char_cnn_train.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import keras, cv2\n",
    "from keras.preprocessing.image import ImageDataGenerator\n",
    "from keras.models import Sequential\n",
    "from keras.layers import Dense, Flatten, MaxPooling2D, Dropout, Conv2D\n",
    "from keras import optimizers\n",
    "from keras.callbacks import ModelCheckpoint\n",
    "import matplotlib.pyplot as plt"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Found 996 images belonging to 36 classes.\n",
      "Found 222 images belonging to 36 classes.\n"
     ]
    }
   ],
   "source": [
    "train_datagen = ImageDataGenerator(rescale=1./255, width_shift_range=0.1, height_shift_range=0.1, \n",
    "                                   rotation_range=5,fill_mode='constant',shear_range=0.1)\n",
    "train_generator = train_datagen.flow_from_directory(\n",
    "        'character_set/train', \n",
    "        target_size=(28,28),  \n",
    "        batch_size=16,\n",
    "        color_mode = 'grayscale',\n",
    "        class_mode='categorical')\n",
    "\n",
    "validation_generator = train_datagen.flow_from_directory(\n",
    "        'character_set/val',  \n",
    "        target_size=(28,28),\n",
    "        batch_size=1,\n",
    "        color_mode = 'grayscale',\n",
    "        class_mode='categorical')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.image.AxesImage at 0x1557b347888>"
      ]
     },
     "execution_count": 5,
     "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": [
    "plt.imshow(np.squeeze(train_generator[0][0][2]), cmap='gray')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential_4\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "conv2d_7 (Conv2D)            (None, 26, 26, 16)        160       \n",
      "_________________________________________________________________\n",
      "conv2d_8 (Conv2D)            (None, 24, 24, 32)        4640      \n",
      "_________________________________________________________________\n",
      "max_pooling2d_4 (MaxPooling2 (None, 12, 12, 32)        0         \n",
      "_________________________________________________________________\n",
      "dropout_7 (Dropout)          (None, 12, 12, 32)        0         \n",
      "_________________________________________________________________\n",
      "flatten_4 (Flatten)          (None, 4608)              0         \n",
      "_________________________________________________________________\n",
      "dense_7 (Dense)              (None, 64)                294976    \n",
      "_________________________________________________________________\n",
      "dropout_8 (Dropout)          (None, 64)                0         \n",
      "_________________________________________________________________\n",
      "dense_8 (Dense)              (None, 36)                2340      \n",
      "=================================================================\n",
      "Total params: 302,116\n",
      "Trainable params: 302,116\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model = Sequential()\n",
    "model.add(Conv2D(16, kernel_size=(3, 3),\n",
    "                 activation='relu',\n",
    "                 input_shape=(28,28,1)))\n",
    "model.add(Conv2D(32, (3, 3), activation='relu'))\n",
    "model.add(MaxPooling2D(pool_size=(2, 2)))\n",
    "model.add(Dropout(0.25))\n",
    "model.add(Flatten())\n",
    "model.add(Dense(64, activation='relu'))\n",
    "model.add(Dropout(0.5))\n",
    "model.add(Dense(36, activation='softmax'))     # 36 classes: A-Z, 0-9\n",
    "\n",
    "model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "# model.load_weights('char_best_weights--0.9820.hdf5')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "ckpt = ModelCheckpoint('char_weights--{val_accuracy:.4f}.hdf5', monitor='val_accuracy', verbose=0, save_best_only=True)\n",
    "model.compile(loss='categorical_crossentropy', optimizer=optimizers.Adam(lr=1e-4), metrics=['accuracy'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/10\n",
      "996/996 [==============================] - 18s 18ms/step - loss: 0.1158 - accuracy: 0.9537 - val_loss: 3.3379e-06 - val_accuracy: 0.9910\n",
      "Epoch 2/10\n",
      "996/996 [==============================] - 18s 18ms/step - loss: 0.1006 - accuracy: 0.9602 - val_loss: 1.7881e-06 - val_accuracy: 0.9820\n",
      "Epoch 3/10\n",
      "996/996 [==============================] - 17s 18ms/step - loss: 0.0985 - accuracy: 0.9629 - val_loss: 1.4305e-06 - val_accuracy: 1.0000\n",
      "Epoch 4/10\n",
      "996/996 [==============================] - 17s 17ms/step - loss: 0.0937 - accuracy: 0.9626 - val_loss: 1.1921e-07 - val_accuracy: 0.9910\n",
      "Epoch 5/10\n",
      "996/996 [==============================] - 18s 18ms/step - loss: 0.0861 - accuracy: 0.9643 - val_loss: 0.0000e+00 - val_accuracy: 0.9910\n",
      "Epoch 6/10\n",
      "996/996 [==============================] - 17s 17ms/step - loss: 0.0862 - accuracy: 0.9646 - val_loss: 0.0000e+00 - val_accuracy: 0.9910\n",
      "Epoch 7/10\n",
      "996/996 [==============================] - 18s 18ms/step - loss: 0.0864 - accuracy: 0.9657 - val_loss: 0.0000e+00 - val_accuracy: 0.9910\n",
      "Epoch 8/10\n",
      "996/996 [==============================] - 18s 18ms/step - loss: 0.0858 - accuracy: 0.9633 - val_loss: 0.0000e+00 - val_accuracy: 0.9955\n",
      "Epoch 9/10\n",
      "996/996 [==============================] - 18s 18ms/step - loss: 0.0867 - accuracy: 0.9654 - val_loss: 0.0000e+00 - val_accuracy: 1.0000\n",
      "Epoch 10/10\n",
      "996/996 [==============================] - 18s 18ms/step - loss: 0.0836 - accuracy: 0.9654 - val_loss: 0.0000e+00 - val_accuracy: 0.9910\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<keras.callbacks.callbacks.History at 0x1556be13e48>"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.fit_generator(\n",
    "      train_generator,\n",
    "      steps_per_epoch = train_generator.samples ,\n",
    "      validation_data = validation_generator, \n",
    "      validation_steps = validation_generator.samples,\n",
    "      epochs = 10,\n",
    "       callbacks = [ckpt])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[0.0, 0.9864864945411682]"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.evaluate_generator(validation_generator)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "img = cv2.imread('test.png', 0)\n",
    "img = img.reshape(1,28,28,1)\n",
    "img = img/255.0"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{0: 'class_0',\n",
       " 1: 'class_1',\n",
       " 2: 'class_2',\n",
       " 3: 'class_3',\n",
       " 4: 'class_4',\n",
       " 5: 'class_5',\n",
       " 6: 'class_6',\n",
       " 7: 'class_7',\n",
       " 8: 'class_8',\n",
       " 9: 'class_9',\n",
       " 10: 'class_A',\n",
       " 11: 'class_B',\n",
       " 12: 'class_C',\n",
       " 13: 'class_D',\n",
       " 14: 'class_E',\n",
       " 15: 'class_F',\n",
       " 16: 'class_G',\n",
       " 17: 'class_H',\n",
       " 18: 'class_I',\n",
       " 19: 'class_J',\n",
       " 20: 'class_K',\n",
       " 21: 'class_L',\n",
       " 22: 'class_M',\n",
       " 23: 'class_N',\n",
       " 24: 'class_O',\n",
       " 25: 'class_P',\n",
       " 26: 'class_Q',\n",
       " 27: 'class_R',\n",
       " 28: 'class_S',\n",
       " 29: 'class_T',\n",
       " 30: 'class_U',\n",
       " 31: 'class_V',\n",
       " 32: 'class_W',\n",
       " 33: 'class_X',\n",
       " 34: 'class_Y',\n",
       " 35: 'class_Z'}"
      ]
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "labels = (train_generator.class_indices)\n",
    "labels = dict((v,k) for k,v in labels.items())\n",
    "labels"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [],
   "source": [
    "idx = model.predict_classes(img)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'I'"
      ]
     },
     "execution_count": 22,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "labels[idx[0]].split('_')[1]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "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": 4
}