Forms_Project.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 199,
   "metadata": {},
   "outputs": [],
   "source": [
    "# import required packages and utilities\n",
    "import numpy as np \n",
    "import pandas as pd \n",
    "import matplotlib.pyplot as plt\n",
    "import tensorflow as tf\n",
    "from tensorflow.keras.models import Sequential\n",
    "from tensorflow.keras.layers import Dense, Flatten, Dropout, Activation, Conv2D, MaxPooling2D, Input\n",
    "from tensorflow.keras.preprocessing.image import load_img\n",
    "from tensorflow.keras.preprocessing.image import img_to_array\n",
    "from sklearn.model_selection import train_test_split\n",
    "from sklearn.model_selection import StratifiedKFold\n",
    "\n",
    "# Vgg16 utils\n",
    "from tensorflow.keras.applications.vgg16 import VGG16\n",
    "from tensorflow.keras.applications.vgg16 import preprocess_input as vgg_preprocess \n",
    "\n",
    "# Resnet utils \n",
    "from tensorflow.keras.applications.resnet50 import ResNet50\n",
    "from tensorflow.keras.applications.resnet50 import preprocess_input as resnet_preprocess\n",
    "\n",
    "# utils to dump \n",
    "from pickle import dump\n",
    "# import glob\n",
    "import glob"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "# load image\n",
    "image = load_img(\"/Users/varun/trunk/kumud/COURSES/EAI6010/Final_Project/images/iCard_021873_1_Daba_Ayehush_H.jpg\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.image.AxesImage at 0xb301112b0>"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# plot how the image looks like\n",
    "plt.imshow(image)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "data_dir = \"/Users/varun/trunk/kumud/COURSES/EAI6010/Final_Project/forms/\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [],
   "source": [
    "X = []\n",
    "Y = []\n",
    "for img_path in glob.glob(data_dir+\"*/*.jpg\"):\n",
    "    img = img_path.split(\"/\")\n",
    "    label = None\n",
    "    if img[-2] == \"Type-1 blocks\":\n",
    "        label = 1\n",
    "    elif img[-2] == \"Type-2 Empty\":\n",
    "        label = 2\n",
    "    elif img[-2] == \"Type-3 lines\":\n",
    "        label = 3\n",
    "    elif img[-2] == \"HandWritten\":\n",
    "        label = 0\n",
    "    else:\n",
    "        print(\"Unknown label type\", img[-2])\n",
    "    Y.append(label)\n",
    "    X.append(img_to_array(load_img(img_path,target_size=(224, 224))))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "120 120\n"
     ]
    }
   ],
   "source": [
    "print(len(Y),len(X))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[0, 0, 0, 0, 0, 0, 1, 1, 1, 1]"
      ]
     },
     "execution_count": 27,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "Y[:10]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(224, 224, 3)"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X[1].shape"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Image calssification\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Train Model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 233,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Build first model\n",
    "model1 = Sequential([\n",
    "    Dense(64,input_shape = (224, 224, 3), activation='relu'),\n",
    "    Dense(128, activation ='relu'),\n",
    "    Flatten(),\n",
    "    Dense(4, activation='softmax')])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 234,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential_253\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "dense_509 (Dense)            (None, 224, 224, 64)      256       \n",
      "_________________________________________________________________\n",
      "dense_510 (Dense)            (None, 224, 224, 128)     8320      \n",
      "_________________________________________________________________\n",
      "flatten_253 (Flatten)        (None, 6422528)           0         \n",
      "_________________________________________________________________\n",
      "dense_511 (Dense)            (None, 4)                 25690116  \n",
      "=================================================================\n",
      "Total params: 25,698,692\n",
      "Trainable params: 25,698,692\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model1.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 235,
   "metadata": {},
   "outputs": [],
   "source": [
    "# compile model\n",
    "model1.compile(optimizer=tf.keras.optimizers.Adam(0.001),\n",
    "                  loss='categorical_crossentropy',\n",
    "                  metrics=['accuracy'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 236,
   "metadata": {},
   "outputs": [],
   "source": [
    "# splitting into train test\n",
    "X_train, X_test, y_train, y_test = train_test_split(X, Y, test_size=0.20, random_state = 123)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 221,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[0, 2, 2, 2, 1, 1, 2, 3, 1, 1, 1, 1, 0, 0, 1, 3, 2, 3, 2, 3, 1, 2, 3, 3]"
      ]
     },
     "execution_count": 221,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# inspect test labels contains all tyoes of forms\n",
    "y_test"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 222,
   "metadata": {},
   "outputs": [],
   "source": [
    "y_train_cat = tf.keras.utils.to_categorical(y_train)\n",
    "y_test_cat = tf.keras.utils.to_categorical(y_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 237,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 96 samples\n",
      "Epoch 1/10\n",
      "96/96 [==============================] - 14s 147ms/sample - loss: 22179.8059 - accuracy: 0.4062\n",
      "Epoch 2/10\n",
      "96/96 [==============================] - 11s 118ms/sample - loss: 4182.6472 - accuracy: 0.5521\n",
      "Epoch 3/10\n",
      "96/96 [==============================] - 11s 112ms/sample - loss: 1191.6945 - accuracy: 0.8750\n",
      "Epoch 4/10\n",
      "96/96 [==============================] - 11s 111ms/sample - loss: 442.1157 - accuracy: 0.9167\n",
      "Epoch 5/10\n",
      "96/96 [==============================] - 10s 105ms/sample - loss: 31.7618 - accuracy: 0.9583\n",
      "Epoch 6/10\n",
      "96/96 [==============================] - 10s 105ms/sample - loss: 0.0000e+00 - accuracy: 1.0000\n",
      "Epoch 7/10\n",
      "96/96 [==============================] - 10s 107ms/sample - loss: 0.0000e+00 - accuracy: 1.0000\n",
      "Epoch 8/10\n",
      "96/96 [==============================] - 11s 111ms/sample - loss: 0.0000e+00 - accuracy: 1.0000\n",
      "Epoch 9/10\n",
      "96/96 [==============================] - 11s 111ms/sample - loss: 0.0000e+00 - accuracy: 1.0000\n",
      "Epoch 10/10\n",
      "96/96 [==============================] - 11s 116ms/sample - loss: 0.0000e+00 - accuracy: 1.0000\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<tensorflow.python.keras.callbacks.History at 0x1aaf030eb8>"
      ]
     },
     "execution_count": 237,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# fitting our first model : Feedforward NN \n",
    "model1.fit(np.array(X_train), np.array(y_train_cat), epochs=10, batch_size=10,verbose =1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 238,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\r",
      "24/1 [================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================] - 1s 52ms/sample - loss: 670.9678 - accuracy: 0.8333\n"
     ]
    }
   ],
   "source": [
    "# Accuracy on test data\n",
    "score = model1.evaluate(np.array(X_test), np.array(y_test_cat), verbose=1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Image classification using Transfer learning \n",
    "## Using pre-trained deep learning models as feature extractor"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 200,
   "metadata": {},
   "outputs": [],
   "source": [
    "# function to extract features using pre-trained models\n",
    "# extract features from each form in the directory\n",
    "\n",
    "def extract_features_vgg(X):\n",
    "    # load the model\n",
    "    in_layer = Input(shape=(224, 224, 3))\n",
    "    model = VGG16(include_top=False, input_tensor=in_layer)\n",
    "    # extract features from each photo\n",
    "    features = []\n",
    "    for image in X:\n",
    "        # reshape data for the model\n",
    "        image = image.reshape((1, image.shape[0], image.shape[1], image.shape[2]))\n",
    "        # prepare the image for the VGG model\n",
    "        image = vgg_preprocess(image)\n",
    "        # get features\n",
    "        feature = model.predict(image, verbose=0)\n",
    "        # store feature\n",
    "        features.append(feature)\n",
    "    return features\n",
    "\n",
    "\n",
    "def extract_features_resnet(X):\n",
    "# load the model\n",
    "    in_layer = Input(shape=(224, 224, 3))\n",
    "    model = ResNet50(include_top=False, input_tensor=in_layer)\n",
    "    # extract features from each photo\n",
    "    features = []\n",
    "    for image in X:\n",
    "        # reshape data for the model\n",
    "        image = image.reshape((1, image.shape[0], image.shape[1], image.shape[2]))\n",
    "        # prepare the image for the ResNet model\n",
    "        image = resnet_preprocess(image)\n",
    "        # get features\n",
    "        feature = model.predict(image, verbose=0)\n",
    "        # store feature\n",
    "        features.append(feature)\n",
    "    return features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 201,
   "metadata": {},
   "outputs": [],
   "source": [
    "# extracting features using VGG\n",
    "X_vgg_features = extract_features_vgg(X)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 120,
   "metadata": {},
   "outputs": [],
   "source": [
    "# convert into categorical\n",
    "y_cat = tf.keras.utils.to_categorical(Y)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# save extracted features \n",
    "dict_to_dump = {\"X\":X_vgg_features, \"Y\": Y}\n",
    "# save to file\n",
    "dump(dict_to_dump, open('vgg_features.pkl', 'wb'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [],
   "source": [
    "# extracting features using VGG\n",
    "X_resnet_features = extract_features_resnet(X)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 202,
   "metadata": {},
   "outputs": [],
   "source": [
    "#Ref: https://www.tensorflow.org/guide/keras/overview\n",
    "# build our own model\n",
    "def build_model(hidden_size = 128):\n",
    "    # build a feed forward NN \n",
    "    import tensorflow.keras as keras\n",
    "\n",
    "    fnn_model = keras.Sequential()\n",
    "    # Add a densely-connected layer \n",
    "    fnn_model.add(keras.layers.Flatten())\n",
    "    fnn_model.add(keras.layers.Dense(hidden_size, activation='relu'))\n",
    "    fnn_model.add(keras.layers.Dropout(0.2))\n",
    "    # Add a softmax layer with 4 output units:\n",
    "    fnn_model.add(keras.layers.Dense(4, activation='softmax'))\n",
    "\n",
    "    fnn_model.compile(optimizer=tf.keras.optimizers.Adam(0.001),\n",
    "                  loss='categorical_crossentropy',\n",
    "                  metrics=['accuracy'])\n",
    "    return fnn_model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 206,
   "metadata": {},
   "outputs": [],
   "source": [
    "# performing k-fold cross-validation\n",
    "kfold = StratifiedKFold(n_splits=6, shuffle=True, random_state=123)\n",
    "cvscores_vgg = []\n",
    "for train, test in kfold.split(X, Y):\n",
    "    fnn_model = build_model(hidden_size=128)\n",
    "    fnn_model.fit(np.array(X_vgg_features)[train], y_cat[train] , epochs=10, batch_size=128,verbose =0)\n",
    "    score = fnn_model.evaluate(np.array(X_vgg_features)[test],y_cat[test],verbose=0)\n",
    "    cvscores_vgg.append(score[1])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 207,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "The accuracy for Vgg fold 1 is 0.95\n",
      "The accuracy for Vgg fold 2 is 0.95\n",
      "The accuracy for Vgg fold 3 is 0.95\n",
      "The accuracy for Vgg fold 4 is 0.95\n",
      "The accuracy for Vgg fold 5 is 1.00\n",
      "The accuracy for Vgg fold 6 is 0.95\n"
     ]
    }
   ],
   "source": [
    "# print 6 fold cross validation acccuracy \n",
    "for i,score in enumerate(cvscores_vgg):\n",
    "    print(\"The accuracy for Vgg fold {} is {:.2f}\".format(i+1,score))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 208,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "The average accuracy on VGG 6-fold cross validation is 0.96 with standard deviation 0.02\n"
     ]
    }
   ],
   "source": [
    "# taking average of accuracy of 6 fold\n",
    "print(\"The average accuracy on VGG 6-fold cross validation is {:.2f} with standard deviation {:.2f}\"\\\n",
    "      .format(np.mean(cvscores_vgg),np.std(cvscores_vgg)))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Classifier using ResNet50 extracted features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 210,
   "metadata": {},
   "outputs": [],
   "source": [
    "cvscores_resnet = []\n",
    "for train, test in kfold.split(X, Y):\n",
    "    fnn_model = build_model(hidden_size=256)\n",
    "    fnn_model.fit(np.array(X_resnet_features)[train], y_cat[train] , epochs=10, batch_size=32,verbose =0)\n",
    "    score = fnn_model.evaluate(np.array(X_resnet_features)[test],y_cat[test],verbose=0)\n",
    "    cvscores_resnet.append(score[1])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 212,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "The accuracy for Resnet fold 1 is 1.00\n",
      "The accuracy for Resnet fold 2 is 1.00\n",
      "The accuracy for Resnet fold 3 is 0.95\n",
      "The accuracy for Resnet fold 4 is 0.95\n",
      "The accuracy for Resnet fold 5 is 0.95\n",
      "The accuracy for Resnet fold 6 is 0.95\n"
     ]
    }
   ],
   "source": [
    "# print 6 fold cross validation acccuracy for Resnet50\n",
    "for i,score in enumerate(cvscores_resnet):\n",
    "    print(\"The accuracy for Resnet fold {} is {:.2f}\".format(i+1,score))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 213,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "The average accuracy on 6-fold cross validation for Resnet is 0.97 with standard deviation 0.02\n"
     ]
    }
   ],
   "source": [
    "# taking average of accuracy of k fold cv\n",
    "print(\"The average accuracy on 6-fold cross validation for Resnet is {:.2f} with standard deviation {:.2f}\"\\\n",
    "      .format(np.mean(cvscores_resnet),np.std(cvscores_resnet)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 174,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[1.0, 0.95238096, 0.95, 0.94736844, 0.94736844, 1.0]"
      ]
     },
     "execution_count": 174,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "cvscores_resnet"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 214,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[0.95454544, 0.95238096, 0.95, 0.94736844, 1.0, 0.94736844]"
      ]
     },
     "execution_count": 214,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "cvscores_vgg"
   ]
  },
  {
   "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": "3.7.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}