Image Data Augmentation with Keras - Complete.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Image Data Augmentation with Keras\n",
    "\n",
    "![Horizontal Flip](assets/horizontal_flip.jpg)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Task 1: Import Libraries"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "%matplotlib inline\n",
    "\n",
    "import os\n",
    "import numpy as np\n",
    "import tensorflow as tf\n",
    "\n",
    "from PIL import Image\n",
    "from matplotlib import pyplot as plt\n",
    "\n",
    "print('Using TensorFlow', tf.__version__)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Task 2: Rotation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "generator = tf.keras.preprocessing.image.ImageDataGenerator(\n",
    "    rotation_range=40\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "image_path = 'images/train/cat/cat.jpg'\n",
    "\n",
    "plt.imshow(plt.imread(image_path));"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "x, y = next(generator.flow_from_directory('images', batch_size=1))\n",
    "plt.imshow(x[0].astype('uint8'));"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Task 3: Width and Height Shifts"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "generator = tf.keras.preprocessing.image.ImageDataGenerator(\n",
    "    width_shift_range=[-40, -20, 0, 20, 40],\n",
    "    height_shift_range=[-50,50]\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "x, y = next(generator.flow_from_directory('images', batch_size=1))\n",
    "plt.imshow(x[0].astype('uint8'));"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Task 4: Brightness"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "generator = tf.keras.preprocessing.image.ImageDataGenerator(\n",
    "    brightness_range=(0., 2.)\n",
    ")\n",
    "\n",
    "x, y = next(generator.flow_from_directory('images', batch_size=1))\n",
    "plt.imshow(x[0].astype('uint8'));"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Task 5: Shear Transformation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "generator = tf.keras.preprocessing.image.ImageDataGenerator(\n",
    "    shear_range=45\n",
    ")\n",
    "\n",
    "x, y = next(generator.flow_from_directory('images', batch_size=1))\n",
    "plt.imshow(x[0].astype('uint8'));"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Task 6: Zoom"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "generator = tf.keras.preprocessing.image.ImageDataGenerator(\n",
    "    zoom_range=0.5\n",
    ")\n",
    "\n",
    "x, y = next(generator.flow_from_directory('images', batch_size=1))\n",
    "plt.imshow(x[0].astype('uint8'));"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Task 7: Channel Shift"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "generator = tf.keras.preprocessing.image.ImageDataGenerator(\n",
    "    channel_shift_range=100\n",
    ")\n",
    "\n",
    "x, y = next(generator.flow_from_directory('images', batch_size=1))\n",
    "plt.imshow(x[0].astype('uint8'));"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Task 8: Flips"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "generator = tf.keras.preprocessing.image.ImageDataGenerator(\n",
    "    horizontal_flip=True,\n",
    "    vertical_flip=True\n",
    ")\n",
    "\n",
    "x, y = next(generator.flow_from_directory('images', batch_size=1))\n",
    "plt.imshow(x[0].astype('uint8'));"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Task 9: Normalization\n",
    "\n",
    "### Featurewise"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "(x_train, y_train), (x_test, y_test) = tf.keras.datasets.cifar10.load_data()\n",
    "\n",
    "generator = tf.keras.preprocessing.image.ImageDataGenerator(\n",
    "    featurewise_center=True,\n",
    "    featurewise_std_normalization=True\n",
    ")\n",
    "\n",
    "generator.fit(x_train)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "x, y = next(generator.flow(x_train, y_train, batch_size=1))\n",
    "print(x.mean(), x.std(), y)\n",
    "print(x_train.mean())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Samplewise"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "generator = tf.keras.preprocessing.image.ImageDataGenerator(\n",
    "    samplewise_center=True,\n",
    "    samplewise_std_normalization=True\n",
    ")\n",
    "\n",
    "x, y = next(generator.flow(x_train, y_train, batch_size=1))\n",
    "print(x.mean(), x.std(), y)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Task 10: Rescale and Preprocessing Function"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "generator = tf.keras.preprocessing.image.ImageDataGenerator(\n",
    "    preprocessing_function=tf.keras.applications.mobilenet_v2.preprocess_input,\n",
    "    rescale=1.\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "x, y = next(generator.flow(x_train, y_train, batch_size=1))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "print(x.mean(), x.std(), y)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Task 11: Using in Model Training"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "model = tf.keras.models.Sequential([\n",
    "    tf.keras.applications.mobilenet_v2.MobileNetV2(include_top=False, input_shape=(32, 32, 3), pooling='avg'),\n",
    "    tf.keras.layers.Dense(10, activation='softmax')\n",
    "])\n",
    "\n",
    "model.compile(loss='sparse_categorical_crossentropy', optimizer='adam', metrics=['accuracy'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "_ = model.fit(\n",
    "    generator.flow(x_train, y_train, batch_size=32),\n",
    "    steps_per_epoch=10, epochs=1\n",
    ")"
   ]
  },
  {
   "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": "3.6.8"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}