face_recognition_gfp.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "37265ad5",
   "metadata": {},
   "outputs": [],
   "source": [
    "# dependencies\n",
    "import os\n",
    "import cv2\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "from deepface import DeepFace\n",
    "import matplotlib.pyplot as plt\n",
    "import matplotlib\n",
    "from matplotlib.ticker import StrMethodFormatter\n",
    "from sklearn.model_selection import train_test_split"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "2c9d60c8",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Call VGGFace model\n",
    "model_name = \"Facenet\"\n",
    "model = DeepFace.build_model(model_name)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "c3d4d151",
   "metadata": {},
   "outputs": [],
   "source": [
    "######################################################\n",
    "# OWN CLASS TO STORE IMAGE PATHS AND CLASSES\n",
    "######################################################\n",
    "\n",
    "# own class to store path separate for intra vs inter class comparison\n",
    "class facepath: \n",
    "    def __init__(self, path, faceno): \n",
    "        self.path = path \n",
    "        self.faceno = faceno"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "2187b42a",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "loaded 75 files from 15 plain face(s)\n",
      "Example: original_and_generated_faces/original/plain/501181/501181_490210.jpg [Face: 1]\n",
      "----------------------------------\n",
      "loaded 75 files from 15 encrypted layer progression order face(s)\n",
      "Example: original_and_generated_faces/original/l_encrypted/501182/501182_490232_j2k_l-W4-O6.png [Face: 2]\n",
      "----------------------------------\n",
      "loaded 75 files from 15 encrypted resolution progression order face(s)\n",
      "Example: original_and_generated_faces/original/r_encrypted/501183/501183_490247_j2k_r-W4-O6.png [Face: 3]\n",
      "----------------------------------\n",
      "loaded 72 files from 15 AI r_enc_trained_r_gfp_pre_enhanced face(s)\n",
      "Example: original_and_generated_faces/gfpgan/psp_r_enc_trained_r_gfp_pre_enhanced/501185/501185_490303_j2k_r-W4-O6.png [Face: 5]\n",
      "----------------------------------\n",
      "loaded 72 files from 15 AI l_enc_trained_l_gfp_pre_enhanced face(s)\n",
      "Example: original_and_generated_faces/gfpgan/psp_l_enc_trained_l_gfp_pre_enhanced/501186/501186_490329_j2k_l-W4-O6.png [Face: 6]\n",
      "----------------------------------\n",
      "loaded 72 files from 15 AI r_enc_trained_r_gfp_post_enhanced face(s)\n",
      "Example: original_and_generated_faces/gfpgan/psp_r_enc_trained_r_gfp_post_enhanced/501185/501185_490303_j2k_r-W4-O6.png [Face: 5]\n",
      "----------------------------------\n",
      "loaded 72 files from 15 AI l_enc_trained_l_gfp_post_enhanced face(s)\n",
      "Example: original_and_generated_faces/gfpgan/psp_l_enc_trained_l_gfp_post_enhanced/501186/501186_490329_j2k_l-W4-O6.png [Face: 6]\n",
      "----------------------------------\n",
      "loaded 72 files from 15 AI l_enc_trained_l_gfp_pre_post_enhanced face(s)\n",
      "Example: /home/nero/GitHub/encrypted_face_reconstruction/original_and_generated_faces/gfpgan/psp_l_enc_trained_l_gfp_pre_post_enhanced/501186/501186_490329_j2k_l-W4-O6.png [Face: 6]\n",
      "----------------------------------\n",
      "loaded 72 files from 15 AI r_enc_trained_r_gfp_pre_post_enhanced face(s)\n",
      "Example: /home/nero/GitHub/encrypted_face_reconstruction/original_and_generated_faces/gfpgan/psp_r_enc_trained_r_gfp_pre_post_enhanced/501186/501186_490327_j2k_r-W4-O6.png [Face: 6]\n",
      "----------------------------------\n",
      "loaded 72 files from 15 AI l_enc_pretrained face(s)\n",
      "Example: original_and_generated_faces/gfpgan/l_pretrained/501188/501188_490376_j2k_l-W4-O6.png [Face: 8]\n",
      "----------------------------------\n",
      "loaded 72 files from 15 AI r_enc_pretrained face(s)\n",
      "Example: original_and_generated_faces/gfpgan/r_pretrained/501189/501189_490395_j2k_r-W4-O6.png [Face: 9]\n",
      "----------------------------------\n"
     ]
    }
   ],
   "source": [
    "######################################################\n",
    "# DATA LOADING\n",
    "######################################################\n",
    "\n",
    "\n",
    "###############################\n",
    "# 1- PLAIN FACES\n",
    "###############################\n",
    "\n",
    "# how many faces / folders should we use as database\n",
    "n_faces = 15\n",
    "n_max_images_per_face = 5\n",
    "\n",
    "# Load all filenames into a list\n",
    "path = 'original_and_generated_faces/original/plain/' \n",
    "plain_faces = []\n",
    "    \n",
    "i = 0\n",
    "for root, dirs, files in sorted(os.walk(path)):\n",
    "    \n",
    "    if i == n_faces+1:\n",
    "        break\n",
    "    j = 0\n",
    "    for name in files:\n",
    "        \n",
    "        if j == n_max_images_per_face:\n",
    "            break  \n",
    "            \n",
    "        plain_faces.append(facepath(str(root) + \"/\" + (str(name)), i))\n",
    "        j += 1\n",
    "        \n",
    "    i += 1\n",
    "        \n",
    "print(\"loaded %d files from %d plain face(s)\" % (len(plain_faces), n_faces))\n",
    "print(\"Example: %s [Face: %s]\" % (plain_faces[0].path, plain_faces[0].faceno))\n",
    "print(\"----------------------------------\")\n",
    "\n",
    "###############################\n",
    "# 2- ENCRYPTED FACES L\n",
    "###############################\n",
    "\n",
    "# Load all filenames into a list\n",
    "path = 'original_and_generated_faces/original/l_encrypted/' \n",
    "encrypted_l_faces = []\n",
    "    \n",
    "i = 0\n",
    "for root, dirs, files in sorted(os.walk(path)):\n",
    "    \n",
    "    if i == n_faces+1:\n",
    "        break\n",
    "    j = 0\n",
    "    for name in files:\n",
    "        \n",
    "        if j == n_max_images_per_face:\n",
    "            break  \n",
    "            \n",
    "        encrypted_l_faces.append(facepath(str(root) + \"/\" + (str(name)), i))\n",
    "        j += 1\n",
    "        \n",
    "    i += 1\n",
    "        \n",
    "print(\"loaded %d files from %d encrypted layer progression order face(s)\" % (len(encrypted_l_faces), n_faces))\n",
    "print(\"Example: %s [Face: %s]\" % (encrypted_l_faces[5].path, encrypted_l_faces[5].faceno))\n",
    "print(\"----------------------------------\")\n",
    "\n",
    "###############################\n",
    "# 3- ENCRYPTED FACES R\n",
    "###############################\n",
    "\n",
    "# Load all filenames into a list\n",
    "path = 'original_and_generated_faces/original/r_encrypted/' \n",
    "encrypted_r_faces = []\n",
    "    \n",
    "i = 0\n",
    "for root, dirs, files in sorted(os.walk(path)):\n",
    "    \n",
    "    if i == n_faces+1:\n",
    "        break\n",
    "    j = 0\n",
    "    for name in files:\n",
    "        \n",
    "        if j == n_max_images_per_face:\n",
    "            break  \n",
    "            \n",
    "        encrypted_r_faces.append(facepath(str(root) + \"/\" + (str(name)), i))\n",
    "        j += 1\n",
    "        \n",
    "    i += 1\n",
    "        \n",
    "print(\"loaded %d files from %d encrypted resolution progression order face(s)\" % (len(encrypted_r_faces), n_faces))\n",
    "print(\"Example: %s [Face: %s]\" % (encrypted_r_faces[10].path, encrypted_r_faces[10].faceno))\n",
    "print(\"----------------------------------\")\n",
    "\n",
    "\n",
    "#######################################################################\n",
    "# 4 - PSP RECONSTRUCTED FACES R WITH TRAINED R - PRE ENHANCED WITH GFP \n",
    "#######################################################################\n",
    "\n",
    "# Load all filenames into a list\n",
    "path = 'original_and_generated_faces/gfpgan/psp_r_enc_trained_r_gfp_pre_enhanced'\n",
    "r_enc_trained_r_gfp_pre_enhanced = []\n",
    "    \n",
    "i = 0\n",
    "for root, dirs, files in sorted(os.walk(path)):\n",
    "    \n",
    "    if i == n_faces+1:\n",
    "        break\n",
    "    j = 0\n",
    "    for name in files:\n",
    "        \n",
    "        if j == n_max_images_per_face:\n",
    "            break  \n",
    "            \n",
    "        r_enc_trained_r_gfp_pre_enhanced.append(facepath(str(root) + \"/\" + (str(name)), i))\n",
    "        j += 1\n",
    "        \n",
    "    i += 1\n",
    "        \n",
    "print(\"loaded %d files from %d AI r_enc_trained_r_gfp_pre_enhanced face(s)\" % (len(r_enc_trained_r_gfp_pre_enhanced), n_faces))\n",
    "print(\"Example: %s [Face: %s]\" % (r_enc_trained_r_gfp_pre_enhanced[20].path, r_enc_trained_r_gfp_pre_enhanced[20].faceno))\n",
    "print(\"----------------------------------\")\n",
    "\n",
    "#######################################################################\n",
    "# 5 - PSP RECONSTRUCTED FACES L WITH TRAINED L - PRE ENHANCED WITH GFP \n",
    "#######################################################################\n",
    "\n",
    "# Load all filenames into a list\n",
    "path = 'original_and_generated_faces/gfpgan/psp_l_enc_trained_l_gfp_pre_enhanced'\n",
    "l_enc_trained_l_gfp_pre_enhanced = []\n",
    "    \n",
    "i = 0\n",
    "for root, dirs, files in sorted(os.walk(path)):\n",
    "    \n",
    "    if i == n_faces+1:\n",
    "        break\n",
    "    j = 0\n",
    "    for name in files:\n",
    "        \n",
    "        if j == n_max_images_per_face:\n",
    "            break  \n",
    "            \n",
    "        l_enc_trained_l_gfp_pre_enhanced.append(facepath(str(root) + \"/\" + (str(name)), i))\n",
    "        j += 1\n",
    "        \n",
    "    i += 1\n",
    "        \n",
    "print(\"loaded %d files from %d AI l_enc_trained_l_gfp_pre_enhanced face(s)\" % (len(l_enc_trained_l_gfp_pre_enhanced), n_faces))\n",
    "print(\"Example: %s [Face: %s]\" % (l_enc_trained_l_gfp_pre_enhanced[25].path, l_enc_trained_l_gfp_pre_enhanced[25].faceno))\n",
    "print(\"----------------------------------\")\n",
    "\n",
    "#######################################################################\n",
    "# 6 - PSP RECONSTRUCTED FACES R WITH TRAINED R - POST ENHANCED WITH GFP \n",
    "#######################################################################\n",
    "\n",
    "# Load all filenames into a list\n",
    "path = 'original_and_generated_faces/gfpgan/psp_r_enc_trained_r_gfp_post_enhanced'\n",
    "r_enc_trained_r_gfp_post_enhanced = []\n",
    "    \n",
    "i = 0\n",
    "for root, dirs, files in sorted(os.walk(path)):\n",
    "    \n",
    "    if i == n_faces+1:\n",
    "        break\n",
    "    j = 0\n",
    "    for name in files:\n",
    "        \n",
    "        if j == n_max_images_per_face:\n",
    "            break  \n",
    "            \n",
    "        r_enc_trained_r_gfp_post_enhanced.append(facepath(str(root) + \"/\" + (str(name)), i))\n",
    "        j += 1\n",
    "        \n",
    "    i += 1\n",
    "        \n",
    "print(\"loaded %d files from %d AI r_enc_trained_r_gfp_post_enhanced face(s)\" % (len(r_enc_trained_r_gfp_post_enhanced), n_faces))\n",
    "print(\"Example: %s [Face: %s]\" % (r_enc_trained_r_gfp_post_enhanced[20].path, r_enc_trained_r_gfp_post_enhanced[20].faceno))\n",
    "print(\"----------------------------------\")\n",
    "\n",
    "#######################################################################\n",
    "# 7 - PSP RECONSTRUCTED FACES L WITH TRAINED L - POST ENHANCED WITH GFP \n",
    "#######################################################################\n",
    "\n",
    "# Load all filenames into a list\n",
    "path = 'original_and_generated_faces/gfpgan/psp_l_enc_trained_l_gfp_post_enhanced'\n",
    "l_enc_trained_l_gfp_post_enhanced = []\n",
    "    \n",
    "i = 0\n",
    "for root, dirs, files in sorted(os.walk(path)):\n",
    "    \n",
    "    if i == n_faces+1:\n",
    "        break\n",
    "    j = 0\n",
    "    for name in files:\n",
    "        \n",
    "        if j == n_max_images_per_face:\n",
    "            break  \n",
    "            \n",
    "        l_enc_trained_l_gfp_post_enhanced.append(facepath(str(root) + \"/\" + (str(name)), i))\n",
    "        j += 1\n",
    "        \n",
    "    i += 1\n",
    "        \n",
    "print(\"loaded %d files from %d AI l_enc_trained_l_gfp_post_enhanced face(s)\" % (len(l_enc_trained_l_gfp_post_enhanced), n_faces))\n",
    "print(\"Example: %s [Face: %s]\" % (l_enc_trained_l_gfp_post_enhanced[25].path, l_enc_trained_l_gfp_post_enhanced[25].faceno))\n",
    "print(\"----------------------------------\")\n",
    "\n",
    "###############################################################################\n",
    "# 8 - PSP RECONSTRUCTED FACES L WITH TRAINED L - PRE AND POST ENHANCED WITH GFP \n",
    "###############################################################################\n",
    "\n",
    "# Load all filenames into a list\n",
    "path = '/home/nero/GitHub/encrypted_face_reconstruction/original_and_generated_faces/gfpgan/psp_l_enc_trained_l_gfp_pre_post_enhanced'\n",
    "l_enc_trained_l_gfp_pre_post_enhanced = []\n",
    "    \n",
    "i = 0\n",
    "for root, dirs, files in sorted(os.walk(path)):\n",
    "    \n",
    "    if i == n_faces+1:\n",
    "        break\n",
    "    j = 0\n",
    "    for name in files:\n",
    "        \n",
    "        if j == n_max_images_per_face:\n",
    "            break  \n",
    "            \n",
    "        l_enc_trained_l_gfp_pre_post_enhanced.append(facepath(str(root) + \"/\" + (str(name)), i))\n",
    "        j += 1\n",
    "        \n",
    "    i += 1\n",
    "        \n",
    "print(\"loaded %d files from %d AI l_enc_trained_l_gfp_pre_post_enhanced face(s)\" % (len(l_enc_trained_l_gfp_pre_post_enhanced), n_faces))\n",
    "print(\"Example: %s [Face: %s]\" % (l_enc_trained_l_gfp_pre_post_enhanced[25].path, l_enc_trained_l_gfp_pre_post_enhanced[25].faceno))\n",
    "print(\"----------------------------------\")\n",
    "\n",
    "\n",
    "###############################################################################\n",
    "# 9 - PSP RECONSTRUCTED FACES R WITH TRAINED R - PRE AND POST ENHANCED WITH GFP \n",
    "###############################################################################\n",
    "\n",
    "# Load all filenames into a list\n",
    "path = '/home/nero/GitHub/encrypted_face_reconstruction/original_and_generated_faces/gfpgan/psp_r_enc_trained_r_gfp_pre_post_enhanced'\n",
    "r_enc_trained_r_gfp_pre_post_enhanced = []\n",
    "    \n",
    "i = 0\n",
    "for root, dirs, files in sorted(os.walk(path)):\n",
    "    \n",
    "    if i == n_faces+1:\n",
    "        break\n",
    "    j = 0\n",
    "    for name in files:\n",
    "        \n",
    "        if j == n_max_images_per_face:\n",
    "            break  \n",
    "            \n",
    "        r_enc_trained_r_gfp_pre_post_enhanced.append(facepath(str(root) + \"/\" + (str(name)), i))\n",
    "        j += 1\n",
    "        \n",
    "    i += 1\n",
    "        \n",
    "print(\"loaded %d files from %d AI r_enc_trained_r_gfp_pre_post_enhanced face(s)\" % (len(r_enc_trained_r_gfp_pre_post_enhanced), n_faces))\n",
    "print(\"Example: %s [Face: %s]\" % (r_enc_trained_r_gfp_pre_post_enhanced[25].path, r_enc_trained_r_gfp_pre_post_enhanced[25].faceno))\n",
    "print(\"----------------------------------\")\n",
    "\n",
    "\n",
    "\n",
    "###########################################\n",
    "# 10 - GFP RECONSTRUCTED FACES L PRETRAINED\n",
    "###########################################\n",
    "\n",
    "# Load all filenames into a list\n",
    "path = 'original_and_generated_faces/gfpgan/l_pretrained/'\n",
    "l_enc_pretrained = []\n",
    "    \n",
    "i = 0\n",
    "for root, dirs, files in sorted(os.walk(path)):\n",
    "    \n",
    "    if i == n_faces+1:\n",
    "        break\n",
    "    j = 0\n",
    "    for name in files:\n",
    "        \n",
    "        if j == n_max_images_per_face:\n",
    "            break  \n",
    "            \n",
    "        l_enc_pretrained.append(facepath(str(root) + \"/\" + (str(name)), i))\n",
    "        j += 1\n",
    "        \n",
    "    i += 1\n",
    "        \n",
    "print(\"loaded %d files from %d AI l_enc_pretrained face(s)\" % (len(l_enc_pretrained), n_faces))\n",
    "print(\"Example: %s [Face: %s]\" % (l_enc_pretrained[35].path, l_enc_pretrained[35].faceno))\n",
    "print(\"----------------------------------\")\n",
    "\n",
    "\n",
    "###########################################\n",
    "# 11 - GFP RECONSTRUCTED FACES R PRETRAINED\n",
    "###########################################\n",
    "\n",
    "# Load all filenames into a list\n",
    "path = 'original_and_generated_faces/gfpgan/r_pretrained'\n",
    "r_enc_pretrained = []\n",
    "    \n",
    "i = 0\n",
    "for root, dirs, files in sorted(os.walk(path)):\n",
    "    \n",
    "    if i == n_faces+1:\n",
    "        break\n",
    "    j = 0\n",
    "    for name in files:\n",
    "        \n",
    "        if j == n_max_images_per_face:\n",
    "            break  \n",
    "            \n",
    "        r_enc_pretrained.append(facepath(str(root) + \"/\" + (str(name)), i))\n",
    "        j += 1\n",
    "        \n",
    "    i += 1\n",
    "        \n",
    "print(\"loaded %d files from %d AI r_enc_pretrained face(s)\" % (len(r_enc_pretrained), n_faces))\n",
    "print(\"Example: %s [Face: %s]\" % (r_enc_pretrained[40].path, r_enc_pretrained[40].faceno))\n",
    "print(\"----------------------------------\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "5cfc8a30",
   "metadata": {},
   "outputs": [],
   "source": [
    "######################################################\n",
    "# OWN CLASS TO STORE IMAGE PATHS AND CLASSES\n",
    "######################################################\n",
    "\n",
    "class facepath: \n",
    "    def __init__(self, path, faceno): \n",
    "        self.path = path \n",
    "        self.faceno = faceno\n",
    "\n",
    "######################################################\n",
    "# INTRA AND INTER CLASS - FACE COMPARISON\n",
    "######################################################\n",
    "        \n",
    "def compare_faces(set_a, set_b, comparison):\n",
    "\n",
    "    '''\n",
    "    function: \n",
    "        compare two sets of faces with an face recognition algorithm\n",
    "\n",
    "    input:\n",
    "        set_a: set of faces containing objects of facepath\n",
    "        set_b: set of faces containing objects of facepath\n",
    "\n",
    "    output:\n",
    "        avg_dist: average distance metric for face comparison\n",
    "\n",
    "    '''\n",
    "                  \n",
    "    # average distance between images\n",
    "    avg_dist = 0.0\n",
    "\n",
    "\n",
    "    k = 0\n",
    "\n",
    "    # iterate over all faces\n",
    "    for face_a in set_a:\n",
    "\n",
    "        for face_b in set_b:\n",
    "\n",
    "            if \"intra\" in comparison:\n",
    "\n",
    "                # only compare intra class, not inter\n",
    "                if face_a.faceno == face_b.faceno:\n",
    "\n",
    "                    # compare faces\n",
    "                    result = DeepFace.verify(face_a.path,\n",
    "                                             face_b.path,\n",
    "                                             enforce_detection=False,\n",
    "                                             model_name =\"Facenet\") \n",
    "                    k += 1\n",
    "\n",
    "                    # compute average face distance\n",
    "                    avg_dist =  avg_dist + result['distance']\n",
    "                    \n",
    "            elif \"inter\" in comparison:\n",
    "\n",
    "                # only compare inter class not intra\n",
    "                if face_a.faceno is not face_b.faceno:\n",
    "\n",
    "                    # compare faces\n",
    "                    result = DeepFace.verify(face_a.path,\n",
    "                                             face_b.path,\n",
    "                                             enforce_detection=False,\n",
    "                                             model_name =\"Facenet\") \n",
    "\n",
    "                    k += 1\n",
    "\n",
    "                    # compute average face distance\n",
    "                    avg_dist =  avg_dist + result['distance']\n",
    "    avg_dist = float(avg_dist / k)\n",
    "                \n",
    "    return avg_dist"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "7f710f05",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Intra Class Comparison:\n",
      "---------------------------------------------------------------------\n",
      "Avg. Intra-Class Distance PLAIN VS PLAIN: \t\t\t0.41\n",
      "Avg. Intra-Class Distance LAYER-ENCRYPTED VS PLAIN: \t\t0.92\n",
      "Avg. Intra-Class Distance RES_ENCRYPTED VS PLAIN: \t\t0.76\n",
      "Avg. Intra-Class Distance AI PRE ENHANCED WITH GFP (PSP R TRAINED) R VS PLAIN: \t\t0.65\n",
      "Avg. Intra-Class Distance AI PRE ENHANCED WITH GFP (PSP L TRAINED) L VS PLAIN: \t\t0.77\n",
      "Avg. Intra-Class Distance AI POST ENHANCED WITH GFP (PSP R TRAINED) R VS PLAIN: \t\t0.68\n",
      "Avg. Intra-Class Distance AI POST ENHANCED WITH GFP (PSP L TRAINED) L VS PLAIN: \t\t0.77\n",
      "Avg. Intra-Class Distance AI PRE & POST ENHANCED WITH GFP (PSP R TRAINED) R VS PLAIN: \t0.80\n",
      "Avg. Intra-Class Distance AI PRE & POST ENHANCED WITH GFP (PSP L TRAINED) L VS PLAIN: \t0.71\n",
      "Avg. Intra-Class Distance AI (PRE TRAINED GFP) L VS PLAIN: \t0.88\n",
      "Avg. Intra-Class Distance AI (PRE TRAINED GFP) L VS PLAIN: \t0.71\n",
      "---------------------------------------------------------------------\n"
     ]
    }
   ],
   "source": [
    "######################################################\n",
    "# COMPUTING INTRA CLASS DISTANCES\n",
    "######################################################\n",
    "\n",
    "print(\"Intra Class Comparison:\")\n",
    "print(\"---------------------------------------------------------------------\")\n",
    "\n",
    "##########################\n",
    "# ORIGINAL\n",
    "##########################\n",
    "\n",
    "avg_pp_intra = compare_faces(plain_faces, plain_faces, \"intra\")\n",
    "print(\"Avg. Intra-Class Distance PLAIN VS PLAIN: \\t\\t\\t%.2f\" % avg_pp_intra)\n",
    "\n",
    "avg_enc_l_p_intra = compare_faces(encrypted_l_faces, plain_faces, \"intra\")\n",
    "print(\"Avg. Intra-Class Distance LAYER-ENCRYPTED VS PLAIN: \\t\\t%.2f\" % avg_enc_l_p_intra)\n",
    "\n",
    "avg_enc_r_p_intra = compare_faces(encrypted_r_faces, plain_faces, \"intra\")\n",
    "print(\"Avg. Intra-Class Distance RES_ENCRYPTED VS PLAIN: \\t\\t%.2f\" % avg_enc_r_p_intra)\n",
    "\n",
    "##########################\n",
    "# GFP PRE - PSP\n",
    "##########################\n",
    "\n",
    "avg_r_enc_trained_r_gfp_pre_enhanced_intra = compare_faces(r_enc_trained_r_gfp_pre_enhanced, plain_faces, \"intra\")\n",
    "print(\"Avg. Intra-Class Distance AI PRE ENHANCED WITH GFP (PSP R TRAINED) R VS PLAIN: \\t\\t%.2f\" % avg_r_enc_trained_r_gfp_pre_enhanced_intra)\n",
    "\n",
    "avg_l_enc_trained_l_gfp_pre_enhanced_intra = compare_faces(l_enc_trained_l_gfp_pre_enhanced, plain_faces, \"intra\")\n",
    "print(\"Avg. Intra-Class Distance AI PRE ENHANCED WITH GFP (PSP L TRAINED) L VS PLAIN: \\t\\t%.2f\" % avg_l_enc_trained_l_gfp_pre_enhanced_intra)\n",
    "\n",
    "##########################\n",
    "# GFP POST - PSP\n",
    "##########################\n",
    "\n",
    "avg_r_enc_trained_r_gfp_post_enhanced_intra = compare_faces(r_enc_trained_r_gfp_post_enhanced, plain_faces, \"intra\")\n",
    "print(\"Avg. Intra-Class Distance AI POST ENHANCED WITH GFP (PSP R TRAINED) R VS PLAIN: \\t\\t%.2f\" % avg_r_enc_trained_r_gfp_post_enhanced_intra)\n",
    "\n",
    "avg_l_enc_trained_l_gfp_post_enhanced_intra = compare_faces(l_enc_trained_l_gfp_post_enhanced, plain_faces, \"intra\")\n",
    "print(\"Avg. Intra-Class Distance AI POST ENHANCED WITH GFP (PSP L TRAINED) L VS PLAIN: \\t\\t%.2f\" % avg_l_enc_trained_l_gfp_post_enhanced_intra)\n",
    "\n",
    "##########################\n",
    "# GFP PRE POST - PSP\n",
    "##########################\n",
    "\n",
    "avg_l_enc_trained_l_gfp_pre_post_enhanced_intra = compare_faces(l_enc_trained_l_gfp_pre_post_enhanced, plain_faces, \"intra\")\n",
    "print(\"Avg. Intra-Class Distance AI PRE & POST ENHANCED WITH GFP (PSP R TRAINED) R VS PLAIN: \\t%.2f\" % avg_l_enc_trained_l_gfp_pre_post_enhanced_intra)\n",
    "\n",
    "avg_r_enc_trained_r_gfp_pre_post_enhanced_intra = compare_faces(r_enc_trained_r_gfp_pre_post_enhanced, plain_faces, \"intra\")\n",
    "print(\"Avg. Intra-Class Distance AI PRE & POST ENHANCED WITH GFP (PSP L TRAINED) L VS PLAIN: \\t%.2f\" % avg_r_enc_trained_r_gfp_pre_post_enhanced_intra)\n",
    "\n",
    "##########################\n",
    "# GFP PRETRAINED W/O PSP\n",
    "##########################\n",
    "\n",
    "avg_l_enc_pretrained_intra = compare_faces(l_enc_pretrained, plain_faces, \"intra\")\n",
    "print(\"Avg. Intra-Class Distance AI (PRE TRAINED GFP) L VS PLAIN: \\t%.2f\" % avg_l_enc_pretrained_intra)\n",
    "\n",
    "avg_r_enc_pretrained_intra = compare_faces(r_enc_pretrained, plain_faces, \"intra\")\n",
    "print(\"Avg. Intra-Class Distance AI (PRE TRAINED GFP) L VS PLAIN: \\t%.2f\" % avg_r_enc_pretrained_intra)\n",
    "\n",
    "print(\"---------------------------------------------------------------------\")\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "e1f0f74b",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1296x720 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "######################################################\n",
    "# INTRA AND INTER CLASS - FACE COMPARISON - PLOT\n",
    "######################################################\n",
    "\n",
    "# Setting font of the Plot\n",
    "font = {'weight' : 'normal',\n",
    "        'size'   : 12\n",
    "       }\n",
    "\n",
    "matplotlib.rc('font', **font)\n",
    "\n",
    "# Lables and Values\n",
    "labels_intra = ['plain', 'l enc', 'res enc', 'AI r(gfp(r))', 'AI l(gfp(l))', 'AI gfp(r(r))',\n",
    "                'AI gfp(l(l))', 'AI gfp(r(gfp(r)))', 'AI gfp(l(gfp(l)))', 'AI pre-train gfp(r)', 'AI pre-train gfp(l)']\n",
    "\n",
    "values_intra = [avg_pp_intra, avg_enc_l_p_intra, avg_enc_r_p_intra,\n",
    "                avg_r_enc_trained_r_gfp_pre_enhanced_intra, avg_l_enc_trained_l_gfp_pre_enhanced_intra,\n",
    "                avg_r_enc_trained_r_gfp_post_enhanced_intra, avg_l_enc_trained_l_gfp_post_enhanced_intra,\n",
    "                avg_r_enc_trained_r_gfp_pre_post_enhanced_intra, avg_l_enc_trained_l_gfp_pre_post_enhanced_intra,\n",
    "                avg_r_enc_pretrained_intra, avg_l_enc_pretrained_intra]\n",
    "\n",
    "# Bar width and generating X values to place the bars\n",
    "x = np.arange(len(labels_intra))  # the label locations\n",
    "width = 0.35  # the width of the bars\n",
    "\n",
    "# create suplots\n",
    "fig, axs = plt.subplots(2,1, figsize=(18,10))\n",
    "rects_intra = axs[0].bar(x - width/2, values_intra, width, align='edge', color='orange')\n",
    "\n",
    "########################\n",
    "# INTRA CLASS PLOT\n",
    "########################\n",
    "\n",
    "# Add some text for labels, tiSle and custom x-axis tick labels, etc.\n",
    "axs[0].set_ylabel('Average Distance')\n",
    "axs[0].set_title('Intra Class Face Comparison - GFPGAN')\n",
    "axs[0].set_xticks(x)\n",
    "axs[0].set_xticklabels(labels_intra)\n",
    "\n",
    "# average distances\n",
    "axs[0].bar_label(rects_intra, fmt='%.2f')\n",
    "\n",
    "fig.tight_layout()\n",
    "\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "2abf8d1e",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "34264d77",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "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": "3.8.10"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}