GTSRB_Attempt12.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "from PIL import ImageFile\n",
    "ImageFile.LOAD_TRUNCATED_IMAGES = True\n",
    "import numpy as np\n",
    "#import tensorflow as tf\n",
    "import h5py\n",
    "import keras\n",
    "from keras import backend as K\n",
    "from keras import layers\n",
    "from keras.layers import Input, Multiply,GlobalAveragePooling2D, Add, Dense, Activation, ZeroPadding2D, BatchNormalization, Flatten, Conv2D, AveragePooling2D, MaxPooling2D, GlobalMaxPooling2D, Lambda\n",
    "from keras.models import Model, load_model\n",
    "from keras.preprocessing import image\n",
    "from keras.utils import layer_utils\n",
    "from keras.utils.data_utils import get_file\n",
    "from keras.applications.imagenet_utils import preprocess_input\n",
    "from keras.models import Sequential\n",
    "from keras.layers import Activation\n",
    "from keras.layers.core import Dense, Flatten, Dropout\n",
    "from keras.optimizers import Adam\n",
    "from keras.metrics import categorical_crossentropy\n",
    "from keras.preprocessing.image import ImageDataGenerator\n",
    "from keras.layers.normalization import BatchNormalization\n",
    "from keras.utils.vis_utils import model_to_dot\n",
    "from keras.utils import plot_model\n",
    "from keras.initializers import glorot_uniform\n",
    "from keras.layers.convolutional import *\n",
    "from keras.applications.resnet50 import ResNet50\n",
    "from keras.models import Model\n",
    "from keras.applications.resnet50 import ResNet50\n",
    "from keras.models import Model\n",
    "import keras.utils\n",
    "from keras import callbacks\n",
    "from keras import utils as np_utils\n",
    "from sklearn.metrics import confusion_matrix\n",
    "import itertools\n",
    "import matplotlib.pyplot as plt\n",
    "%matplotlib inline\n",
    "import os\n",
    "from sklearn.metrics import roc_curve, auc\n",
    "from sklearn.model_selection import cross_val_score\n",
    "from sklearn.model_selection import StratifiedKFold, KFold, RepeatedKFold\n",
    "import sys\n",
    "import gc\n",
    "from IPython.display import SVG\n",
    "import scipy.misc\n",
    "from matplotlib.pyplot import imshow"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "### Hyperparameters\n",
    "\n",
    "batch_size = 32\n",
    "image_size = 224\n",
    "epoch =30\n",
    "train_size = 39209\n",
    "test_size = 12630\n",
    "num_classes = 43"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Found 39209 images belonging to 43 classes.\n",
      "Found 12630 images belonging to 43 classes.\n",
      "39209\n"
     ]
    }
   ],
   "source": [
    "#files = os.listdir(os.curdir)\n",
    "#print (os.listdir(\"../input/oct2017/OCT2017\"))\n",
    "#K.clear_session()\n",
    "#gc.collect()\n",
    "train_path =  '/home/sharif-sourajit/Desktop/traffic_sign_classification/Data/GTSRB/Final_Training/Images'\n",
    "test_path =  '/home/sharif-sourajit/Desktop/traffic_sign_classification/Data/GTSRB/Final_Testing/Images'\n",
    "\n",
    "classes=['00000','00001','00002','00003','00004','00005','00006','00007','00008','00009','00010','00011','00012','00013','00014','00015','00016','00017','00018','00019','00020','00021','00022','00023','00024','00025','00026','00027','00028','00029','00030','00031','00032','00033','00034','00035','00036','00037','00038','00039','00040','00041','00042']\n",
    "\n",
    "train_datagen = ImageDataGenerator(rotation_range=30,\n",
    "                             width_shift_range=0.2,\n",
    "                             height_shift_range=0.2,\n",
    "                             shear_range=0.2,\n",
    "                             zoom_range=0.2,\n",
    "                             rescale=1.0/255,\n",
    "                             horizontal_flip=False,\n",
    "                             fill_mode='nearest'\n",
    "                                  )\n",
    "\n",
    "test_datagen = ImageDataGenerator(rescale=1.0/255)\n",
    "\n",
    "train_batches = train_datagen.flow_from_directory(train_path, target_size=(image_size,image_size),color_mode='rgb', classes=classes, batch_size=batch_size,class_mode='categorical')\n",
    "\n",
    "test_batches = test_datagen.flow_from_directory(test_path, target_size=(image_size,image_size),color_mode='rgb', classes=classes, batch_size=batch_size, class_mode='categorical')\n",
    "\n",
    "X = train_batches\n",
    "\n",
    "print (X.samples)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "def dilated_block(X, num_channel, base):\n",
    "    \n",
    "    X_shortcut = X\n",
    "    \n",
    "    X = BatchNormalization(axis=-1, name=base + '/Branch1/bn_1')(X)\n",
    "    X = Activation('relu', name=base + '/Branch1/relu_1')(X)\n",
    "    X = Conv2D(num_channel, (3,3), strides=(1,1), padding='same', name=base + '/Branch1/conv_1', kernel_initializer=glorot_uniform(seed=0))(X)\n",
    "\n",
    "    X = BatchNormalization(axis=-1, name=base + '/Branch1/bn_2')(X)\n",
    "    X= Activation('relu', name=base + '/Branch1/relu_2')(X)\n",
    "    X = Conv2D(num_channel, (3,3), strides=(1,1), padding='same', name=base + '/Branch1/conv_2', kernel_initializer=glorot_uniform(seed=0))(X)\n",
    "    \n",
    "    \n",
    "    ##### Branch2 Start #####\n",
    "    X_shortcut1 = BatchNormalization(axis=-1, name=base + '/Branch2/bn_1')(X_shortcut)\n",
    "    X_shortcut1 = Activation('relu', name=base + '/Branch2/relu_1')(X_shortcut1)\n",
    "    X_shortcut1 = Conv2D(num_channel, (3,3), strides=(1,1), dilation_rate=2, padding='same', name=base + '/Branch2/conv_1', kernel_initializer=glorot_uniform(seed=0))(X_shortcut1)\n",
    "    X = Add(name=base + '/Add1')([X, X_shortcut1])\n",
    "    ##### Branch2 End #####\n",
    "    \n",
    "    X = BatchNormalization(axis=-1, name=base + '/Branch1/bn_3')(X)\n",
    "    X= Activation('relu', name=base + '/Branch1/relu_3')(X)\n",
    "    X = Conv2D(num_channel, (3,3), strides=(1,1), padding='same', name=base + '/Branch1/conv_3', kernel_initializer=glorot_uniform(seed=0))(X)\n",
    "\n",
    "    ##### Branch3 Start #####\n",
    "    X_shortcut2 = BatchNormalization(axis=-1, name=base + '/Branch3/bn_1')(X_shortcut)\n",
    "    X_shortcut2 = Activation('relu', name=base + '/Branch3/relu_1')(X_shortcut2)\n",
    "    X_shortcut2 = Conv2D(num_channel, (3,3), strides=(1,1), dilation_rate=3, padding='same', name=base + '/Branch3/conv_1', kernel_initializer=glorot_uniform(seed=0))(X_shortcut2)\n",
    "    X = Add(name=base + '/Add2')([X, X_shortcut2])\n",
    "    ##### Branch3 End #####    \n",
    "    \n",
    "    return X"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "def CNN_model():\n",
    "    input_shape=(image_size, image_size, 3)\n",
    "    X_input= Input(input_shape)\n",
    "    \n",
    "    X = Conv2D(64, (3,3), strides=(1,1), padding='same', name='conv_1', kernel_initializer=glorot_uniform(seed=0))(X_input)\n",
    "    X = BatchNormalization(axis=-1, name='bn_1')(X)\n",
    "    X = Activation('relu', name='relu_1')(X)\n",
    "    \n",
    "    X = dilated_block(X, 64, 'Block1')\n",
    "    X = dilated_block(X, 64, 'Block2')\n",
    "    #X = dilated_conv_block(X, 16, 'Block2_pool')\n",
    "    \n",
    "    X = MaxPooling2D((3,3), strides=(2,2), padding='same' ,name='pool_1')(X)\n",
    "    \n",
    "    X = dilated_block(X, 64, 'Block3')\n",
    "    X = dilated_block(X, 128, 'Block4')\n",
    "    #X = dilated_conv_block(X, 64, 'Block5_pool')\n",
    "    \n",
    "    X = MaxPooling2D((3,3), strides=(2,2), padding='same' ,name='pool_2')(X)\n",
    "    \n",
    "    \n",
    "    X = dilated_block(X, 256, 'Block7')\n",
    "    #X = dilated_conv_block(X, 128, 'Block7_pool')\n",
    "    \n",
    "    X = MaxPooling2D((3,3), strides=(2,2), padding='same' ,name='pool_3')(X)\n",
    "    \n",
    "    X = dilated_block(X, 256, 'Block8')\n",
    "    #X = Conv2D(43, (3,3), strides=(1,1), padding='same', name='conv_9', kernel_initializer=glorot_uniform(seed=0))(X)\n",
    "    X = AveragePooling2D((8,8), strides=(1,1), name='avg_pool')(X)\n",
    "    \n",
    "    X = Flatten(name='Flatten')(X)\n",
    "    X = Dense(512, name='Dense_1')(X)\n",
    "    #X = Dropout(0.2)(X)\n",
    "    X = Dense(num_classes, name='Dense_2')(X)\n",
    "    X = Activation('softmax', name='classifier')(X)\n",
    "    \n",
    "    model = Model(inputs=X_input, outputs=X, name='Dilated_skipped_model')\n",
    "    #model.compile(Adam(lr=.0001), loss='categorical_crossentropy', metrics=['accuracy'])\n",
    "    model.summary()\n",
    "    \n",
    "    return model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "def callback_for_training(tf_log_dir_name='./tf-log/',patience_lr=2):\n",
    "    cb = [None] * 3\n",
    "    \"\"\"\n",
    "    Tensorboard log callback\n",
    "    \"\"\"\n",
    "    tb = callbacks.TensorBoard(log_dir=tf_log_dir_name, histogram_freq=0)\n",
    "    cb[0]=tb\n",
    "   \n",
    "    \"\"\"\n",
    "    Early Stopping callback\n",
    "    \"\"\"\n",
    "    #Uncomment for usage\n",
    "    # early_stop = callbacks.EarlyStopping(monitor='val_acc', min_delta=0, patience=5, verbose=1, mode='auto',save_best_only=True)\n",
    "    # cb.apppend(early_stop)\n",
    "    \n",
    "    \"\"\"\n",
    "    Model Checkpointer\n",
    "    \"\"\"\n",
    "    mdc = callbacks.ModelCheckpoint('bestmodel-{epoch:02d}-{val_loss:.2f}.hdf5',\n",
    "                                verbose=0,\n",
    "                                monitor='val_acc',\n",
    "                                save_best_only=True)\n",
    "    cb[1]= mdc\n",
    "    \n",
    "    \"\"\"\n",
    "    Reduce Learning Rate\n",
    "    \"\"\"    \n",
    "    reduce_lr_loss = callbacks.ReduceLROnPlateau(monitor='val_loss', factor=1e-1, patience=patience_lr, verbose=1, mode='min', min_delta=1e-4, min_lr=1e-12)\n",
    "    #cb.append(reduce_lr_loss)\n",
    "    cb[2]=reduce_lr_loss\n",
    "    \n",
    "    return cb"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/30\n",
      "1225/1225 [==============================] - 531s 433ms/step - loss: 0.2195 - acc: 0.9414 - val_loss: 0.0835 - val_acc: 0.9768\n",
      "Epoch 2/30\n",
      "1225/1225 [==============================] - 524s 428ms/step - loss: 0.0325 - acc: 0.9916 - val_loss: 0.0970 - val_acc: 0.9748\n",
      "Epoch 3/30\n",
      "1225/1225 [==============================] - 525s 429ms/step - loss: 0.0270 - acc: 0.9931 - val_loss: 0.0706 - val_acc: 0.9806\n",
      "Epoch 4/30\n",
      "1225/1225 [==============================] - 526s 429ms/step - loss: 0.0258 - acc: 0.9940 - val_loss: 0.0488 - val_acc: 0.9860\n",
      "Epoch 5/30\n",
      "1225/1225 [==============================] - 527s 430ms/step - loss: 0.0185 - acc: 0.9955 - val_loss: 0.1062 - val_acc: 0.9743\n",
      "Epoch 6/30\n",
      "1225/1225 [==============================] - 528s 431ms/step - loss: 0.0164 - acc: 0.9955 - val_loss: 0.1373 - val_acc: 0.9606\n",
      "\n",
      "Epoch 00006: ReduceLROnPlateau reducing learning rate to 9.999999747378752e-06.\n",
      "Epoch 7/30\n",
      "1225/1225 [==============================] - 529s 432ms/step - loss: 0.0036 - acc: 0.9990 - val_loss: 0.0318 - val_acc: 0.9913\n",
      "Epoch 8/30\n",
      "1225/1225 [==============================] - 529s 432ms/step - loss: 0.0016 - acc: 0.9996 - val_loss: 0.0337 - val_acc: 0.9911\n",
      "Epoch 9/30\n",
      "1225/1225 [==============================] - 530s 432ms/step - loss: 7.5947e-04 - acc: 0.9998 - val_loss: 0.0320 - val_acc: 0.9915\n",
      "\n",
      "Epoch 00009: ReduceLROnPlateau reducing learning rate to 9.999999747378752e-07.\n",
      "Epoch 10/30\n",
      "1225/1225 [==============================] - 530s 432ms/step - loss: 8.3753e-04 - acc: 0.9998 - val_loss: 0.0299 - val_acc: 0.9922\n",
      "Epoch 11/30\n",
      "1225/1225 [==============================] - 531s 433ms/step - loss: 7.4444e-04 - acc: 0.9998 - val_loss: 0.0305 - val_acc: 0.9919\n",
      "Epoch 12/30\n",
      "1225/1225 [==============================] - 532s 434ms/step - loss: 4.8645e-04 - acc: 0.9998 - val_loss: 0.0288 - val_acc: 0.9928\n",
      "Epoch 13/30\n",
      "1225/1225 [==============================] - 532s 434ms/step - loss: 8.8008e-04 - acc: 0.9998 - val_loss: 0.0281 - val_acc: 0.9930\n",
      "Epoch 14/30\n",
      "1225/1225 [==============================] - 532s 435ms/step - loss: 3.4246e-04 - acc: 1.0000 - val_loss: 0.0273 - val_acc: 0.9932\n",
      "Epoch 15/30\n",
      "1225/1225 [==============================] - 532s 434ms/step - loss: 5.2387e-04 - acc: 0.9998 - val_loss: 0.0263 - val_acc: 0.9929\n",
      "Epoch 16/30\n",
      "1225/1225 [==============================] - 533s 435ms/step - loss: 5.7112e-04 - acc: 0.9998 - val_loss: 0.0267 - val_acc: 0.9933\n",
      "Epoch 17/30\n",
      "1225/1225 [==============================] - 534s 436ms/step - loss: 3.8387e-04 - acc: 0.9999 - val_loss: 0.0270 - val_acc: 0.9927\n",
      "\n",
      "Epoch 00017: ReduceLROnPlateau reducing learning rate to 9.999999974752428e-08.\n",
      "Epoch 18/30\n",
      "1225/1225 [==============================] - 535s 437ms/step - loss: 4.8942e-04 - acc: 0.9999 - val_loss: 0.0264 - val_acc: 0.9929\n",
      "Epoch 19/30\n",
      "1225/1225 [==============================] - 536s 437ms/step - loss: 3.7719e-04 - acc: 0.9999 - val_loss: 0.0262 - val_acc: 0.9930\n",
      "Epoch 20/30\n",
      "1225/1225 [==============================] - 534s 436ms/step - loss: 4.8082e-04 - acc: 0.9999 - val_loss: 0.0270 - val_acc: 0.9927\n",
      "Epoch 21/30\n",
      "1225/1225 [==============================] - 535s 436ms/step - loss: 6.7937e-04 - acc: 0.9998 - val_loss: 0.0264 - val_acc: 0.9930\n",
      "\n",
      "Epoch 00021: ReduceLROnPlateau reducing learning rate to 1.0000000116860975e-08.\n",
      "Epoch 22/30\n",
      "1225/1225 [==============================] - 533s 435ms/step - loss: 3.2889e-04 - acc: 0.9999 - val_loss: 0.0272 - val_acc: 0.9929\n",
      "Epoch 23/30\n",
      "1225/1225 [==============================] - 532s 435ms/step - loss: 6.8164e-04 - acc: 0.9999 - val_loss: 0.0268 - val_acc: 0.9928\n",
      "\n",
      "Epoch 00023: ReduceLROnPlateau reducing learning rate to 9.999999939225292e-10.\n",
      "Epoch 24/30\n",
      "1225/1225 [==============================] - 534s 436ms/step - loss: 2.3637e-04 - acc: 1.0000 - val_loss: 0.0265 - val_acc: 0.9929\n",
      "Epoch 25/30\n",
      "1225/1225 [==============================] - 534s 436ms/step - loss: 4.9491e-04 - acc: 0.9999 - val_loss: 0.0267 - val_acc: 0.9929\n",
      "\n",
      "Epoch 00025: ReduceLROnPlateau reducing learning rate to 9.999999717180686e-11.\n",
      "Epoch 26/30\n",
      "1225/1225 [==============================] - 535s 436ms/step - loss: 2.3547e-04 - acc: 0.9999 - val_loss: 0.0262 - val_acc: 0.9931\n",
      "Epoch 27/30\n",
      "1225/1225 [==============================] - 536s 437ms/step - loss: 3.5911e-04 - acc: 0.9999 - val_loss: 0.0267 - val_acc: 0.9929\n",
      "\n",
      "Epoch 00027: ReduceLROnPlateau reducing learning rate to 9.99999943962493e-12.\n",
      "Epoch 28/30\n",
      "1225/1225 [==============================] - 536s 437ms/step - loss: 2.9634e-04 - acc: 0.9999 - val_loss: 0.0276 - val_acc: 0.9929\n",
      "Epoch 29/30\n",
      "1225/1225 [==============================] - 536s 437ms/step - loss: 2.6736e-04 - acc: 1.0000 - val_loss: 0.0267 - val_acc: 0.9929\n",
      "\n",
      "Epoch 00029: ReduceLROnPlateau reducing learning rate to 1e-12.\n",
      "Epoch 30/30\n",
      "1225/1225 [==============================] - 533s 435ms/step - loss: 2.5227e-04 - acc: 0.9999 - val_loss: 0.0262 - val_acc: 0.9929\n"
     ]
    }
   ],
   "source": [
    "K.clear_session()\n",
    "gc.collect()\n",
    "\n",
    "cb = callback_for_training()\n",
    "\n",
    "model = CNN_model()\n",
    "\n",
    "#model.compile(Adam(lr=.0001), loss='categorical_crossentropy', metrics=['accuracy'])\n",
    "\n",
    "history = model.fit_generator(train_batches, shuffle=True, steps_per_epoch=train_size //batch_size, validation_data=test_batches, validation_steps= test_size//batch_size, epochs=epoch, verbose=1, callbacks=cb)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "#model.save('GTSRB_Train01.h5')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "#model = load_model('GTSRB_Train01.h5')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f588b42b588>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f588b2ea8d0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "def  plot_loss_acc(history):\n",
    "    loss = history.history['loss']\n",
    "    val_loss = history.history['val_loss']\n",
    "    epochs = range(len(loss))\n",
    "    plt.plot(epochs, loss, 'bo')\n",
    "    plt.plot(epochs, val_loss, 'g')\n",
    "    plt.title('Training and validation loss')\n",
    "    plt.legend(['train', 'val'], loc='upper right')\n",
    "    filename= 'GTSRB_Approach08_loss.png'\n",
    "    plt.savefig(filename)\n",
    "    plt.show()\n",
    "    \n",
    "    acc = history.history['acc']\n",
    "    val_acc = history.history['val_acc']\n",
    "    epochs = range(len(acc))\n",
    "    plt.plot(epochs, acc, 'b')\n",
    "    plt.plot(epochs, val_acc, 'g')\n",
    "    plt.title('Training and validation accuracy')\n",
    "    plt.legend(['train', 'val'], loc='lower right')\n",
    "    filename= 'GTSRB_Approach08_accuracy.png'\n",
    "    plt.savefig(filename)\n",
    "    plt.show()\n",
    "    \n",
    "plot_loss_acc(history)"
   ]
  },
  {
   "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": "3.6.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}