train_and_evaluate.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Train Feed-Forward network on pairwise distance between facial landmarks"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Modules\n",
    "from utils import *\n",
    "from data_preprocess import *\n",
    "from models import *\n",
    "\n",
    "# Tensorflow / Keras / sklearn\n",
    "from sklearn.model_selection import train_test_split\n",
    "from tensorflow.keras.utils import to_categorical\n",
    "from tensorflow.keras.applications.vgg16 import preprocess_input\n",
    "from tensorflow.keras.preprocessing import image\n",
    "\n",
    "# Utils\n",
    "from ipywidgets import IntProgress\n",
    "from matplotlib.ticker import MaxNLocator"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Load dataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Load annotations\n",
    "annotations = load_annotations('./data/annotations.csv')\n",
    "# Convert labels to categorical\n",
    "annotations['class'] = labels_to_categorical(annotations[['smile', 'mouth_open']])\n",
    "labels = annotations['class'].values"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Calculate pairwise distance between facial landmarks\n",
    "# distances = np.zeros((len(labels), 2278))\n",
    "# for i, p in enumerate(annotations['points']):\n",
    "#     points = str_to_array(p)\n",
    "#     distances[i] = points_to_distances(points, normalize=True)\n",
    "\n",
    "# np.save('./data/features/distances.npy', distances)\n",
    "distances = np.load('./data/features/distances.npy')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "((865, 2278), (865,))"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "distances.shape, labels.shape"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Analyze distribution by classes"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "calm_closed: 473 images\n",
      "calm_open: 199 images\n",
      "smile_closed: 36 images\n",
      "smile_open: 157 images\n"
     ]
    }
   ],
   "source": [
    "counts = annotations['class'].value_counts()\n",
    "for clss in classes_categorical.keys():\n",
    "    print('{}: {} images'.format(classes_categorical[clss], counts[clss]))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "In order to balance `smile_closed` class, let's add more examples of this class to the dataset."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "def bootstrap(X, indicies, n_repeats):\n",
    "    # Add to features\n",
    "    extras = np.copy(X[indicies])\n",
    "    extras = np.repeat(extras, n_repeats, axis=0)\n",
    "    X_new = np.concatenate([X, extras], axis=0)\n",
    "    return X_new"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Now 108 images in smile_closed\n"
     ]
    }
   ],
   "source": [
    "distances = bootstrap(distances, labels==2, 2)\n",
    "labels = bootstrap(labels, labels==2, 2)\n",
    "\n",
    "print('Now {} images in smile_closed'.format((labels == 2).sum()))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "# # Split into train and test datasets\n",
    "X_train, X_test, y_train, y_test = train_test_split(distances, to_categorical(labels), test_size=0.15, shuffle=True, stratify=labels)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Train Feed-Forward model on pairwise distances"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "# model = FeedForward(input_shape=(2278,), output_shape=4, layers_dims=[1000, 700, 500, 500], dropout_rate=0.4)\n",
    "# optimizer = keras.optimizers.Adam(lr=0.0001, decay=0.0005)\n",
    "# model.compile(optimizer=optimizer, loss='categorical_crossentropy', metrics=['accuracy'])\n",
    "# # model.summary()\n",
    "model = tf.keras.models.load_model('./models/ff_1000_700_500_500.h5')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 95,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "717de4d3660342fc9a40e0733c03621c",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "IntProgress(value=1, description='Initializing...', min=1)"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1152x432 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "epochs = 100\n",
    "history = model.fit(\n",
    "    x=X_train, \n",
    "    y=y_train, \n",
    "    batch_size=32, \n",
    "    epochs=epochs, \n",
    "    verbose=0, \n",
    "    callbacks=[TrainProgressBar(max_value=epochs)], \n",
    "    # validation_split=0.2, \n",
    "    shuffle=True)\n",
    "plot_history(history)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "141/141 [==============================] - 0s 681us/step\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[0.3451385633319828, 0.8794326249589312]"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.evaluate(X_test, y_test)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Test on new images"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "8de571c734424e3187a207e4531ad0fb",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "interactive(children=(Dropdown(description='image', options=(PosixPath('data/test/1290921548_1.jpg'), PosixPat…"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "test_path = './data/test/'\n",
    "predictor_path = './dlib_landmarks_predictor/shape_predictor_68_face_landmarks.dat'\n",
    "\n",
    "@interact(image = list(Path(test_path).glob('*.jpg')))\n",
    "def predict(image):\n",
    "    predictions = []\n",
    "    img = imread(image.as_posix())\n",
    "    _, landmarks = detect_faces(img, predictor_path, plot=True, plot_landmarks=False)\n",
    "    for i, points in enumerate(landmarks):\n",
    "        distances = points_to_distances(points, normalize=True)\n",
    "        model_output = model.predict(distances[np.newaxis]).squeeze()\n",
    "        label = np.argmax(model_output)\n",
    "        class_name = classes_categorical[label]\n",
    "        print('[{}] {} ({})'.format(i, class_name, model_output))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Save the trained model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 92,
   "metadata": {},
   "outputs": [],
   "source": [
    "# model.save('./models/ff_1000_700_500_500.h5')\n",
    "# model.save_weights('./models/ff_1000_700_500_500_weigths.h5')\n",
    "# tf.keras.utils.plot_model(model, to_file='./models/ff_1000_700_500_500.png', show_shapes=True, show_layer_names=True)"
   ]
  }
 ],
 "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
}