D):\n",
    "    train['emotion'] = train['emotion'].apply(lambda x: e_ind(x))\n",
    "    test['emotion'] = test['emotion'].apply(lambda x: e_ind(x))\n",
    "train_ind = np.array(train[\"emotion\"])\n",
    "test_ind = np.array(test[\"emotion\"])\n",
    "train_ind.shape, test_ind.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "((28709, 5), (3589, 5))"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_pix = train_pix.reshape(-1,48,48,1)\n",
    "train_ind = np_utils.to_categorical(train_ind)\n",
    "test_pix = test_pix.reshape(-1,48,48,1)\n",
    "test_ind = np_utils.to_categorical(test_ind)\n",
    "train_ind.shape, test_ind.shape\n",
    "#train_pix.shape, test_pix.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "conv2d_29 (Conv2D)           (None, 44, 44, 64)        1664      \n",
      "_________________________________________________________________\n",
      "batch_normalization_22 (Batc (None, 44, 44, 64)        256       \n",
      "_________________________________________________________________\n",
      "activation_43 (Activation)   (None, 44, 44, 64)        0         \n",
      "_________________________________________________________________\n",
      "conv2d_30 (Conv2D)           (None, 41, 41, 64)        65600     \n",
      "_________________________________________________________________\n",
      "batch_normalization_23 (Batc (None, 41, 41, 64)        256       \n",
      "_________________________________________________________________\n",
      "activation_44 (Activation)   (None, 41, 41, 64)        0         \n",
      "_________________________________________________________________\n",
      "max_pooling2d_17 (MaxPooling (None, 20, 20, 64)        0         \n",
      "_________________________________________________________________\n",
      "dropout_13 (Dropout)         (None, 20, 20, 64)        0         \n",
      "_________________________________________________________________\n",
      "conv2d_31 (Conv2D)           (None, 18, 18, 32)        18464     \n",
      "_________________________________________________________________\n",
      "batch_normalization_24 (Batc (None, 18, 18, 32)        128       \n",
      "_________________________________________________________________\n",
      "activation_45 (Activation)   (None, 18, 18, 32)        0         \n",
      "_________________________________________________________________\n",
      "conv2d_32 (Conv2D)           (None, 16, 16, 32)        9248      \n",
      "_________________________________________________________________\n",
      "batch_normalization_25 (Batc (None, 16, 16, 32)        128       \n",
      "_________________________________________________________________\n",
      "activation_46 (Activation)   (None, 16, 16, 32)        0         \n",
      "_________________________________________________________________\n",
      "max_pooling2d_18 (MaxPooling (None, 8, 8, 32)          0         \n",
      "_________________________________________________________________\n",
      "dropout_14 (Dropout)         (None, 8, 8, 32)          0         \n",
      "_________________________________________________________________\n",
      "flatten_8 (Flatten)          (None, 2048)              0         \n",
      "_________________________________________________________________\n",
      "dense_15 (Dense)             (None, 128)               262272    \n",
      "_________________________________________________________________\n",
      "batch_normalization_26 (Batc (None, 128)               512       \n",
      "_________________________________________________________________\n",
      "activation_47 (Activation)   (None, 128)               0         \n",
      "_________________________________________________________________\n",
      "dropout_15 (Dropout)         (None, 128)               0         \n",
      "_________________________________________________________________\n",
      "dense_16 (Dense)             (None, 5)                 645       \n",
      "_________________________________________________________________\n",
      "activation_48 (Activation)   (None, 5)                 0         \n",
      "=================================================================\n",
      "Total params: 359,173\n",
      "Trainable params: 358,533\n",
      "Non-trainable params: 640\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model = Sequential()\n",
    "model.add(Conv2D(64, 5, data_format=\"channels_last\", kernel_initializer=\"he_normal\", \n",
    "                 input_shape=(48, 48, 1)))\n",
    "model.add(BatchNormalization())\n",
    "model.add(Activation(\"relu\"))\n",
    "\n",
    "model.add(Conv2D(64, 4))\n",
    "model.add(BatchNormalization())\n",
    "model.add(Activation(\"relu\"))\n",
    "model.add(MaxPool2D(pool_size=(2, 2), strides=2))\n",
    "model.add(Dropout(0.5))\n",
    "\n",
    "model.add(Conv2D(32, 3))\n",
    "model.add(BatchNormalization())\n",
    "model.add(Activation(\"relu\"))\n",
    "\n",
    "model.add(Conv2D(32, 3))\n",
    "model.add(BatchNormalization())\n",
    "model.add(Activation(\"relu\"))\n",
    "\n",
    "model.add(MaxPool2D(pool_size=(2, 2), strides=2))\n",
    "model.add(Dropout(0.5))\n",
    "\n",
    "model.add(Flatten())\n",
    "model.add(Dense(128))\n",
    "model.add(BatchNormalization())\n",
    "model.add(Activation(\"relu\"))\n",
    "model.add(Dropout(0.2))\n",
    "\n",
    "if(F_S_D):\n",
    "    model.add(Dense(5))\n",
    "else:\n",
    "    model.add(Dense(7))\n",
    "model.add(Activation('softmax'))\n",
    "model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])\n",
    "model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 28709 samples, validate on 3589 samples\n",
      "Epoch 1/16\n",
      "\n",
      "Epoch 00001: val_loss improved from inf to 1.18156, saving model to chkPt1.h5\n",
      " - 1189s - loss: 1.3142 - acc: 0.4424 - val_loss: 1.1816 - val_acc: 0.4684\n",
      "Epoch 2/16\n",
      "\n",
      "Epoch 00002: val_loss improved from 1.18156 to 1.02509, saving model to chkPt1.h5\n",
      " - 1089s - loss: 1.1040 - acc: 0.5325 - val_loss: 1.0251 - val_acc: 0.5837\n",
      "Epoch 3/16\n",
      "\n",
      "Epoch 00003: val_loss improved from 1.02509 to 0.95903, saving model to chkPt1.h5\n",
      " - 1154s - loss: 1.0263 - acc: 0.5712 - val_loss: 0.9590 - val_acc: 0.6144\n",
      "Epoch 4/16\n",
      "\n",
      "Epoch 00004: val_loss did not improve\n",
      " - 1119s - loss: 0.9729 - acc: 0.5935 - val_loss: 0.9751 - val_acc: 0.5979\n",
      "Epoch 5/16\n",
      "\n",
      "Epoch 00005: val_loss improved from 0.95903 to 0.90675, saving model to chkPt1.h5\n",
      " - 2065s - loss: 0.9350 - acc: 0.6130 - val_loss: 0.9067 - val_acc: 0.6364\n",
      "Epoch 6/16\n",
      "\n",
      "Epoch 00006: val_loss improved from 0.90675 to 0.89092, saving model to chkPt1.h5\n",
      " - 1171s - loss: 0.9102 - acc: 0.6239 - val_loss: 0.8909 - val_acc: 0.6350\n",
      "Epoch 7/16\n",
      "\n",
      "Epoch 00007: val_loss improved from 0.89092 to 0.87333, saving model to chkPt1.h5\n",
      " - 3102s - loss: 0.8864 - acc: 0.6386 - val_loss: 0.8733 - val_acc: 0.6478\n",
      "Epoch 8/16\n",
      "\n",
      "Epoch 00008: val_loss improved from 0.87333 to 0.86625, saving model to chkPt1.h5\n",
      " - 1521s - loss: 0.8641 - acc: 0.6496 - val_loss: 0.8663 - val_acc: 0.6542\n",
      "Epoch 9/16\n",
      "\n",
      "Epoch 00009: val_loss improved from 0.86625 to 0.84629, saving model to chkPt1.h5\n",
      " - 6372s - loss: 0.8430 - acc: 0.6570 - val_loss: 0.8463 - val_acc: 0.6573\n",
      "Epoch 10/16\n",
      "\n",
      "Epoch 00010: val_loss did not improve\n",
      " - 1214s - loss: 0.8320 - acc: 0.6627 - val_loss: 0.8478 - val_acc: 0.6567\n",
      "Epoch 11/16\n",
      "\n",
      "Epoch 00011: val_loss improved from 0.84629 to 0.82408, saving model to chkPt1.h5\n",
      " - 4998s - loss: 0.8197 - acc: 0.6693 - val_loss: 0.8241 - val_acc: 0.6734\n",
      "Epoch 12/16\n",
      "\n",
      "Epoch 00012: val_loss improved from 0.82408 to 0.82149, saving model to chkPt1.h5\n",
      " - 1139s - loss: 0.8007 - acc: 0.6771 - val_loss: 0.8215 - val_acc: 0.6693\n",
      "Epoch 13/16\n",
      "\n",
      "Epoch 00013: val_loss improved from 0.82149 to 0.81216, saving model to chkPt1.h5\n",
      " - 1898s - loss: 0.7862 - acc: 0.6840 - val_loss: 0.8122 - val_acc: 0.6718\n",
      "Epoch 14/16\n",
      "\n",
      "Epoch 00014: val_loss improved from 0.81216 to 0.80925, saving model to chkPt1.h5\n",
      " - 1082s - loss: 0.7794 - acc: 0.6882 - val_loss: 0.8093 - val_acc: 0.6732\n",
      "Epoch 15/16\n",
      "\n",
      "Epoch 00015: val_loss improved from 0.80925 to 0.80548, saving model to chkPt1.h5\n",
      " - 1124s - loss: 0.7643 - acc: 0.6988 - val_loss: 0.8055 - val_acc: 0.6737\n",
      "Epoch 16/16\n",
      "\n",
      "Epoch 00016: val_loss improved from 0.80548 to 0.79962, saving model to chkPt1.h5\n",
      " - 1131s - loss: 0.7517 - acc: 0.7006 - val_loss: 0.7996 - val_acc: 0.6868\n"
     ]
    }
   ],
   "source": [
    "# checkpoint\n",
    "checkPoint = ModelCheckpoint(filepath='chkPt1.h5', verbose=1, save_best_only=True)\n",
    "\n",
    "res = model.fit(train_pix, train_ind, epochs=16,\n",
    "                 shuffle=True,\n",
    "                 batch_size=100, \n",
    "                 validation_data=(test_pix, test_ind),\n",
    "                 callbacks=[checkPoint], \n",
    "                 verbose=2)\n",
    "\n",
    "# save model to json\n",
    "model_json = model.to_json()\n",
    "with open(\"model1.json\", \"w\") as json_file:\n",
    "    json_file.write(model_json)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAA0kAAADYCAYAAAA+q6g7AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nIYQQQggh7Pw/5Y7Skz01MPEAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x1466d18d0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "import seaborn as sns\n",
    "plt.figure(figsize=(14,3))\n",
    "plt.subplot(1, 2, 1)\n",
    "plt.suptitle('Optimizer : Adam', fontsize=10)\n",
    "plt.ylabel('Loss', fontsize=16)\n",
    "plt.plot(res.history['loss'], color='b', label='Training Loss')\n",
    "plt.plot(res.history['val_loss'], color='r', label='Validation Loss')\n",
    "plt.legend(loc='upper right')\n",
    "\n",
    "plt.subplot(1, 2, 2)\n",
    "plt.ylabel('Accuracy', fontsize=16)\n",
    "plt.plot(res.history['acc'], color='b', label='Training Accuracy')\n",
    "plt.plot(res.history['val_acc'], color='r', label='Validation Accuracy')\n",
    "plt.legend(loc='lower right')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[0.79961926936605043, 0.68682084146001676]"
      ]
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "score = model.evaluate(test_pix, test_ind, verbose=0)\n",
    "score"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from keras.models import model_from_json\n",
    "import numpy as np\n",
    "\n",
    "class FacialExpressionModel(object):\n",
    "    EMOTIONS_LIST = [\"Angry\", \"Disgust/fear/surprise\",\"Sad\",\"Happy\", \"Neutral\"]\n",
    "\n",
    "    def __init__(self, model_json_file, model_weights_file):\n",
    "        # load model from JSON file\n",
    "        with open(model_json_file, \"r\") as json_file:\n",
    "            loaded_model_json = json_file.read()\n",
    "            self.loaded_model = model_from_json(loaded_model_json)\n",
    "\n",
    "        # load weights into the new model\n",
    "        self.loaded_model.load_weights(model_weights_file)\n",
    "        print(\"Model loaded from disk\")\n",
    "        self.loaded_model.summary()\n",
    "\n",
    "    def predict_emotion(self, img):\n",
    "        self.preds = self.loaded_model.predict(img)\n",
    "\n",
    "        return FacialExpressionModel.EMOTIONS_LIST[np.argmax(self.preds)]\n",
    "\n",
    "\n",
    "if __name__ == '__main__':\n",
    "    pass"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model loaded from disk\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "conv2d_29 (Conv2D)           (None, 44, 44, 64)        1664      \n",
      "_________________________________________________________________\n",
      "batch_normalization_22 (Batc (None, 44, 44, 64)        256       \n",
      "_________________________________________________________________\n",
      "activation_43 (Activation)   (None, 44, 44, 64)        0         \n",
      "_________________________________________________________________\n",
      "conv2d_30 (Conv2D)           (None, 41, 41, 64)        65600     \n",
      "_________________________________________________________________\n",
      "batch_normalization_23 (Batc (None, 41, 41, 64)        256       \n",
      "_________________________________________________________________\n",
      "activation_44 (Activation)   (None, 41, 41, 64)        0         \n",
      "_________________________________________________________________\n",
      "max_pooling2d_17 (MaxPooling (None, 20, 20, 64)        0         \n",
      "_________________________________________________________________\n",
      "dropout_13 (Dropout)         (None, 20, 20, 64)        0         \n",
      "_________________________________________________________________\n",
      "conv2d_31 (Conv2D)           (None, 18, 18, 32)        18464     \n",
      "_________________________________________________________________\n",
      "batch_normalization_24 (Batc (None, 18, 18, 32)        128       \n",
      "_________________________________________________________________\n",
      "activation_45 (Activation)   (None, 18, 18, 32)        0         \n",
      "_________________________________________________________________\n",
      "conv2d_32 (Conv2D)           (None, 16, 16, 32)        9248      \n",
      "_________________________________________________________________\n",
      "batch_normalization_25 (Batc (None, 16, 16, 32)        128       \n",
      "_________________________________________________________________\n",
      "activation_46 (Activation)   (None, 16, 16, 32)        0         \n",
      "_________________________________________________________________\n",
      "max_pooling2d_18 (MaxPooling (None, 8, 8, 32)          0         \n",
      "_________________________________________________________________\n",
      "dropout_14 (Dropout)         (None, 8, 8, 32)          0         \n",
      "_________________________________________________________________\n",
      "flatten_8 (Flatten)          (None, 2048)              0         \n",
      "_________________________________________________________________\n",
      "dense_15 (Dense)             (None, 128)               262272    \n",
      "_________________________________________________________________\n",
      "batch_normalization_26 (Batc (None, 128)               512       \n",
      "_________________________________________________________________\n",
      "activation_47 (Activation)   (None, 128)               0         \n",
      "_________________________________________________________________\n",
      "dropout_15 (Dropout)         (None, 128)               0         \n",
      "_________________________________________________________________\n",
      "dense_16 (Dense)             (None, 5)                 645       \n",
      "_________________________________________________________________\n",
      "activation_48 (Activation)   (None, 5)                 0         \n",
      "=================================================================\n",
      "Total params: 359,173\n",
      "Trainable params: 358,533\n",
      "Non-trainable params: 640\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "import cv2\n",
    "import numpy as np\n",
    "\n",
    "rgb = cv2.VideoCapture(0)\n",
    "facec = cv2.CascadeClassifier('haarcascade_frontalface_default.xml')\n",
    "font = cv2.FONT_HERSHEY_SIMPLEX\n",
    "\n",
    "def __get_data__():\n",
    "    _, fr = rgb.read()\n",
    "    gray = cv2.cvtColor(fr, cv2.COLOR_BGR2GRAY)\n",
    "    faces = facec.detectMultiScale(gray, 1.3, 5)\n",
    "    return faces, fr, gray\n",
    "\n",
    "def start_app(cnn):\n",
    "    skip_frame = 10\n",
    "    data = []\n",
    "    flag = False\n",
    "    ix = 0\n",
    "    while True:\n",
    "        ix += 1\n",
    "        \n",
    "        faces, fr, gray_fr = __get_data__()\n",
    "        for (x, y, w, h) in faces:\n",
    "            fc = gray_fr[y:y+h, x:x+w]\n",
    "            \n",
    "            roi = cv2.resize(fc, (48, 48))\n",
    "            pred = cnn.predict_emotion(roi[np.newaxis, :, :, np.newaxis])\n",
    "\n",
    "            cv2.putText(fr, pred, (x, y), font, 1, (255, 255, 0), 2)\n",
    "            cv2.rectangle(fr,(x,y),(x+w,y+h),(255,0,0),2)\n",
    "\n",
    "        if cv2.waitKey(1) == 27:\n",
    "            break\n",
    "        cv2.imshow('Filter', fr)\n",
    "    cv2.destroyAllWindows()\n",
    "\n",
    "\n",
    "if __name__ == '__main__':\n",
    "    model = FacialExpressionModel(\"model1.json\", \"chkPt1.h5\")\n",
    "    start_app(model)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "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": "3.6.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}