Face Recognition - SP2 - V2.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import warnings\n",
    "warnings.filterwarnings('ignore')\n",
    "import pickle\n",
    "from helper_functions import show_distances_distributions, get_performance\n",
    "import numpy as np\n",
    "import face_recognition\n",
    "import matplotlib.pyplot as plt\n",
    "import cv2\n",
    "import matplotlib.image as imreader\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Loading training data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "face_db_path = \"/home/asoriano/Stratio/spaceAI/evolution-vision/src/resources/faces-dbs/sp2-v1/spaceai-evolution-train-v2-raw.db\"\n",
    "with open(face_db_path, 'rb') as fp:\n",
    "    face_db = pickle.load(fp)\n",
    "train_face_names = face_db['names']\n",
    "train_face_embeddings = face_db['embedding']\n",
    "train_img_paths = face_db['images_path']\n",
    "train_img_sizes = face_db['sizes']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# · Getting number of samples and faces\n",
    "n_samples = len(train_face_names)\n",
    "n_faces = len(set(train_face_names))\n",
    "\n",
    "print(\"- Number of samples: \" + str(n_samples))\n",
    "print(\"- Number of faces: \" + str(n_faces))\n",
    "print(set(train_face_names))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Image paths"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_img_paths[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "tais_path_prefix  = \"/mnt/hdd1tb/evolution-vision/face-recognition/resources/train_dir\"\n",
    "local_path_prefix = \"/home/asoriano/Escritorio/Facial-sp2-v1\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "def local_path(path):\n",
    "    return path.replace(tais_path_prefix, local_path_prefix)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Distances distribution"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "show_distances_distributions(train_face_names, train_face_embeddings)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Loading eval data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "face_db_path = \"/home/asoriano/Stratio/spaceAI/evolution-vision/src/resources/faces-dbs/sp2-v1/spaceai-evolution-evaluation-v1.db\"\n",
    "with open(face_db_path, 'rb') as fp:\n",
    "    face_db = pickle.load(fp)\n",
    "eval_face_names = face_db['names']\n",
    "eval_face_embeddings = face_db['embedding']\n",
    "eval_img_paths = face_db['images_path']\n",
    "eval_img_sizes = face_db['sizes']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "eval_face_names = [\"Hugo Dominguez\" if n==\"Hugo QA\" else n for n in eval_face_names]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# · Getting number of samples and faces\n",
    "n_samples = len(eval_face_names)\n",
    "n_faces = len(set(eval_face_names))\n",
    "print(\"- Number of samples: \" + str(n_samples))\n",
    "print(\"- Number of faces: \" + str(n_faces))\n",
    "print(set(eval_face_names))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Distances distribution"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "show_distances_distributions(eval_face_names, eval_face_embeddings)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Getting unknown people in evaluation video"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "unknown_people"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "unknown_people = list(set(eval_face_names) - set(train_face_names))\n",
    "eval_face_names_unk = [\"unknown\" if n in unknown_people else n for n in eval_face_names]\n",
    "set(eval_face_names_unk)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Performance analysis"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "opt_tau, f1, accuracy, labeled_name, predicted_name, all_distances, all_indices_th = \\\n",
    "    get_performance(1, np.arange(0.3, 0.7, 0.05), train_face_names, train_face_embeddings, eval_face_names_unk, eval_face_embeddings, \"unknown\")\n",
    "\n",
    "print(\"Optimal threshold\" + str(opt_tau))\n",
    "print(\"F1-Score\" + str(f1))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Optimal threshold: 0.45\n",
      "F1-Score: [0.145089778953989, 0.32834222394002227, 0.6619780596748418, 0.81881411304981211, 0.78994252558934219, 0.79093556458469061, 0.79093556458469061, 0.79093556458469061]\n"
     ]
    }
   ],
   "source": [
    "opt_tau, f1, accuracy, labeled_name, predicted_name, all_distances, all_indices_th = \\\n",
    "    get_performance(3, np.arange(0.3, 0.7, 0.05), train_face_names, train_face_embeddings, eval_face_names_unk, eval_face_embeddings, \"unknown\")\n",
    "\n",
    "print(\"Optimal threshold: \" + str(opt_tau))\n",
    "print(\"F1-Score: \" + str(f1))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Optimal threshold: 0.45\n",
      "F1-Score: [0.13492385503022672, 0.28711134037191438, 0.59809112209079629, 0.82195076561912384, 0.79371754240473746, 0.79588932751532437, 0.79588932751532437, 0.79588932751532437]\n"
     ]
    }
   ],
   "source": [
    "opt_tau, f1, accuracy, labeled_name, predicted_name, all_distances, all_indices_th = \\\n",
    "    get_performance(5, np.arange(0.3, 0.7, 0.05), train_face_names, train_face_embeddings, eval_face_names_unk, eval_face_embeddings, \"unknown\")\n",
    "\n",
    "print(\"Optimal threshold: \" + str(opt_tau))\n",
    "print(\"F1-Score: \" + str(f1))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Error analysis"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "unrecognized_errors = []\n",
    "misclassified_errors = []\n",
    "false_positive_errors = []\n",
    "\n",
    "for test_idx, (tag, pred, distances, indexes) in enumerate(zip(labeled_name, predicted_name, all_distances, all_indices_th)):\n",
    "    if tag != pred:\n",
    "        if tag == \"unknown\":\n",
    "            false_positive_errors.append((test_idx, distances, indexes))\n",
    "        if tag != \"unknown\" and pred != \"unknown\":\n",
    "            misclassified_errors.append((test_idx, distances, indexes))\n",
    "        if pred == \"unknown\":\n",
    "            unrecognized_errors.append((test_idx, distances, indexes))\n",
    "\n",
    "print(\"-----------------------------------------------------------\")\n",
    "print(\" => Unrecognized errors\")\n",
    "print(\"-----------------------------------------------------------\")\n",
    "\n",
    "for unrecognized_error in unrecognized_errors:\n",
    "    test_idx, distances, indexes = unrecognized_error\n",
    "    plt.imshow(plt.imread(local_path(eval_img_paths[test_idx])))\n",
    "    plt.title(\"Idx: \" + str(test_idx) + \" \" + eval_img_paths[test_idx].replace(tais_path_prefix,\"\"))\n",
    "    plt.xlabel(eval_img_sizes[test_idx])\n",
    "    plt.show()\n",
    "\n",
    "print(\"-----------------------------------------------------------\")\n",
    "print(\" => Misclassified errors\")\n",
    "print(\"-----------------------------------------------------------\")\n",
    "\n",
    "for misclassified_error in misclassified_errors:\n",
    "    test_idx, distances, indexes = misclassified_error\n",
    "    f, ax = plt.subplots(1, 2, figsize=(10,6)) \n",
    "    ax[0].imshow( plt.imread(local_path(eval_img_paths[test_idx])) )\n",
    "    ax[0].set_xlabel(eval_img_sizes[test_idx])\n",
    "    ax[0].set_title(\"Idx: \" + str(test_idx))\n",
    "    ax[1].imshow( plt.imread(local_path(train_img_paths[indexes[0]])) )\n",
    "    ax[1].set_xlabel(train_img_sizes[indexes[0]])\n",
    "    ax[1].set_title(\"Idx: \" + str(indexes[0]))\n",
    "    plt.show()\n",
    "\n",
    "print(\"-----------------------------------------------------------\")\n",
    "print(\" => False positive errors\")\n",
    "print(\"-----------------------------------------------------------\")\n",
    "\n",
    "for false_positive_error in false_positive_errors:\n",
    "    test_idx, distances, indexes = false_positive_error\n",
    "    f, ax = plt.subplots(1, 2, figsize=(10,6)) \n",
    "    ax[0].imshow( plt.imread(local_path(eval_img_paths[test_idx])) )\n",
    "    ax[0].set_xlabel(eval_img_sizes[test_idx])\n",
    "    ax[0].set_title(\"Idx: \" + str(test_idx))\n",
    "    ax[1].imshow( plt.imread(local_path(train_img_paths[indexes[0]])) ) \n",
    "    ax[1].set_xlabel(train_img_sizes[indexes[0]])\n",
    "    ax[1].set_title(\"Idx: \" + str(indexes[0]))\n",
    "    plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Error - size distribution"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [],
   "source": [
    "eval_errors = []\n",
    "eval_errors.extend(unrecognized_errors)\n",
    "eval_errors.extend(misclassified_errors)\n",
    "eval_errors.extend(false_positive_errors)\n",
    "\n",
    "eval_sizes = [ eval_img_sizes[idx] for idx, _, _ in eval_errors ]\n",
    "\n",
    "train_errors = []\n",
    "train_errors.extend(misclassified_errors)\n",
    "train_errors.extend(false_positive_errors)\n",
    "train_sizes = [ train_img_sizes[idx[0]] for _, _, idx in train_errors ]\n",
    "\n",
    "error_sizes = eval_sizes\n",
    "error_sizes.extend(train_sizes)\n",
    "w_e, h_e = zip(*error_sizes)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "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.hist(w_e, 20, alpha=0.9, density=False, label=\"Width - error distribution\")\n",
    "plt.show()\n",
    "\n",
    "plt.hist(h_e, 20, alpha=0.9, density=False, label=\"Height - error distribution\")\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Remove small faces"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [],
   "source": [
    "min_size = 110"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [],
   "source": [
    "idx_train_small = [i for i, (w,h) in enumerate(train_img_sizes) if w > min_size]\n",
    "\n",
    "train_face_names = np.array(train_face_names)[idx_train_small].tolist()\n",
    "train_face_embeddings = [e for i, e in enumerate(train_face_embeddings) if i in idx_train_small]\n",
    "train_img_paths = np.array(train_img_paths)[idx_train_small].tolist()\n",
    "train_img_sizes = np.array(train_img_sizes)[idx_train_small].tolist()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "3722"
      ]
     },
     "execution_count": 31,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(train_face_names)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [],
   "source": [
    "new_face_db = {}\n",
    "new_face_db['names'] = train_face_names\n",
    "new_face_db['embedding'] = train_face_embeddings\n",
    "new_face_db['images_path'] = train_img_paths\n",
    "new_face_db['sizes'] = train_img_sizes"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [],
   "source": [
    "with open(str(\"spaceai-evolution-train-v2-no-small.db\"), 'wb') as fp:\n",
    "    pickle.dump(new_face_db, fp)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [],
   "source": [
    "idx_eval_small = [i for i, (w,h) in enumerate(eval_img_sizes) if w > min_size]\n",
    "\n",
    "eval_face_names = np.array(eval_face_names)[idx_eval_small].tolist()\n",
    "eval_face_names_unk = np.array(eval_face_names_unk)[idx_eval_small].tolist()\n",
    "eval_face_embeddings = [e for i, e in enumerate(eval_face_embeddings) if i in idx_eval_small]\n",
    "eval_img_paths = np.array(eval_img_paths)[idx_eval_small].tolist()\n",
    "eval_img_sizes = np.array(eval_img_sizes)[idx_eval_small].tolist()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [
    {
     "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": [
    "show_distances_distributions(train_face_names, train_face_embeddings)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [
    {
     "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": [
    "show_distances_distributions(eval_face_names, eval_face_embeddings)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Optimal threshold0.45\n",
      "F1-Score[0.22915977900689699, 0.4644423827775967, 0.76111402133392436, 0.86570473237732892, 0.84640663889841616, 0.83714940809472893, 0.83714940809472893, 0.83714940809472893]\n"
     ]
    }
   ],
   "source": [
    "opt_tau, f1, accuracy, labeled_name, predicted_name, all_distances, all_indices_th = \\\n",
    "    get_performance(1, np.arange(0.3, 0.7, 0.05), train_face_names, train_face_embeddings, eval_face_names_unk, eval_face_embeddings, \"unknown\")\n",
    "\n",
    "print(\"Optimal threshold\" + str(opt_tau))\n",
    "print(\"F1-Score\" + str(f1))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Optimal threshold0.5\n",
      "F1-Score[0.20121316946705486, 0.39385551624203891, 0.67722742966563776, 0.83064487966296818, 0.8549242176250158, 0.83737434684093648, 0.83737434684093648, 0.83737434684093648]\n"
     ]
    }
   ],
   "source": [
    "opt_tau, f1, accuracy, labeled_name, predicted_name, all_distances, all_indices_th = \\\n",
    "    get_performance(3, np.arange(0.3, 0.7, 0.05), train_face_names, train_face_embeddings, eval_face_names_unk, eval_face_embeddings, \"unknown\")\n",
    "\n",
    "print(\"Optimal threshold\" + str(opt_tau))\n",
    "print(\"F1-Score\" + str(f1))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Optimal threshold0.5\n",
      "F1-Score[0.19228412677697343, 0.37365700163263621, 0.62033311604462393, 0.81168245400001304, 0.85837638552449036, 0.84027873038082335, 0.84182971705521792, 0.84182971705521792]\n"
     ]
    }
   ],
   "source": [
    "opt_tau, f1, accuracy, labeled_name, predicted_name, all_distances, all_indices_th = \\\n",
    "    get_performance(5, np.arange(0.3, 0.7, 0.05), train_face_names, train_face_embeddings, eval_face_names_unk, eval_face_embeddings, \"unknown\")\n",
    "\n",
    "print(\"Optimal threshold\" + str(opt_tau))\n",
    "print(\"F1-Score\" + str(f1))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "unrecognized_errors = []\n",
    "misclassified_errors = []\n",
    "false_positive_errors = []\n",
    "\n",
    "for test_idx, (tag, pred, distances, indexes) in enumerate(zip(labeled_name, predicted_name, all_distances, all_indices_th)):\n",
    "    if tag != pred:\n",
    "        if tag == \"unknown\":\n",
    "            false_positive_errors.append((test_idx, distances, indexes))\n",
    "        if tag != \"unknown\" and pred != \"unknown\":\n",
    "            misclassified_errors.append((test_idx, distances, indexes))\n",
    "        if pred == \"unknown\":\n",
    "            unrecognized_errors.append((test_idx, distances, indexes))\n",
    "\n",
    "print(\"-----------------------------------------------------------\")\n",
    "print(\" => Unrecognized errors\")\n",
    "print(\"-----------------------------------------------------------\")\n",
    "\n",
    "for unrecognized_error in unrecognized_errors:\n",
    "    test_idx, distances, indexes = unrecognized_error\n",
    "    plt.imshow(plt.imread(local_path(eval_img_paths[test_idx])))\n",
    "    plt.title(\"Idx: \" + str(test_idx) + \" \" + eval_img_paths[test_idx].replace(tais_path_prefix,\"\"))\n",
    "    plt.xlabel(eval_img_sizes[test_idx])\n",
    "    plt.show()\n",
    "\n",
    "print(\"-----------------------------------------------------------\")\n",
    "print(\" => Misclassified errors\")\n",
    "print(\"-----------------------------------------------------------\")\n",
    "\n",
    "for misclassified_error in misclassified_errors:\n",
    "    test_idx, distances, indexes = misclassified_error\n",
    "    f, ax = plt.subplots(1, 2, figsize=(10,6)) \n",
    "    ax[0].imshow( plt.imread(local_path(eval_img_paths[test_idx])) )\n",
    "    ax[0].set_xlabel(eval_img_sizes[test_idx])\n",
    "    ax[0].set_title(\"Idx: \" + str(test_idx))\n",
    "    ax[1].imshow( plt.imread(local_path(train_img_paths[indexes[0]])) )\n",
    "    ax[1].set_xlabel(train_img_sizes[indexes[0]])\n",
    "    ax[1].set_title(\"Idx: \" + str(indexes[0]))\n",
    "    plt.show()\n",
    "\n",
    "print(\"-----------------------------------------------------------\")\n",
    "print(\" => False positive errors\")\n",
    "print(\"-----------------------------------------------------------\")\n",
    "\n",
    "for false_positive_error in false_positive_errors:\n",
    "    test_idx, distances, indexes = false_positive_error\n",
    "    f, ax = plt.subplots(1, 2, figsize=(10,6)) \n",
    "    ax[0].imshow( plt.imread(local_path(eval_img_paths[test_idx])) )\n",
    "    ax[0].set_xlabel(eval_img_sizes[test_idx])\n",
    "    ax[0].set_title(\"Idx: \" + str(test_idx))\n",
    "    ax[1].imshow( plt.imread(local_path(train_img_paths[indexes[0]])) ) \n",
    "    ax[1].set_xlabel(train_img_sizes[indexes[0]])\n",
    "    ax[1].set_title(\"Idx: \" + str(indexes[0]))\n",
    "    plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Resize small faces"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "def resize_small_faces(names, embedings, paths, sizes, min_size=110, to_size=200, model=\"hog\"):\n",
    "    out_names = []\n",
    "    out_embedings = []\n",
    "    out_paths = []\n",
    "    out_sizes = []\n",
    "    \n",
    "    resized = 0\n",
    "    \n",
    "    for i, (name, emb, path, size) in enumerate(zip(names, embedings, paths, sizes)):\n",
    "        w, h = size\n",
    "        if w < min_size:\n",
    "            resized +=1\n",
    "            img = cv2.imread(local_path(path))\n",
    "            ih, iw, _ = img.shape\n",
    "            r = to_size/w\n",
    "            img = cv2.resize(img,(int(r*iw), int(r*ih)))\n",
    "            img_rgb = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)\n",
    "            \n",
    "            try:\n",
    "                l = face_recognition.face_locations(img_rgb, number_of_times_to_upsample=0, model=model)\n",
    "                emb = face_recognition.face_encodings(img_rgb, l)[0]\n",
    "            \n",
    "                out_embedings.append(emb)\n",
    "                out_sizes.append((int(r*w), int(r*h)))\n",
    "                out_names.append(name)\n",
    "                out_paths.append(path)\n",
    "            except:\n",
    "                print(\"Error getting embedding\")\n",
    "            \n",
    "        else:\n",
    "            out_names.append(name)\n",
    "            out_embedings.append(emb)\n",
    "            out_paths.append(path)\n",
    "            out_sizes.append(sizes)\n",
    "\n",
    "    print(\"Resized images: \" + str(resized))\n",
    "    return out_names, out_embedings, out_paths, out_sizes"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "resized_train_face_names, resized_train_face_embeddings, resized_train_img_paths, resized_train_img_sizes = \\\n",
    "    resize_small_faces(train_face_names, train_face_embeddings, train_img_paths, train_img_sizes, 110)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "show_distances_distributions(resized_train_face_names, resized_train_face_embeddings)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "resized_eval_face_names, resized_eval_face_embeddings, resized_eval_img_paths, resized_eval_img_sizes = \\\n",
    "    resize_small_faces(eval_face_names, eval_face_embeddings, eval_img_paths, eval_img_sizes, 110)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "show_distances_distributions(resized_eval_face_names, resized_eval_face_embeddings)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "unknown_people = list(set(resized_eval_face_names) - set(train_face_names))\n",
    "resized_eval_face_names_unk = [\"unknown\" if n in unknown_people else n for n in resized_eval_face_names]\n",
    "set(resized_eval_face_names_unk)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "opt_tau, f1, accuracy, labeled_name, predicted_name, all_distances, all_indices_th = \\\n",
    "    get_performance(1, np.arange(0.3, 0.7, 0.05), resized_train_face_names, resized_train_face_embeddings, resized_eval_face_names_unk, resized_eval_face_embeddings, \"unknown\")\n",
    "\n",
    "print(\"Optimal threshold\" + str(opt_tau))\n",
    "print(\"F1-Score\" + str(f1))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "scrolled": false
   },
   "outputs": [],
   "source": [
    "unrecognized_errors = []\n",
    "misclassified_errors = []\n",
    "false_positive_errors = []\n",
    "\n",
    "for test_idx, (tag, pred, distances, indexes) in enumerate(zip(labeled_name, predicted_name, all_distances, all_indices_th)):\n",
    "    if tag != pred:\n",
    "        if tag == \"unknown\":\n",
    "            false_positive_errors.append((test_idx, distances, indexes))\n",
    "        if tag != \"unknown\" and pred != \"unknown\":\n",
    "            misclassified_errors.append((test_idx, distances, indexes))\n",
    "        if pred == \"unknown\":\n",
    "            unrecognized_errors.append((test_idx, distances, indexes))\n",
    "\n",
    "print(\"-----------------------------------------------------------\")\n",
    "print(\" => Unrecognized errors\")\n",
    "print(\"-----------------------------------------------------------\")\n",
    "\n",
    "for unrecognized_error in unrecognized_errors:\n",
    "    test_idx, distances, indexes = unrecognized_error\n",
    "    plt.imshow(plt.imread(local_path(eval_img_paths[test_idx])))\n",
    "    plt.title(\"Idx: \" + str(test_idx) + \" \" + eval_img_paths[test_idx].replace(tais_path_prefix,\"\"))\n",
    "    plt.xlabel(eval_img_sizes[test_idx])\n",
    "    plt.show()\n",
    "\n",
    "print(\"-----------------------------------------------------------\")\n",
    "print(\" => Misclassified errors\")\n",
    "print(\"-----------------------------------------------------------\")\n",
    "\n",
    "for misclassified_error in misclassified_errors:\n",
    "    test_idx, distances, indexes = misclassified_error\n",
    "    f, ax = plt.subplots(1, 2, figsize=(10,6)) \n",
    "    ax[0].imshow( plt.imread(local_path(eval_img_paths[test_idx])) )\n",
    "    ax[0].set_xlabel(eval_img_sizes[test_idx])\n",
    "    ax[0].set_title(\"Idx: \" + str(test_idx))\n",
    "    ax[1].imshow( plt.imread(local_path(train_img_paths[indexes[0]])) )\n",
    "    ax[1].set_xlabel(train_img_sizes[indexes[0]])\n",
    "    ax[1].set_title(\"Idx: \" + str(indexes[0]))\n",
    "    plt.show()\n",
    "\n",
    "print(\"-----------------------------------------------------------\")\n",
    "print(\" => False positive errors\")\n",
    "print(\"-----------------------------------------------------------\")\n",
    "\n",
    "for false_positive_error in false_positive_errors:\n",
    "    test_idx, distances, indexes = false_positive_error\n",
    "    f, ax = plt.subplots(1, 2, figsize=(10,6)) \n",
    "    ax[0].imshow( plt.imread(local_path(eval_img_paths[test_idx])) )\n",
    "    ax[0].set_xlabel(eval_img_sizes[test_idx])\n",
    "    ax[0].set_title(\"Idx: \" + str(test_idx))\n",
    "    ax[1].imshow( plt.imread(local_path(train_img_paths[indexes[0]])) ) \n",
    "    ax[1].set_xlabel(train_img_sizes[indexes[0]])\n",
    "    ax[1].set_title(\"Idx: \" + str(indexes[0]))\n",
    "    plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Faces landmarks analysis"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "plt.imshow(plt.imread(local_path(eval_img_paths[925])))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "plt.imshow(plt.imread(local_path(eval_img_paths[921])))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "img_frontal = plt.imread(local_path(eval_img_paths[921]))\n",
    "face_landmarks_frontal = face_recognition.face_landmarks(img_frontal)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "face_landmarks_frontal"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "img_nofrontal = plt.imread(local_path(eval_img_paths[925]))\n",
    "face_landmarks_nofrontal = face_recognition.face_landmarks(img_nofrontal)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "scrolled": false
   },
   "outputs": [],
   "source": [
    "face_landmarks_nofrontal"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from PIL import Image, ImageDraw\n",
    "import face_recognition\n",
    "\n",
    "for face_landmarks in face_landmarks_frontal:\n",
    "    pil_image = Image.fromarray(img_frontal)\n",
    "    d = ImageDraw.Draw(pil_image, 'RGBA')\n",
    "\n",
    "    # Make the eyebrows into a nightmare\n",
    "    d.polygon(face_landmarks['left_eyebrow'], fill=(68, 54, 39, 128))\n",
    "    d.polygon(face_landmarks['right_eyebrow'], fill=(68, 54, 39, 128))\n",
    "    d.line(face_landmarks['left_eyebrow'], fill=(68, 54, 39, 150), width=5)\n",
    "    d.line(face_landmarks['right_eyebrow'], fill=(68, 54, 39, 150), width=5)\n",
    "\n",
    "    # Gloss the lips\n",
    "    d.polygon(face_landmarks['top_lip'], fill=(150, 0, 0, 128))\n",
    "    d.polygon(face_landmarks['bottom_lip'], fill=(150, 0, 0, 128))\n",
    "    d.line(face_landmarks['top_lip'], fill=(150, 0, 0, 64), width=8)\n",
    "    d.line(face_landmarks['bottom_lip'], fill=(150, 0, 0, 64), width=8)\n",
    "\n",
    "    # Sparkle the eyes\n",
    "    d.polygon(face_landmarks['left_eye'], fill=(255, 255, 255, 30))\n",
    "    d.polygon(face_landmarks['right_eye'], fill=(255, 255, 255, 30))\n",
    "\n",
    "    # Apply some eyeliner\n",
    "    d.line(face_landmarks['left_eye'] + [face_landmarks['left_eye'][0]], fill=(0, 0, 0, 110), width=6)\n",
    "    d.line(face_landmarks['right_eye'] + [face_landmarks['right_eye'][0]], fill=(0, 0, 0, 110), width=6)\n",
    "\n",
    "pix = np.array(pil_image)\n",
    "plt.imshow(pix)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from PIL import Image, ImageDraw\n",
    "import face_recognition\n",
    "\n",
    "for face_landmarks in face_landmarks_nofrontal:\n",
    "    pil_image = Image.fromarray(img_nofrontal)\n",
    "    d = ImageDraw.Draw(pil_image, 'RGBA')\n",
    "\n",
    "    # Make the eyebrows into a nightmare\n",
    "    d.polygon(face_landmarks['left_eyebrow'], fill=(68, 54, 39, 128))\n",
    "    d.polygon(face_landmarks['right_eyebrow'], fill=(68, 54, 39, 128))\n",
    "    d.line(face_landmarks['left_eyebrow'], fill=(68, 54, 39, 150), width=5)\n",
    "    d.line(face_landmarks['right_eyebrow'], fill=(68, 54, 39, 150), width=5)\n",
    "\n",
    "    # Gloss the lips\n",
    "    d.polygon(face_landmarks['top_lip'], fill=(150, 0, 0, 128))\n",
    "    d.polygon(face_landmarks['bottom_lip'], fill=(150, 0, 0, 128))\n",
    "    d.line(face_landmarks['top_lip'], fill=(150, 0, 0, 64), width=8)\n",
    "    d.line(face_landmarks['bottom_lip'], fill=(150, 0, 0, 64), width=8)\n",
    "\n",
    "    # Sparkle the eyes\n",
    "    d.polygon(face_landmarks['left_eye'], fill=(255, 255, 255, 30))\n",
    "    d.polygon(face_landmarks['right_eye'], fill=(255, 255, 255, 30))\n",
    "\n",
    "    # Apply some eyeliner\n",
    "    d.line(face_landmarks['left_eye'] + [face_landmarks['left_eye'][0]], fill=(0, 0, 0, 110), width=6)\n",
    "    d.line(face_landmarks['right_eye'] + [face_landmarks['right_eye'][0]], fill=(0, 0, 0, 110), width=6)\n",
    "\n",
    "pix = np.array(pil_image)\n",
    "plt.imshow(pix)\n",
    "plt.show()"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3.0
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.5.0+"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 0
}