notebooks/Create Cropped Faces.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "This code ingests:\n",
    "\n",
    "- A folder of `.json` files generated by OpenFace.\n",
    "- A folder of images in `Teenie_Harris_PNG1600`\n",
    "\n",
    "And outputs:\n",
    "\n",
    "- `images.npy` the cropped imags\n",
    "- `indices.npy` the index of face within the photo\n",
    "- `descriptors.npy` the OpenFace descriptor for the face\n",
    "- `filenames.csv` the filename the face was taken from\n",
    "\n",
    "Each file has the same number of rows."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "import re\n",
    "import json\n",
    "import matplotlib.pyplot as plt\n",
    "from utils.imutil import *\n",
    "from utils.list_all_files import *\n",
    "from utils.crop import *\n",
    "from utils.progress import *\n",
    "from utils.mosaic import *\n",
    "from utils.draw_shapes import *"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "input_dir = '../data/openface/json/'\n",
    "output_dir = '../data/openface/npy32/'\n",
    "output_side = 32\n",
    "output_dtype = np.uint8 # the png1600 images are uint16 for some reason"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "59278 0:03:25 288.77/s\n"
     ]
    }
   ],
   "source": [
    "def natural_sort_key(s, _nsre=re.compile('([0-9]+)')):\n",
    "    return [int(text) if text.isdigit() else text.lower()\n",
    "            for text in _nsre.split(s)]\n",
    "\n",
    "def get_id(fn):\n",
    "    return os.path.splitext(os.path.basename(fn))[0]\n",
    "\n",
    "tasks = []\n",
    "for input_fn in list_all_files(input_dir):\n",
    "    tasks.append(input_fn)\n",
    "\n",
    "tasks.sort(key=lambda x: natural_sort_key(get_id(x)))\n",
    "\n",
    "def job(task):\n",
    "    try:\n",
    "        metadata = json.load(open(task))\n",
    "        out = []\n",
    "        img = imread(fn)\n",
    "        for i, face in enumerate(metadata['faces']):\n",
    "                w,n,e,s = face['box']\n",
    "                # force it square\n",
    "                rows = s - n\n",
    "                cols = e - w\n",
    "                side = min(rows, cols)\n",
    "                s = n + side\n",
    "                e = w + side\n",
    "                face_img = safe_crop(img, n, s, w, e, fill=0)\n",
    "                face_img = imresize(face_img, max_side=output_side)\n",
    "                if output_dtype is not None:\n",
    "                    face_img = face_img.astype(output_dtype)\n",
    "                face_rep = np.asarray(face['rep'])\n",
    "                out.append((task, i, face_img, face_rep))\n",
    "        return out\n",
    "    except:\n",
    "        print(task, metadata['path'], face['box'])\n",
    "        raise\n",
    "\n",
    "results = progress_parallel(job, tasks)\n",
    "results = [item for sublist in results for item in sublist] # flatten results"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "input_filenames, face_indices, face_images, face_descriptors = list(zip(*results))\n",
    "face_images = np.asarray(face_images)\n",
    "face_descriptors = np.asarray(face_descriptors)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "os.makedirs(output_dir, exist_ok=True)\n",
    "np.save(os.path.join(output_dir, 'images.npy'), face_images)\n",
    "np.save(os.path.join(output_dir, 'indices.npy'), face_indices)\n",
    "np.save(os.path.join(output_dir, 'descriptors.npy'), face_descriptors)\n",
    "np.savetxt(os.path.join(output_dir, 'filenames.csv'), input_filenames, fmt='%s')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "# imshow(make_mosaic(face_images[:495*495]), fmt='jpg')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYcAAAD8CAYAAACcjGjIAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADl0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uIDIuMi4yLCBodHRwOi8vbWF0cGxvdGxpYi5vcmcvhp/UCwAAFxlJREFUeJzt3X2MXfV95/H3Z81D2DzZwCSitlmTxu2GRBtDRuCI1So1CRhS1VQiWqMqWJFX7kZmlWijbUxXWpoHJCJtQxcppesWF1Nl47AkWSxw6rpAVEUKYBMcwBDkCXjD1BQ7ayDJRkvW9Lt/3J/pXZ9rz3hm8J0p75d0dM/5nt8593uSwZ85D/dOqgpJkvr9k2E3IEmafQwHSVKH4SBJ6jAcJEkdhoMkqcNwkCR1TDocksxL8miSe9ryeUkeSrI3ydeTnNbqp7flsbZ+Sd8+rm/1p5Nc3ldf2WpjSTbM3OFJkqbiRM4cPgU81bf8JeDmqloKvAisbfW1wItV9W7g5jaOJOcDq4H3AiuBP26BMw/4CnAFcD5wTRsrSRqSSYVDkkXAR4E/a8sBVgB3tSGbgava/Kq2TFt/aRu/CthSVa9U1bPAGHBRm8aq6pmq+iWwpY2VJA3JKZMc90fA7wFvbctnAS9V1eG2PA4sbPMLgecAqupwkpfb+IXAg3377N/muaPqF0/U0Nlnn11LliyZZPuSJIBHHnnkJ1U1MtG4CcMhyW8CB6rqkSQfOlIeMLQmWHes+qCzl4Hf6ZFkHbAO4Nxzz2XXrl3H6VySdLQk/3My4yZzWekS4LeS7KN3yWcFvTOJ+UmOhMsiYH+bHwcWtyZOAd4OHOqvH7XNseodVbWxqkaranRkZMLgkyRN0YThUFXXV9WiqlpC74by/VX1O8ADwNVt2Brg7ja/tS3T1t9fvW/32wqsbk8znQcsBR4GdgJL29NPp7X32DojRydJmpLJ3nMY5LPAliRfBB4Fbmv124C/SDJG74xhNUBV7UlyJ/AkcBhYX1WvAiS5DtgOzAM2VdWeafQlSZqmzNWv7B4dHS3vOUjSiUnySFWNTjTOT0hLkjoMB0lSh+EgSeowHCRJHYaDJKljOo+yvmEs2XDva/P7bvroEDuRpJPDMwdJUofhIEnqMBwkSR2GgySpw3CQJHUYDpKkDsNBktRhOEiSOgwHSVKHn5Du4yehJanHMwdJUofhIEnqmDAckrwpycNJfpBkT5LPtfrtSZ5NsrtNy1o9SW5JMpbksSQX9u1rTZK9bVrTV/9AksfbNrckyetxsJKkyZnMPYdXgBVV9fMkpwLfTfLttu4/VNVdR42/AljapouBW4GLk5wJ3ACMAgU8kmRrVb3YxqwDHgS2ASuBbyNJGooJzxyq5+dt8dQ21XE2WQXc0bZ7EJif5BzgcmBHVR1qgbADWNnWva2qvldVBdwBXDWNY5IkTdOk7jkkmZdkN3CA3j/wD7VVN7ZLRzcnOb3VFgLP9W0+3mrHq48PqEuShmRSj7JW1avAsiTzgW8leR9wPfB3wGnARuCzwOeBQfcLagr1jiTr6F1+4txzz51M668rH32V9I/VCT2tVFUvAd8BVlbV8+3S0SvAnwMXtWHjwOK+zRYB+yeoLxpQH/T+G6tqtKpGR0ZGTqR1SdIJmMzTSiPtjIEkZwAfBn7Y7hXQniy6CniibbIVuLY9tbQceLmqnge2A5clWZBkAXAZsL2t+1mS5W1f1wJ3z+xhSpJOxGQuK50DbE4yj16Y3FlV9yS5P8kIvctCu4F/28ZvA64ExoBfAJ8AqKpDSb4A7GzjPl9Vh9r8J4HbgTPoPaX0uj6p5OUgSTq+CcOhqh4DLhhQX3GM8QWsP8a6TcCmAfVdwPsm6kWSdHL4CWlJUofhIEnqMBwkSR2GgySpw3CQJHUYDpKkDsNBktRhOEiSOgwHSVKH4SBJ6jAcJEkdhoMkqcNwkCR1GA6SpA7DQZLUYThIkjoMB0lSh+EgSeowHCRJHROGQ5I3JXk4yQ+S7EnyuVY/L8lDSfYm+XqS01r99LY81tYv6dvX9a3+dJLL++orW20syYaZP0xJ0ok4ZRJjXgFWVNXPk5wKfDfJt4F/D9xcVVuS/AmwFri1vb5YVe9Oshr4EvCvk5wPrAbeC/wK8NdJfq29x1eAjwDjwM4kW6vqyRk8zmNasuHek/E2kjSnTHjmUD0/b4untqmAFcBdrb4ZuKrNr2rLtPWXJkmrb6mqV6rqWWAMuKhNY1X1TFX9EtjSxkqShmRS9xySzEuyGzgA7AB+BLxUVYfbkHFgYZtfCDwH0Na/DJzVXz9qm2PVB/WxLsmuJLsOHjw4mdYlSVMwqXCoqlerahmwiN5v+u8ZNKy95hjrTrQ+qI+NVTVaVaMjIyMTNy5JmpITelqpql4CvgMsB+YnOXLPYhGwv82PA4sB2vq3A4f660dtc6y6JGlIJvO00kiS+W3+DODDwFPAA8DVbdga4O42v7Ut09bfX1XV6qvb00znAUuBh4GdwNL29NNp9G5ab52Jg5MkTc1knlY6B9icZB69MLmzqu5J8iSwJckXgUeB29r424C/SDJG74xhNUBV7UlyJ/AkcBhYX1WvAiS5DtgOzAM2VdWeGTtCSdIJmzAcquox4IIB9Wfo3X84uv5/gI8dY183AjcOqG8Dtk2iX0nSSeAnpCVJHYaDJKnDcJAkdRgOkqQOw0GS1GE4SJI6DAdJUofhIEnqMBwkSR2GgySpw3CQJHUYDpKkDsNBktRhOEiSOgwHSVKH4SBJ6jAcJEkdhoMkqWPCcEiyOMkDSZ5KsifJp1r9D5L8bZLdbbqyb5vrk4wleTrJ5X31la02lmRDX/28JA8l2Zvk60lOm+kDlSRN3oR/Qxo4DHymqr6f5K3AI0l2tHU3V9V/7h+c5HxgNfBe4FeAv07ya231V4CPAOPAziRbq+pJ4EttX1uS/AmwFrh1ugc3LEs23Pva/L6bPjrETiRpaiY8c6iq56vq+23+Z8BTwMLjbLIK2FJVr1TVs8AYcFGbxqrqmar6JbAFWJUkwArgrrb9ZuCqqR6QJGn6TuieQ5IlwAXAQ610XZLHkmxKsqDVFgLP9W023mrHqp8FvFRVh4+qD3r/dUl2Jdl18ODBE2ldknQCJh0OSd4CfAP4dFX9lN5ln18FlgHPA394ZOiAzWsK9W6xamNVjVbV6MjIyGRblySdoMnccyDJqfSC4atV9U2Aqnqhb/2fAve0xXFgcd/mi4D9bX5Q/SfA/CSntLOH/vGSpCGYzNNKAW4DnqqqL/fVz+kb9tvAE21+K7A6yelJzgOWAg8DO4Gl7cmk0+jdtN5aVQU8AFzdtl8D3D29w5IkTcdkzhwuAT4OPJ5kd6v9PnBNkmX0LgHtA34XoKr2JLkTeJLek07rq+pVgCTXAduBecCmqtrT9vdZYEuSLwKP0gsjSdKQTBgOVfVdBt8X2HacbW4EbhxQ3zZou6p6ht7TTJKkWcBPSEuSOgwHSVKH4SBJ6jAcJEkdhoMkqcNwkCR1GA6SpA7DQZLUYThIkjoMB0lSh+EgSeowHCRJHYaDJKnDcJAkdRgOkqQOw0GS1GE4SJI6DAdJUseE4ZBkcZIHkjyVZE+ST7X6mUl2JNnbXhe0epLckmQsyWNJLuzb15o2fm+SNX31DyR5vG1zS5JBf5ZUknSSTObM4TDwmap6D7AcWJ/kfGADcF9VLQXua8sAVwBL27QOuBV6YQLcAFxM7+9F33AkUNqYdX3brZz+oUmSpmrCcKiq56vq+23+Z8BTwEJgFbC5DdsMXNXmVwF3VM+DwPwk5wCXAzuq6lBVvQjsAFa2dW+rqu9VVQF39O1LkjQEJ3TPIckS4ALgIeCdVfU89AIEeEcbthB4rm+z8VY7Xn18QH3Q+69LsivJroMHD55I65KkEzDpcEjyFuAbwKer6qfHGzqgVlOod4tVG6tqtKpGR0ZGJmpZkjRFkwqHJKfSC4avVtU3W/mFdkmI9nqg1ceBxX2bLwL2T1BfNKAuSRqSyTytFOA24Kmq+nLfqq3AkSeO1gB399WvbU8tLQdebpedtgOXJVnQbkRfBmxv636WZHl7r2v79iVJGoJTJjHmEuDjwONJdrfa7wM3AXcmWQv8GPhYW7cNuBIYA34BfAKgqg4l+QKws437fFUdavOfBG4HzgC+3SZJ0pBMGA5V9V0G3xcAuHTA+ALWH2Nfm4BNA+q7gPdN1Isk6eTwE9KSpA7DQZLUYThIkjoMB0lSh+EgSeowHCRJHYaDJKnDcJAkdRgOkqQOw0GS1GE4SJI6DAdJUsdkvpVVr4MlG+59bX7fTR8dYieS1OWZgySpw3CQJHUYDpKkDsNBktRhOEiSOiYMhySbkhxI8kRf7Q+S/G2S3W26sm/d9UnGkjyd5PK++spWG0uyoa9+XpKHkuxN8vUkp83kAUqSTtxkzhxuB1YOqN9cVcvatA0gyfnAauC9bZs/TjIvyTzgK8AVwPnANW0swJfavpYCLwJrp3NAkqTpmzAcqupvgEOT3N8qYEtVvVJVzwJjwEVtGquqZ6rql8AWYFWSACuAu9r2m4GrTvAYJEkzbDr3HK5L8li77LSg1RYCz/WNGW+1Y9XPAl6qqsNH1SVJQzTVcLgV+FVgGfA88IetngFjawr1gZKsS7Irya6DBw+eWMeSpEmbUjhU1QtV9WpV/T3wp/QuG0HvN//FfUMXAfuPU/8JMD/JKUfVj/W+G6tqtKpGR0ZGptK6JGkSphQOSc7pW/xt4MiTTFuB1UlOT3IesBR4GNgJLG1PJp1G76b11qoq4AHg6rb9GuDuqfQkSZo5E37xXpKvAR8Czk4yDtwAfCjJMnqXgPYBvwtQVXuS3Ak8CRwG1lfVq20/1wHbgXnApqra097is8CWJF8EHgVum7GjkyRNyYThUFXXDCgf8x/wqroRuHFAfRuwbUD9Gf7hspQkaRbwE9KSpA7DQZLUYThIkjoMB0lSh+EgSeowHCRJHYaDJKnDcJAkdRgOkqQOw0GS1GE4SJI6DAdJUofhIEnqMBwkSR2GgySpw3CQJHUYDpKkDsNBktQxYTgk2ZTkQJIn+mpnJtmRZG97XdDqSXJLkrEkjyW5sG+bNW383iRr+uofSPJ42+aWJJnpg5QknZjJnDncDqw8qrYBuK+qlgL3tWWAK4ClbVoH3Aq9MAFuAC6m9/eibzgSKG3Mur7tjn6vN7QlG+59bZKkk2XCcKiqvwEOHVVeBWxu85uBq/rqd1TPg8D8JOcAlwM7qupQVb0I7ABWtnVvq6rvVVUBd/TtS5I0JFO95/DOqnoeoL2+o9UXAs/1jRtvtePVxwfUJUlDNNM3pAfdL6gp1AfvPFmXZFeSXQcPHpxii5KkiUw1HF5ol4RorwdafRxY3DduEbB/gvqiAfWBqmpjVY1W1ejIyMgUW5ckTWSq4bAVOPLE0Rrg7r76te2ppeXAy+2y03bgsiQL2o3oy4Dtbd3PkixvTyld27cvSdKQnDLRgCRfAz4EnJ1knN5TRzcBdyZZC/wY+Fgbvg24EhgDfgF8AqCqDiX5ArCzjft8VR25yf1Jek9EnQF8u02SpCGaMByq6ppjrLp0wNgC1h9jP5uATQPqu4D3TdSHJOnk8RPSkqQOw0GS1GE4SJI6DAdJUofhIEnqMBwkSR2GgySpw3CQJHUYDpKkDsNBktRhOEiSOgwHSVKH4SBJ6jAcJEkdE35lt+aWJRvufW1+300fHWInkuYyzxwkSR2GgySpw3CQJHVMKxyS7EvyeJLdSXa12plJdiTZ214XtHqS3JJkLMljSS7s28+aNn5vkjXTOyRJ0nTNxJnDb1TVsqoabcsbgPuqailwX1sGuAJY2qZ1wK3QCxPgBuBi4CLghiOBIkkajtfjstIqYHOb3wxc1Ve/o3oeBOYnOQe4HNhRVYeq6kVgB7DydehLkjRJ0w2HAv4qySNJ1rXaO6vqeYD2+o5WXwg817fteKsdqy5JGpLpfs7hkqran+QdwI4kPzzO2Ayo1XHq3R30AmgdwLnnnnuivUqSJmlaZw5Vtb+9HgC+Re+ewQvtchHt9UAbPg4s7tt8EbD/OPVB77exqkaranRkZGQ6rUuSjmPK4ZDkzUneemQeuAx4AtgKHHniaA1wd5vfClzbnlpaDrzcLjttBy5LsqDdiL6s1SRJQzKdy0rvBL6V5Mh+/ltV/WWSncCdSdYCPwY+1sZvA64ExoBfAJ8AqKpDSb4A7GzjPl9Vh6bRlyRpmqYcDlX1DPD+AfX/BVw6oF7A+mPsaxOwaaq9aGb4vUySjvAT0pKkDsNBktRhOEiSOgwHSVKHf+xHM84b29Lc55mDJKnDcJAkdXhZSXNC/6Uq8HKV9HrzzEGS1GE4SJI6DAdJUofhIEnqMBwkSR0+rSQN4Af59EbnmYMkqcNwkCR1GA6SpA7DQZLUMWvCIcnKJE8nGUuyYdj9SNIb2ax4WinJPOArwEeAcWBnkq1V9eRwO5NOLp+S0mwxW84cLgLGquqZqvolsAVYNeSeJOkNa1acOQALgef6lseBi4fUi6Q+ns28MaWqht0DST4GXF5V/6Ytfxy4qKr+3VHj1gHr2uKvA0+f1EZnxtnAT4bdxDTY//DN9WOw/+H6Z1U1MtGg2XLmMA4s7lteBOw/elBVbQQ2nqymXg9JdlXV6LD7mCr7H765fgz2PzfMlnsOO4GlSc5LchqwGtg65J4k6Q1rVpw5VNXhJNcB24F5wKaq2jPktiTpDWtWhANAVW0Dtg27j5NgTl8Ww/5ng7l+DPY/B8yKG9KSpNllttxzkCTNIobDDEuyKcmBJE/01c5MsiPJ3va6oNWT5Jb2lSGPJblweJ1DksVJHkjyVJI9ST41l/pvPb0pycNJftCO4XOtfl6Sh9oxfL09+ECS09vyWFu/ZJj9H5FkXpJHk9zTludM/0n2JXk8ye4ku1ptzvwMASSZn+SuJD9s/z18cK4dw3QZDjPvdmDlUbUNwH1VtRS4ry0DXAEsbdM64NaT1OOxHAY+U1XvAZYD65Ocz9zpH+AVYEVVvR9YBqxMshz4EnBzO4YXgbVt/Frgxap6N3BzGzcbfAp4qm95rvX/G1W1rO+Rz7n0MwTwX4C/rKp/Dryf3v8Xc+0YpqeqnGZ4ApYAT/QtPw2c0+bPAZ5u8/8VuGbQuNkwAXfT+76rudr/PwW+T+/T9j8BTmn1DwLb2/x24INt/pQ2LkPuexG9f3xWAPcAmWP97wPOPqo2Z36GgLcBzx79v+NcOoaZmDxzODneWVXPA7TXd7T6oK8NWXiSexuoXZ64AHiIOdZ/uySzGzgA7AB+BLxUVYfbkP4+XzuGtv5l4KyT23HHHwG/B/x9Wz6LudV/AX+V5JH2rQYwt36G3gUcBP68Xdr7syRvZm4dw7QZDsOVAbWhPz6W5C3AN4BPV9VPjzd0QG3o/VfVq1W1jN5v4BcB7xk0rL3OqmNI8pvAgap6pL88YOis7L+5pKoupHe5ZX2Sf3WcsbOx/1OAC4Fbq+oC4H/zD5eQBpmNxzBthsPJ8UKScwDa64FWn9TXhpxMSU6lFwxfrapvtvKc6b9fVb0EfIfe/ZP5SY58rqe/z9eOoa1/O3Do5Hb6/7kE+K0k++h9O/EKemcSc6V/qmp/ez0AfIteQM+ln6FxYLyqHmrLd9ELi7l0DNNmOJwcW4E1bX4NvWv5R+rXtqcdlgMvHzltHYYkAW4DnqqqL/etmhP9AyQZSTK/zZ8BfJjezcQHgKvbsKOP4cixXQ3cX+3C8TBU1fVVtaiqltD7Gpn7q+p3mCP9J3lzkrcemQcuA55gDv0MVdXfAc8l+fVWuhR4kjl0DDNi2Dc9/rFNwNeA54H/S+83irX0rgHfB+xtr2e2saH3R45+BDwOjA65939J73T4MWB3m66cK/23nv4F8Gg7hieA/9Tq7wIeBsaA/w6c3upvastjbf27hn0MfcfyIeCeudR/6/MHbdoD/MdWnzM/Q62vZcCu9nP0P4AFc+0Ypjv5CWlJUoeXlSRJHYaDJKnDcJAkdRgOkqQOw0GS1GE4SJI6DAdJUofhIEnq+H9CdgoaAJxQXQAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.88043982329\n"
     ]
    }
   ],
   "source": [
    "plt.hist(results, bins=100)\n",
    "plt.show()\n",
    "results.sort()\n",
    "\n",
    "# plt.plot(results)\n",
    "# plt.yscale('log')\n",
    "# plt.show()\n",
    "\n",
    "# 88% of faces are smaller than 128 pixels\n",
    "print(np.sum(np.asarray(results) < output_side) / len(results))"
   ]
  }
 ],
 "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
}