Tutorial - Evaluate Sparse Attacks on CIFAR10.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "78193fe6",
   "metadata": {},
   "source": [
    "### 0. Import libraries & functions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "5ef22921",
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "import sys\n",
    "os.environ[\"CUDA_DEVICE_ORDER\"]=\"PCI_BUS_ID\"\n",
    "os.environ[\"CUDA_VISIBLE_DEVICES\"]=\"0\";\n",
    "#from pudb import set_trace\n",
    "import numpy as np\n",
    "import torch\n",
    "from torchvision import models\n",
    "\n",
    "import torch.backends.cudnn as cudnn\n",
    "import random\n",
    "from tqdm.auto import tqdm\n",
    "import matplotlib.pyplot as plt\n",
    "import seaborn as sns\n",
    "\n",
    "from xmodels import *\n",
    "from utils_se import *\n",
    "\n",
    "from pointwise_attack import PointWiseAtt\n",
    "from spaevo_attack import SpaEvoAtt\n",
    "import pgd_attacks_pt2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "3a988c7c",
   "metadata": {},
   "outputs": [],
   "source": [
    "def imshow(img):\n",
    "    \n",
    "    npimg = img[0].cpu().numpy()\n",
    "    npimg = np.transpose(npimg,(1, 2, 0))\n",
    "    npimg = np.clip(npimg, 0, 1)\n",
    "    plt.imshow(npimg)\n",
    "    plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "06f31f62",
   "metadata": {},
   "outputs": [],
   "source": [
    "def perturbation_heat_map(xo,xa):\n",
    "    \n",
    "    fig_dims = (5, 5)\n",
    "    fig, ax = plt.subplots(figsize=fig_dims)\n",
    "    x=torch.abs(xo-xa).sum(dim=1).cpu()[0]\n",
    "    sns.heatmap(x,ax=ax,xticklabels=False, yticklabels=False,cbar = False)\n",
    "    plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e1b25425",
   "metadata": {},
   "source": [
    "### 1. Load data, pretrained model, evaluation set"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "f57bef7f",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Files already downloaded and verified\n"
     ]
    }
   ],
   "source": [
    "# a. Load dataset\n",
    "batch_size = 1\n",
    "dataset = 'cifar10' # or 'imagenet'\n",
    "datapath = '../datasets/cifar10'#'../datasets/ImageNet-val'\n",
    "testloader, testset = load_data(dataset,data_path=datapath,batch_size=batch_size)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "3dcdced9",
   "metadata": {},
   "outputs": [],
   "source": [
    "# b. Load pre-trained model\n",
    "\n",
    "# 'resnet50' if pre-trained model from Pytorch. 'resnet18' if using pre-trained cifar10 model\n",
    "arch = 'resnet18' \n",
    "\n",
    "# None means using pre-traineded model from Pytorch or default path. Otherwise, please change model_path = '...'\n",
    "model_path = None \n",
    "\n",
    "if dataset == 'cifar10':\n",
    "    num_classes = 10\n",
    "    norm = False # True means pre-trained model does \"not\" normalized data while training.\n",
    "    \n",
    "elif dataset == 'imagenet':\n",
    "    num_classes = 1000\n",
    "    norm = False # True means pre-trained model does \"not\" normalized data while training.\n",
    "\n",
    "net = load_model(arch,model_path)\n",
    "model = PretrainedModel(net,dataset,arch,norm)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "0b63ed2e",
   "metadata": {},
   "outputs": [],
   "source": [
    "# c. Load evaluation set\n",
    "targeted = True # True means targeted attack. False means untargeted attack.\n",
    "seed = 999\n",
    "\n",
    "ID_set = get_evalset(model,dataset,arch,testset,seed,targeted)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "f84a2f5b",
   "metadata": {},
   "source": [
    "### 2. PointWise Evaluation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "3f3637c4",
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "nqry = 400; Reset value to original -> new distance: 11.239893913269043; L0 = 947; pred label: tensor([5], device='cuda:0')\n",
      "nqry = 600; Reset value to original -> new distance: 9.634395599365234; L0 = 782; pred label: tensor([5], device='cuda:0')\n",
      "refine stage!\n",
      "nquery = 1000; Set value at 88 from [0.8352941  0.5568628  0.91764706 0.8862745 ] to [0.8352941 0.5568628 0.8827512 0.8862745] (original has [0.8352941 0.5568628 0.654902  0.8862745]) -> new distance: 7.551494598388672; \n",
      "pred label:tensor([5], device='cuda:0'); L0:465\n",
      "nquery = 1200; Set value at 167 from [0.76862746 0.73333335 0.50980395 0.7882353 ] to [0.76862746 0.73333335 0.50980395 0.786466  ] (original has [0.76862746 0.73333335 0.50980395 0.7647059 ]) -> new distance: 7.122283935546875; \n",
      "pred label:tensor([5], device='cuda:0'); L0:435\n",
      "nquery = 1600; Set value at 258 from [0.827451   0.8        0.39215687 0.84705883] to [0.827451   0.8        0.39178157 0.84705883] (original has [0.827451   0.8        0.2        0.84705883]) -> new distance: 6.866913795471191; \n",
      "pred label:tensor([5], device='cuda:0'); L0:405\n"
     ]
    }
   ],
   "source": [
    "i = 31#114\n",
    "o = ID_set[i,1] #oID\n",
    "oimg, olabel = testset[o]\n",
    "oimg = torch.unsqueeze(oimg, 0).cuda()\n",
    "query_limit = 2000\n",
    "nquery = 0\n",
    "n_pix = 4\n",
    "seed = 0\n",
    "\n",
    "if targeted:\n",
    "    t = ID_set[i,3] #tID, 3 is index acrross dataset - 4 is sample index in a class (not accross dataset)\n",
    "    timg, tlabel = testset[t]\n",
    "    timg = torch.unsqueeze(timg, 0).cuda()\n",
    "else: \n",
    "    init_mode = 'salt_pepper_att' \n",
    "    timg, nqry,_ = gen_starting_point(model,oimg,olabel,seed,dataset,init_mode)\n",
    "    tlabel = None\n",
    "    nquery += nqry\n",
    "\n",
    "attack = PointWiseAtt(model,flag=targeted)\n",
    "\n",
    "#adv, nqry, Dt = attack.pw_perturb(oimg.cpu().numpy(),timg.cpu().numpy(),olabel,tlabel,max_query=query_limit)\n",
    "adv, nqry, Dt = attack.pw_perturb_multiple(oimg.cpu().numpy(),timg.cpu().numpy(),olabel,tlabel,n_pix,max_query=query_limit)\n",
    "    \n",
    "adv = torch.from_numpy(adv.reshape(oimg.shape)).cuda()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "070cf7bc",
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Source image:\n"
     ]
    },
    {
     "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": [
      "Starting image:\n"
     ]
    },
    {
     "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": [
      "Adversarial Example:\n"
     ]
    },
    {
     "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": [
      "Perturbation Heat Map:\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 360x360 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "print('Source image:')\n",
    "imshow(oimg)\n",
    "\n",
    "print('Starting image:')\n",
    "timg, _ = testset[t]\n",
    "timg = torch.unsqueeze(timg, 0).cuda()\n",
    "imshow(timg)\n",
    "\n",
    "print('Adversarial Example:')\n",
    "imshow(adv)\n",
    "\n",
    "print('Perturbation Heat Map:')\n",
    "perturbation_heat_map(oimg,adv)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "5e26f832",
   "metadata": {},
   "outputs": [],
   "source": [
    "DPw = Dt"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "6ff4b3ce",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "- sample ID: 5013\n",
      "- ground-truth: 0 \n",
      "- target label: 5 \n",
      "- AdvEx label: 5\n",
      "- L0 dist:355\n"
     ]
    }
   ],
   "source": [
    "# check prediction of adversarial example\n",
    "alabel = model.predict_label(adv).item()\n",
    "dist = l0b(oimg,adv)#DPw[-1]\n",
    "print(f'- sample ID: {o}\\n- ground-truth: {olabel} \\n- target label: {tlabel} \\n- AdvEx label: {alabel}\\n- L0 dist:{dist}')"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b20c0fde",
   "metadata": {},
   "source": [
    "### 3. PGD_L0 Evaluation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "9ae06155",
   "metadata": {},
   "outputs": [],
   "source": [
    "if dataset == 'imagenet':\n",
    "    mu = (0.0, 0.0, 0.0)\n",
    "    std = (1.0, 1.0, 1.0)\n",
    "\n",
    "    if targeted:\n",
    "        num_steps = 5000\n",
    "        step_size = 224*224*3*(0.025)\n",
    "        LB_L0 = 0\n",
    "        UB_L0 = 4000\n",
    "    else:\n",
    "        num_steps = 2000\n",
    "        step_size = 224*224*3*(0.05)\n",
    "        LB_L0 = 0\n",
    "        UB_L0 = 800\n",
    "\n",
    "elif dataset == 'cifar10':\n",
    "    mu = (0.0, 0.0, 0.0)\n",
    "    std = (1.0, 1.0, 1.0)\n",
    "\n",
    "    if targeted:\n",
    "        num_steps = 250\n",
    "        step_size = 120000.0/255.0\n",
    "        LB_L0 = 0\n",
    "        UB_L0 = 200\n",
    "    else:\n",
    "        num_steps = 100\n",
    "        step_size = 120000.0/255.0\n",
    "        LB_L0 = 0\n",
    "        UB_L0 = 100\n",
    "\n",
    "pgd_args = {'type_attack': 'L0',\n",
    "            'n_restarts': 1,\n",
    "            'num_steps': num_steps,\n",
    "            'step_size': step_size,\n",
    "            'epsilon': -1,\n",
    "            'sparsity': 100,\n",
    "            'mu': mu,\n",
    "            'std':std}\n",
    "\n",
    "i = 31\n",
    "o = ID_set[i,1] #oID\n",
    "img, label = testset[o]\n",
    "img = torch.unsqueeze(img, 0).cuda()\n",
    "oimg = img.permute(0,2,3,1) # equal to np.transpose(img,(0,2, 3, 1))\n",
    "olabel = np.expand_dims(label,axis = 0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "0d041874",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[0] 5 100 100\n",
      "[0] 5 50 50\n",
      "[0] 5 25 25\n",
      "[0] 5 12 12\n",
      "[0] 5 6 6\n",
      "[0] 5 3 3\n"
     ]
    }
   ],
   "source": [
    "seed = 0\n",
    "np.random.seed(seed)\n",
    "torch.manual_seed(seed)\n",
    "\n",
    "ub = UB_L0\n",
    "lb = LB_L0  \n",
    "\n",
    "while (ub - lb) > 5:\n",
    "    k = int((ub + lb)//2)\n",
    "       \n",
    "    # 2. Run attack\n",
    "    # not \"net\" but \"model\" instead because the net is drafted in PreTrainedModel \n",
    "    # and mu,std are feeded to pgd.attack with 0 & 1 since input is transformered PreTrainModel\n",
    "    pgd_args['sparsity'] = k\n",
    "    attack = pgd_attacks_pt2.PGDattack(model, pgd_args) \n",
    "\n",
    "    if targeted:\n",
    "        label = np.expand_dims(ID_set[i,2].cpu().numpy(),axis=0) #tcla\n",
    "        tlabel = label.item()\n",
    "        adv, pgd_adv_acc, pixel_changed = attack.perturb_norm_target(oimg.cpu().numpy(), label)\n",
    "    else: #untaget\n",
    "        label = olabel\n",
    "        adv, pgd_adv_acc, pixel_changed = attack.perturb_norm(oimg.cpu().numpy(), label)\n",
    "                \n",
    "    adv = torch.from_numpy(adv).cuda()\n",
    "    adv = adv.permute(0,3,1,2)\n",
    "    alabel = model.predict_label(adv)\n",
    "    if alabel.cpu().numpy()!=olabel.item():\n",
    "        ub = k\n",
    "        L0 = pixel_changed.item()\n",
    "        alabel_fin = alabel.item()\n",
    "        adv_fin = adv.clone()\n",
    "        print(olabel,alabel_fin,L0,k)\n",
    "    else:\n",
    "        lb = k"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "d8c37d04",
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Source image:\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAPsAAAD5CAYAAADhukOtAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4yLjIsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+WH4yJAAAdXUlEQVR4nO2da4yd13We33Vucx+Sw+uIulChFNm07NDSmHVgO3CsSFVsNbKBWrB/OAJqhEkQAzGQ/BBcoHaB/nCK2oaBFi7oWo5SuJaV2K6FRE3tKgGEGIgsiqGomyVKFCWSGl7nyrmd2+qPc4hS6n7XDOdyhtF+H4Dgmb1mf986+3zrfOfsd9Za5u4QQrzzKay3A0KIzqBgFyITFOxCZIKCXYhMULALkQkKdiEyobSSyWZ2N4BvAigC+G/u/tXo9wc3DPrW7dtWcsq3nh8WGZdjWuSE6ZnR8cJzLduRKz8ocX3NWN75OuzkMlgLD5n6vRxZ/PToaUxMTCbdXHawm1kRwH8BcCeAkwCeMrNH3f0FNmfr9m346n/+etLW9CY9V8HSH0CKBb70pSK3FaIrMbAVC8wPfrhicDwL/I8uq8h/oz4GTgbHazYbwTR+MZZKRTaLzikE62HkGgB4sETnC9+MguMVonBv8onR+RqN9BrX63ztGf/md/+A2lbyMX4fgFfc/Zi7VwE8DODeFRxPCLGGrCTYdwI4cdnPJ9tjQoirkDXfoDOz/WZ20MwOTk1OrfXphBCElQT7KQDXXfbzte2xt+DuB9x9xN1HBjcMruB0QoiVsJJgfwrAzWZ2o5lVAHwGwKOr45YQYrVZ9m68u9fN7AsA/jda0tuD7v58NMcAFJHeYSwGW6Alst1d6SoHc7gfzWDXlO8iA8Vi2lYpRTvnXGWIdqaj3dtCsLPO5Jpo55w9r9bxuB/u3MlyqZIcbwS7+9HJCgXuY7RTz9Y4Un9CP4L1iNSVSEYzso4WXMOw9DqGCklwuEVx98cAPLaSYwghOoP+gk6ITFCwC5EJCnYhMkHBLkQmKNiFyIQV7cZfKcWCYWN/d9LWVUlLNQCXXcpd3P1ikAgTySAeSIBU1WjW6JxGnUs8pRL3P5LXooQRnvgRJaDwc1kgeUXqFcsmadb4WtXqdWorlbjMWiwG1w553lGCUiF4XaJkF2/yBWkESS3s5WwG12KTXnOBbEgtQoh3FAp2ITJBwS5EJijYhcgEBbsQmdDh3fgCBnu7kraonhzbOLUm370tBrWiGs53RmsLC9TGCA6HcrCLXADf6W7U+EEbQTJGkSTyhGWYguyOSDGIynE5WZRKhR+vXI7uPdzWJGWdAJ701GzwnfNyJXrNgrJl4Q5/YGukbdF1VUDaxyiOdGcXIhMU7EJkgoJdiExQsAuRCQp2ITJBwS5EJnRUerMC0F1JS0MLC1U6jyXCRF1OIlkokic8mEj9qKTlRADwZqB5hXXJljUNDZIfUY4kL1IvDoiTO4IcmdaLnaAU1LuLEnKien31+pV3YqkHSTfR+pajJJlgXj1IAGJrHF2nzQZZq6AuoO7sQmSCgl2ITFCwC5EJCnYhMkHBLkQmKNiFyIQVSW9mdhzANIAGgLq7jyzy+yiVSUZRmJaVJpKFLJC8KhbIUEQaBIBymbQ0CuqLzc7NUVuUeVUIJMCoTt58NS3x1Br8eTWb6bqAi/lhQdYea9kVvc6NYD3CNlRhSb60H83g2qkHWXSRBDg/z+Xj6Bphh6zXuDzIfGwGC7UaOvtvuvv5VTiOEGIN0cd4ITJhpcHuAH5qZk+b2f7VcEgIsTas9GP8h939lJltA/AzM/uluz9x+S+03wT2A8A112xf4emEEMtlRXd2dz/V/v8sgB8D2Jf4nQPuPuLuI5uGNq7kdEKIFbDsYDezPjMbuPQYwF0Anlstx4QQq8tKPsZvB/DjdluhEoD/4e5/u9yDNUK5Iw1r7QPEhQELRI4BECkrMCprcLmjp5tnxM3MzFDb9NQ0tW3dzr8OsdZFxWVmm1WC4ovF4F7BsvYiKS+6BuYW5qnNg3ZYZSL1Rq2movZg1SqfNz/Pi5VGx2Qy4EJQ/JS18/KgJ9eyg93djwH4teXOF0J0FklvQmSCgl2ITFCwC5EJCnYhMkHBLkQmdLTgZLPRxMWZi0nb3CzPDmOyxeDAAJ8TVEOMZBAmaQBcYItkrSi76tChQ9T29NNPU9tv3XkXtd2+7//7uyYA8fOqBcUQI6I+diwjcTlyEgD09PRQW1TTk8l50WsWXR+VclBcNMjMm5qaorbZ2dnkeFcXP1dfX19yvBBJrNQihHhHoWAXIhMU7EJkgoJdiExQsAuRCR3djS8UCxjo70/aop1d1qqnGOy4R7vg0W5rlBTCkw+CbdiAHTt2UFu0Q/7www9T2zTZ2f3IRz5C50TPuVrlddWKQdYQa80V7bhHfkS75wiSP5wcM3rNotZQpSI/V0T0vHt7e5Pjy72uGLqzC5EJCnYhMkHBLkQmKNiFyAQFuxCZoGAXIhM6Kr1F7Z+6ApWhROqqhQ2jwlZC/GRhkgwZj5IPikHNtb1791JbJEM9/Mgj1Pbd7343OT4+Pk7nfOITn6C27m7eGqoZSVTkNYueVyiXUssiyTXkfIWgbl2UgBK91r29QWJQcF0xWTGaM30xnVAWysrUIoR4R6FgFyITFOxCZIKCXYhMULALkQkKdiEyYVHpzcweBHAPgLPufmt7bAjADwDsAnAcwH3uzrWdSziAZloaiGS0MpFxmkG7oGrYTiqoMVaoUFuDyBqRBOXVSDTiLY1uueVmarv/dz9HbY/85Y+S448//nd0TtTS6K67/iW1bRhIZzACQI2s//wcf86RhFbp4q9LocgvYyb1BSpf2IaqVAwktEDO6+9LZ7YBXC4rlbjMVymnbaUV1qD7cwB3v23sAQCPu/vNAB5v/yyEuIpZNNjb/dbH3jZ8L4CH2o8fAvDJVfZLCLHKLPc7+3Z3H20/Po1WR1chxFXMijfovPWFg36RMbP9ZnbQzA6OjU2s9HRCiGWy3GA/Y2bDAND+/yz7RXc/4O4j7j4yNLRxmacTQqyU5Qb7owDubz++H8BPVscdIcRasRTp7fsAPgpgi5mdBPBlAF8F8IiZfR7A6wDuW+oJC+QTvwXZOtX5tFwzM8dbRi1UuXxSrvDspNkFXuhxgLSbMuPvmfML3MdigT/nDf0bqG1jIHltGBhMjk9MTNM5P//5P1Lb2GQ6uwoAPnXv71DbDdddmxyv1vjr0gzW/sLYJLWxgo0A0E8KnFpwn6tVuQQYvGTwIpcwg3qTNDOPydQAUCGRG51n0WB3988S0x2LzRVCXD3oL+iEyAQFuxCZoGAXIhMU7EJkgoJdiEzoaMHJZrOJ6Zl0L7Jalcsus7MzyfGZed6HrOY8+8fmgv5lQdaQlXuS46XgLbMayDgb+vuobWGBZ4c99dQvqI2tFZMNAaBY4evxwgsvUts86SsHAJ+5L63G3nDDDdyPYtTvj6epzQWZdFVyXUUZdn19/HWpBMUol1swkxlZ5iAANJrp10wFJ4UQCnYhckHBLkQmKNiFyAQFuxCZoGAXIhM6Kr01mk1MTaeloWaQ4QNLy2E9fekMLwCoznJpZWqaZ3JZUDTwwmQ6c6wreMvs7+FyUiXQ7OZmuBx28uTJK7bteS/vK1cP1v6XL79CbU8++RS1TU9MJcf/1e/wTLn3vOc91BZltjWbXKKamUlfb7VaJPVySXFykmffRYUqe4KeeYMb0hmO9Qb3sdpIX9+NQP7TnV2ITFCwC5EJCnYhMkHBLkQmKNiFyIQOJ8I4ZkkSSrnMd637etOJCV4IklZKvF3QPEkWAYDpYKfePb3T2TPAEyeGhoaobeNGXm23v5uvx623vpfaXjp6PDl+/fXX0zmvHX+d2s6dpYWDMR7UhfvHJw8mx59//pd0zr338l4jd97Fq6B1Bckp7LqK5oyP805mM8FO/ZkzZ7gfFX493kSUhllSexEApufSu/H1hnbjhcgeBbsQmaBgFyITFOxCZIKCXYhMULALkQlLaf/0IIB7AJx191vbY18B8HsAzrV/7Uvu/thix3J3VOvpZIG5BZ74UelO134rgice9Jb5+9imfp6UMNDFl+Sandckx3uIfwBQCBJruoNzLSxwiSdq8XPbbbcnx585/Aydc/DQYWo7P8ZlqHrQYqtG6gOePPEmnfOrv/oStX3605+mtlKZJ/KwpJZIetu8eTO1DWzgcmmhzI9ZDxJvJqbScu+Zc+fpnLPjaVluPmihtZQ7+58DuDsx/g1339v+t2igCyHWl0WD3d2fADDWAV+EEGvISr6zf8HMjpjZg2a2adU8EkKsCcsN9m8B2A1gL4BRAF9jv2hm+83soJkdnJxMFzQQQqw9ywp2dz/j7g1v/bH4twHsC373gLuPuPvIhg28sowQYm1ZVrCb2fBlP34KwHOr444QYq1YivT2fQAfBbDFzE4C+DKAj5rZXrQa1xwH8PtLOVmxWMTQhnQboijrjdnm57g85UFdtf5unoFU6ue1znq70n4EHaNQr/HMpTmuXMGbdWqbJnXVAOBvHvub5PjFWS5t3nzLu6itn9RHA4DRk1xGmyBtr7bv2ELn/NadH6W2vgEua83N8+ugSuq4zU3y1yVq1rQQSFu93dxH7+LX3Ph4ev977ALfFz83Npccr9f5dbNosLv7ZxPD31lsnhDi6kJ/QSdEJijYhcgEBbsQmaBgFyITFOxCZEJHC06WS0UMb00XYLQglevixXRWUMG5dlUuc6ljcCAt/wEteZDhpLVOs5aWQQCgUeeS11yV2wYCH28fSWe2AcD/+tvHk+O9M1xqqpT5ZRC1O6rVuP9m6bX6wAfeT+eMjLyP2qo1Lq/NEZkPAObJGs/N8desXufyWk+FZ0x2lYNrJ+huNkQKj5rxe/Hg5vRz7g6y+XRnFyITFOxCZIKCXYhMULALkQkKdiEyQcEuRCZ0VHprkdYgZme5FDJPel4VCst7r2oSCW2xY9J5xnWVapVnIVUXuG2mwNdjmhQoBIDh4eHkePNN3ocsktc2b+a96roDeXN+Li0d3nPPPfxcW3ihx5kg069e4xIse81mLvLjNRr8eMNbd1BbKZBta0HBSSe6XCQDlyrTyfFyiV+/urMLkQkKdiEyQcEuRCYo2IXIBAW7EJnQ0d34ZrNJd92jHVBGuFtZ4k8tSrqJduOZLZrTFSRONBpcFagGyR3u3P9du3Ylxyem+O7z7Bw/V63JE2iiHfIP3H5benwfLUQc7ljPznE/Fub5PHj6GqnX+dpPT3O142LwnPt7ef3CapD0xOrGzc7yczVJEhLb2Qd0ZxciGxTsQmSCgl2ITFCwC5EJCnYhMkHBLkQmLKX903UA/gLAdrSyWA64+zfNbAjADwDsQqsF1H3uPh4dy92ppFQq8fZPlUo64WJ2ltcli2ShTZt4h+lIlmOyRlcXl9fKJV4TrK+PJ5KceOMEtZ0M2i7tefee5HhPL69p90/PHKG28eOvU1skOd5xxx3pOcuoNQgAFwM57PgJnuRjSJ9vfoFLeYUCl3THxyaorR7Ia8uRicuBbHthIn3tN4O2Z0u5s9cB/Im77wHwQQB/ZGZ7ADwA4HF3vxnA4+2fhRBXKYsGu7uPuvuh9uNpAC8C2AngXgAPtX/tIQCfXCsnhRAr54q+s5vZLgDvB/AkgO3uPto2nUbrY74Q4iplycFuZv0Afgjgi+4+dbnNW19mk18WzGy/mR00s4Pj47xIghBibVlSsJtZGa1A/567/6g9fMbMhtv2YQBnU3Pd/YC7j7j7yKZNvNe3EGJtWTTYrbU9/R0AL7r71y8zPQrg/vbj+wH8ZPXdE0KsFkvJevsQgM8BeNbMDrfHvgTgqwAeMbPPA3gdwH1LOSGTtupB1luZSBNRllSURbewwLO8zp0/T2093WkppKuL15LrqnAfq0H7pMPPPMvnBRlxQ0PpmnG7b7qJzjlxikt5h488R2233voeatu8OV1P7ugrr9A5/f091DYWfAU8c/octQ1uSH+a3LGd15Lr6uZyaW2eS4AXLlygtigLkyWqNcHlujdGp5Lj1aAe36LB7u7/ABCxEkiLqUKIqw79BZ0QmaBgFyITFOxCZIKCXYhMULALkQkdLTjZaDRwYSItGURZQb2kkF+TigRAX18/tdUCeaJe4zKa9aTfG0+dOU3nzM1zeW1qkss4p89zqemm3buobYi0UHrpJS55HX3pZWorBVlq99xzF7U1Lf28X3n1DX6uSh+1nRpN/s0WAGAmKDg5RFTK4WtvoHN6gmzEWo1nU/YO8Guu0eDZaC//8rXk+LnzPKvzjTPp6yMqHqo7uxCZoGAXIhMU7EJkgoJdiExQsAuRCQp2ITKho9JboVBEX39anugmGWVAq0dcir5+LtV0laMsIy6DlMq8GCXLXGo0uZQ3NsEltFqVy3y9/YPUNrfApaaLpAjnkWd5Ft3x48ep7b3v20ttu3bvpraJ6bTEOjXHfT91jBe3nKsGffGC1/PM2KnkeC3ol7djB5fQhgaCay4omnrh/Bi3XUjLeecD6Y1KeXwpdGcXIhcU7EJkgoJdiExQsAuRCQp2ITKho7vxpXIJW7duSdqqVb5LywhyNFAsciPb3QeAUp3vkLM6edvIcwKAnh6+e/tykJzS18PrsfUGtoNPH0qOP/MsryXX1Z1ONAKAd72b15l76hDf4b8wmd6NHyUJHAAwOcXXvtrkr2c9SIhq1tKv9QtHj9M5p8/zRJh337iL2rZu2Eht1UBNWCDqyrZt2+icjeQ2XalwRUB3diEyQcEuRCYo2IXIBAW7EJmgYBciExTsQmTCotKbmV0H4C/QasnsAA64+zfN7CsAfg/Apd47X3L3xxY7XqGQrjVXLHJpolBIvyfNzc3RORcvBrW4yPEAYHKSS0MsWWdoiCfPbN/OO1n39XDJa+xCIFGRJBMAeOnoq8nxN07yFk93fIzXkvsXv/4hanv+5Reo7bXjJ5Ljc1Vea7DR5G2XFgLpKhJtvc56K/GMkfEJXhvw2DFeQ6+ymz+3zZv5NTJ+YTo5PjnNE2Ga5bTEFiZ5Ucv/ow7gT9z9kJkNAHjazH7Wtn3D3f/TEo4hhFhnltLrbRTAaPvxtJm9CGDnWjsmhFhdrug7u5ntAvB+AE+2h75gZkfM7EEz459ThBDrzpKD3cz6AfwQwBfdfQrAtwDsBrAXrTv/18i8/WZ20MwOjo2Nr4LLQojlsKRgN7MyWoH+PXf/EQC4+xl3b7h7E8C3AexLzXX3A+4+4u4j0UaWEGJtWTTYzcwAfAfAi+7+9cvGhy/7tU8B4JkWQoh1Zym78R8C8DkAz5rZ4fbYlwB81sz2oiXHHQfw+4sdyN1RraYlsUolkF0W0nNmZubpnKlpbpuZ4S18Ll7kssvOa65JjpdLPEvKG1wY2jLEs6S6g/Xo7uW2W/aks9Ree4NLb9fesIvaSoEft+7ZQ20DAxuS40ee45l+rx7jbbTKJV6jsGD8nlXz9PrffOONdM4N16dbaAHAYDfPKtu2Of2cAaB/IHg9yTF/8RS/f752Kr1WtRpvN7aU3fh/AJI5hItq6kKIqwf9BZ0QmaBgFyITFOxCZIKCXYhMULALkQkdLThZq9Vw+szZpG37Np4dNjmVzvJ64wSXakgXJAC8cCQAzM9z6aLeOJcc7+3h2U693XyJg7qXWJjnWXvNwP+tZB3f9W4uk01d5FLk0WOvUVtvF79XDPami2IOb+HFOcfO8Wy+eVI4EgCMtOUCgNlaWnorOl/Dvi4upS4scGnWjWcxFoPWUIMb07Lc7pt4Csq5ybQfUUan7uxCZIKCXYhMULALkQkKdiEyQcEuRCYo2IXIhM5Kb/U6Rs9eSNqmLkYZbOnii2+cOEPnVGv8qe28lksac9O839jUubRs2D/Ie41du3Mrtc3PcXltocolwOkga6+LyEY7SMYeALw5ep7a/u6Jnwfn4lJTV1c6S61U4tlffRt5vYMLJ05Sm4OvVbGQvg5Gz/LnfPpcWmIFgA39/LXevoPLivU6v6/WF9Iy4JZNA3TOvttvSY7/9H8G2YHUIoR4R6FgFyITFOxCZIKCXYhMULALkQkKdiEyoaPSW6PRxNR0WjZy45ljF8bS0tvY+ASd48YlntHTo9RWI1lSAFAopGUXK3LfyxUuhdR44hWmgxr7tSDrzT39/r07KLBYC2ShMy8cpbaxCS6XNprptWLSIAAMDHCpqcFbmGFulvf86+pKXwcWZIexOQCwbfg6apud55l5zQZPw+wup33p6uqjc4Z3pGXPMukBB+jOLkQ2KNiFyAQFuxCZoGAXIhMU7EJkwqK78WbWDeAJAF3t3/8rd/+ymd0I4GEAmwE8DeBz7s4zEgCUyxXs2JFOyGjUox3m9M5uTw9PxLgwzmuF1es82aUY7Kz39KR31ht1vgt7liT+AEBvL/cfCNSJ8zxRY+Ziete3HOzsbh3i7Y6u2T5NbafPpVUSAJibT6saMzN8V7pa5a+LJZsStejpSde7i2wDA/10TuTjiVPpZCgAGB3liVkbB/n5brw+nZhVKfPwLJTS9+l6EEdLubMvAPiYu/8aWu2Z7zazDwL4MwDfcPebAIwD+PwSjiWEWCcWDXZvcek2WW7/cwAfA/BX7fGHAHxyTTwUQqwKS+3PXmx3cD0L4GcAXgUw4e6XPnedBMCTxIUQ686Sgt3dG+6+F8C1APYBeNdST2Bm+83soJkdnJrk3/GEEGvLFe3Gu/sEgL8H8OsANprZpR2EawGcInMOuPuIu48MbuD9q4UQa8uiwW5mW81sY/txD4A7AbyIVtD/6/av3Q/gJ2vlpBBi5SwlEWYYwENmVkTrzeERd/9rM3sBwMNm9h8A/BOA7yx2oGazidmZdN21SA7rqqQlqmuGeeJEuYsnyUTJAjNBfbfBwfT5FhZ48szMRd7SqL+fyzGDg/xTUG83l5rePPVmcvzMuSBpaJbLNeViIP9YcPlYOnOlGNxenCTPAIAbt20ZGqK2IWKLJNb5oPXWhXF+fXjQz+vceS5hnie2wQEuzW4aSl8fkXy5aLC7+xEA70+MH0Pr+7sQ4p8B+gs6ITJBwS5EJijYhcgEBbsQmaBgFyITzD0o7rXaJzM7B+D19o9bAPAePJ1DfrwV+fFW/rn5cYO7J3uOdTTY33Jis4PuPrIuJ5cf8iNDP/QxXohMULALkQnrGewH1vHclyM/3or8eCvvGD/W7Tu7EKKz6GO8EJmwLsFuZneb2Utm9oqZPbAePrT9OG5mz5rZYTM72MHzPmhmZ83sucvGhszsZ2Z2tP3/pnXy4ytmdqq9JofN7OMd8OM6M/t7M3vBzJ43sz9uj3d0TQI/OromZtZtZr8ws2fafvz79viNZvZkO25+YGa8l1YKd+/oP7TKpr4K4FcAVAA8A2BPp/1o+3IcwJZ1OO9vALgNwHOXjf1HAA+0Hz8A4M/WyY+vAPjTDq/HMIDb2o8HALwMYE+n1yTwo6NrAsAA9LcflwE8CeCDAB4B8Jn2+H8F8IdXctz1uLPvA/CKux/zVunphwHcuw5+rBvu/gSAsbcN34tW4U6gQwU8iR8dx91H3f1Q+/E0WsVRdqLDaxL40VG8xaoXeV2PYN8J4MRlP69nsUoH8FMze9rM9q+TD5fY7u6X2sueBrB9HX35gpkdaX/MX/OvE5djZrvQqp/wJNZxTd7mB9DhNVmLIq+5b9B92N1vA/DbAP7IzH5jvR0CWu/saL0RrQffArAbrR4BowC+1qkTm1k/gB8C+KK7v6XETyfXJOFHx9fEV1DklbEewX4KwOVNrmmxyrXG3U+1/z8L4MdY38o7Z8xsGADa//PWI2uIu59pX2hNAN9Gh9bEzMpoBdj33P1H7eGOr0nKj/Vak/a5r7jIK2M9gv0pADe3dxYrAD4D4NFOO2FmfWY2cOkxgLsAPBfPWlMeRatwJ7COBTwvBVebT6EDa2JmhlYNwxfd/euXmTq6JsyPTq/JmhV57dQO49t2Gz+O1k7nqwD+7Tr58CtoKQHPAHi+k34A+D5aHwdraH33+jxaPfMeB3AUwP8BMLROfvx3AM8COIJWsA13wI8Po/UR/QiAw+1/H+/0mgR+dHRNALwPrSKuR9B6Y/l3l12zvwDwCoC/BNB1JcfVX9AJkQm5b9AJkQ0KdiEyQcEuRCYo2IXIBAW7EJmgYBciExTsQmSCgl2ITPi/JA8lmhi6O8YAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Adversarial Example:\n"
     ]
    },
    {
     "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": [
      "Perturbation Heat Map:\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 360x360 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "print('Source image:')\n",
    "imshow(img)\n",
    "\n",
    "print('Adversarial Example:')\n",
    "imshow(adv_fin)\n",
    "\n",
    "print('Perturbation Heat Map:')\n",
    "perturbation_heat_map(img,adv_fin)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "c6e250dc",
   "metadata": {},
   "source": [
    "### 4. SparseEvo Evaluation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "df3d426f",
   "metadata": {},
   "outputs": [],
   "source": [
    "if dataset == 'cifar10':\n",
    "    n_pix = 4 # 4, 16, 64, 256 only required for uni_rand: 4/(32*32) = 196/(224*224) = 0.004\n",
    "    if targeted:\n",
    "        mu = 0.01\n",
    "    else:\n",
    "        mu = 0.04\n",
    "elif dataset == 'imagenet':\n",
    "    n_pix = 196 # 49, 196, 784, 3136 only required for uni_rand: 4/(32*32) = 196/(224*224) = 0.004\n",
    "    if targeted:\n",
    "        mu = 0.001\n",
    "    else:\n",
    "        mu = 0.004\n",
    "\n",
    "query_limit = 2000\n",
    "nquery = 0\n",
    "pop_size = 10\n",
    "cr = 0.5\n",
    "seed = 0\n",
    "n = n_pix\n",
    "\n",
    "attack = SpaEvoAtt(model,n,pop_size,cr,mu,seed,flag=targeted)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "1316559b",
   "metadata": {},
   "outputs": [],
   "source": [
    "# change i to select sample from selected (generated) evaluation set\n",
    "i = 31# 115,51,113,999,12,123,114,123\n",
    "o = ID_set[i,1] #oID\n",
    "oimg, olabel = testset[o]\n",
    "oimg = torch.unsqueeze(oimg, 0).cuda()\n",
    "\n",
    "if targeted:\n",
    "    t = ID_set[i,3] #tID, 3 is index acrross dataset - 4 is sample index in a class (not accross dataset)\n",
    "    timg, tlabel = testset[t]\n",
    "    timg = torch.unsqueeze(timg, 0).cuda()\n",
    "else:\n",
    "    init_mode = 'salt_pepper_att' #'gauss_rand' #'salt_pepper'\n",
    "    timg, nqry,_ = gen_starting_point(model,oimg,olabel,seed,dataset,init_mode)\n",
    "    tlabel = None\n",
    "    nquery += nqry\n",
    "    \n",
    "adv, nqry,Dt = attack.evo_perturb(oimg,timg,olabel,tlabel,max_query=query_limit)\n",
    "\n",
    "DSpaEvo = Dt"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "7dd943ad",
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Source image:\n"
     ]
    },
    {
     "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": [
      "Starting image:\n"
     ]
    },
    {
     "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": [
      "Adversarial Example:\n"
     ]
    },
    {
     "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": [
      "Perturbation Heat Map:\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 360x360 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "print('Source image:')\n",
    "imshow(oimg)\n",
    "\n",
    "print('Starting image:')\n",
    "timg, _ = testset[t]\n",
    "timg = torch.unsqueeze(timg, 0).cuda()\n",
    "imshow(timg)\n",
    "\n",
    "print('Adversarial Example:')\n",
    "imshow(adv)\n",
    "\n",
    "print('Perturbation Heat Map:')\n",
    "perturbation_heat_map(oimg,adv)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "802fd04e",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "- sample ID: 5013\n",
      "- ground-truth: 0 \n",
      "- target label: 5 \n",
      "- AdvEx label: 5\n",
      "- L0 dist:9\n"
     ]
    }
   ],
   "source": [
    "# check prediction of adversarial example\n",
    "alabel = model.predict_label(adv).item()\n",
    "dist = l0b(oimg,adv)#DSpaEvo[-1]\n",
    "print(f'- sample ID: {o}\\n- ground-truth: {olabel} \\n- target label: {tlabel} \\n- AdvEx label: {alabel}\\n- L0 dist:{dist}')"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "059ce735",
   "metadata": {},
   "source": [
    "### 5. Plot Sparsity vs Queries"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "6ccfcc77",
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 576x360 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "img_size = 32*32\n",
    "nq = query_limit\n",
    "q = np.arange(nq)\n",
    "df = pd.DataFrame({'qry':q,'dPW':DPw[:nq]/img_size,\n",
    "                   'dSpaE':DSpaEvo[:nq].cpu().numpy()/img_size})\n",
    "\n",
    "fig_dims = (8, 5)\n",
    "fig, ax = plt.subplots(figsize=fig_dims)\n",
    "f = sns.lineplot(x = 'qry', y = 'dPW',ax=ax,linewidth = 3, color = 'red',label = 'PW',data=df)\n",
    "f = sns.lineplot(x = 'qry', y = 'dSpaE',ax=ax,linewidth = 3, color = 'blue',label = 'SpaEvo',data=df)\n",
    "\n",
    "f.axes.set_title(\"Distortion vs Queries\",fontsize=20)\n",
    "f.set_xlabel(\"Queries\",fontsize=18)\n",
    "f.set_ylabel(\"Sparsity\",fontsize=18)\n",
    "f.tick_params(labelsize=15)\n",
    "#plt.legend(labels=['RamBoAttack (HSJA)'])\n",
    "plt.legend(title='Attack', fontsize='14', title_fontsize='18',bbox_to_anchor= (1.5,1))\n",
    "plt.show()"
   ]
  }
 ],
 "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": "3.8.12"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}