eval_test_cam_adv.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "import numpy as np\n",
    "import torch\n",
    "import torchvision\n",
    "#import foolbox\n",
    "import tqdm\n",
    "import imageio\n",
    "from torchvision import transforms\n",
    "from torch.autograd import Variable\n",
    "import torch.nn as nn\n",
    "import os\n",
    "from torch.utils.data import Dataset,DataLoader\n",
    "import skimage.transform\n",
    "import skimage\n",
    "from skimage import io\n",
    "import cv2\n",
    "from torch.nn import functional as F\n",
    "import copy"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "torch.cuda.set_device(3)\n",
    "DEVICE = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "#calcyulating the mean and the standard deviation\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "#model = torchvision.models.resnet18(pretrained=True, )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "#model.eval()\n",
    "#_=model.to(DEVICE)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "class ResnetNetwork(nn.Module):\n",
    "    def __init__(self):\n",
    "        super(ResnetNetwork, self).__init__()\n",
    "        cnn = torchvision.models.resnet152(pretrained=True)\n",
    "        #lastlayer_in = cnn.fc.in_features\n",
    "        #cnn.fc = nn.Linear(lastlayer_in, 1000)\n",
    "        self.sim_model = cnn\n",
    "    def forward(self, input1 = None):\n",
    "        output1 = self.sim_model(input1)\n",
    "        #output2 = self.fc1(output1)\n",
    "        return output1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "mean = np.array([0.5, 0.5, 0.5]).reshape((3, 1, 1))\n",
    "std = np.array([0.5, 0.5, 0.5]).reshape((3, 1, 1))\n",
    "transform = transforms.Compose(\n",
    "    [transforms.ToTensor(),\n",
    "     transforms.Normalize(mean=(0.485, 0.456, 0.406),std= (0.229, 0.224, 0.225))\n",
    "    ]\n",
    ")\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "class cam_dataset(Dataset):\n",
    "    def __init__(self,img_transform1):\n",
    "        #/workspace/data_fine_grain/cub200/raw/CUB_200_2011/(base) ironman@bigbox:~$ CUB_200_2011\n",
    "        self.dir_path = './test/'\n",
    "        self.check_path='./imagenet_fgsm/imagenet/fgsm/'\n",
    "        self.files_list = []\n",
    "        self.original_label = []\n",
    "        self.misclassified_label = []\n",
    "        self.img_transform1 = img_transform1\n",
    "        self.original_image= []\n",
    "        \n",
    "        for r, d, f in os.walk(self.dir_path):\n",
    "            for file in f:\n",
    "                if '.png' in file and \"-checkpoint.png\" not in file:\n",
    "                    #print(\"here\")\n",
    "                    #print(file)\n",
    "                    #print(file.split('.')[-2].split('_')[-2])\n",
    "                    #print(file.split('.')[-2].split('_')[-1])\n",
    "                    #if(int(file.split('.')[-2].split('_')[-2])==int(file.split('.')[-2].split('_')[-1])):\n",
    "                    #print(file)\n",
    "                    self.files_list.append(os.path.join(r,file))\n",
    "                    self.original_label.append(int(file.split('.')[-2].split('_')[-2]))\n",
    "                    self.misclassified_label.append(int(file.split('.')[-2].split('_')[-1]))\n",
    "                    self.original_image.append(os.path.join(self.check_path,file))\n",
    "                    #print(os.path.join(self.check_path,file.split('.')[-2]+'.png'))\n",
    "\n",
    "        #print(self.original_image)\n",
    "            \n",
    "    def __getitem__(self,index):\n",
    "        #if(self.train):\n",
    "            img2_path = self.original_image[index]\n",
    "            img2_path = img2_path.rstrip()\n",
    "            img2 = io.imread(img2_path)\n",
    "            \n",
    "            #print(img1.shape)\n",
    "            if(img2.shape.__len__() == 2):\n",
    "                img2 = np.repeat(img2[:, :, np.newaxis], 3, axis=2)\n",
    "            img2 = img2.astype(np.float)\n",
    "            img2 = cv2.resize(img2,(256,256),interpolation = cv2.INTER_AREA)\n",
    "            img2 = img2/255.0\n",
    "            #img2 = img2/255\n",
    "           \n",
    "            img2 = self.img_transform1(img2)\n",
    "            #print(\"check the image\")\n",
    "            #print(np.unique(img2))\n",
    "            #img2 = self.img_transform1(img2)\n",
    "        \n",
    "        \n",
    "           \n",
    "            img1_path = self.files_list[index]\n",
    "            img1_path = img1_path.rstrip()\n",
    "            img1 = io.imread(img1_path)\n",
    "            \n",
    "            #print(img1.shape)\n",
    "            if(img1.shape.__len__() == 2):\n",
    "                img1 = np.repeat(img1[:, :, np.newaxis], 3, axis=2) \n",
    "            #print(img1_path)\n",
    "            #plt.imshow(img1)\n",
    "            img1 = img1.astype(np.float)\n",
    "            img1 = cv2.resize(img1,(256,256),interpolation = cv2.INTER_AREA)\n",
    "            img1 = img1/255.0\n",
    "           \n",
    "            img1 = self.img_transform1(img1)\n",
    "            label = self.original_label[index]\n",
    "            label = int(label)\n",
    "            mislabel = self.misclassified_label[index]\n",
    "            mislabel = int(mislabel)\n",
    "            return {\"img\": img1,\"img2\":img2,\"label\": label,\"mislabel\":mislabel,\"path\": img1_path}\n",
    "   \n",
    "        #return img1,img2,label\n",
    "    def __len__(self):\n",
    "        return len(self.original_label)\n",
    "\n",
    "\n",
    "test_set = cam_dataset(transform)\n",
    "\n",
    "testloader = DataLoader(test_set,batch_size =1,shuffle = True, num_workers =0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "data = test_set[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "#print(data['img2'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "wrong\n",
      "['./test/fgsm_n04367480_25111_840_840.png']\n",
      "original_label\n",
      "463\n",
      "841\n",
      "wrong\n",
      "['./test/fgsm_n13133613_13488_998_998.png']\n",
      "original_label\n",
      "999\n",
      "85\n",
      "wrong\n",
      "['./test/fgsm_n04367480_10144_840_840.png']\n",
      "original_label\n",
      "841\n",
      "420\n",
      "wrong\n",
      "['./test/fgsm_n07711569_1558_935_935.png']\n",
      "original_label\n",
      "936\n",
      "545\n",
      "wrong\n",
      "['./test/fgsm_n01631663_4835_27_27.png']\n",
      "original_label\n",
      "28\n",
      "27\n",
      "wrong\n",
      "['./test/fgsm_n03710637_9473_638_638.png']\n",
      "original_label\n",
      "639\n",
      "579\n",
      "wrong\n",
      "['./test/fgsm_n07711569_7523_935_935.png']\n",
      "original_label\n",
      "936\n",
      "464\n",
      "wrong\n",
      "['./test/fgsm_n02102177_2388_218_218.png']\n",
      "original_label\n",
      "219\n",
      "162\n",
      "wrong\n",
      "['./test/fgsm_n02102177_5418_218_218.png']\n",
      "original_label\n",
      "219\n",
      "248\n",
      "wrong\n",
      "['./test/fgsm_n09468604_22268_979_979.png']\n",
      "original_label\n",
      "959\n",
      "526\n",
      "wrong\n",
      "['./test/fgsm_n02454379_13804_363_363.png']\n",
      "original_label\n",
      "364\n",
      "52\n",
      "wrong\n",
      "['./test/fgsm_n02102177_2695_218_218.png']\n",
      "original_label\n",
      "219\n",
      "197\n",
      "wrong\n",
      "['./test/fgsm_n02102177_5597_218_218.png']\n",
      "original_label\n",
      "219\n",
      "218\n",
      "wrong\n",
      "['./test/fgsm_n02219486_30176_310_310.png']\n",
      "original_label\n",
      "311\n",
      "76\n",
      "wrong\n",
      "['./test/fgsm_n02106662_4323_235_235.png']\n",
      "original_label\n",
      "236\n",
      "249\n",
      "wrong\n",
      "['./test/fgsm_n02422699_3313_352_352.png']\n",
      "original_label\n",
      "353\n",
      "354\n",
      "wrong\n",
      "['./test/fgsm_n12267677_8941_988_988.png']\n",
      "original_label\n",
      "989\n",
      "113\n",
      "wrong\n",
      "['./test/fgsm_n04525038_6185_885_885.png']\n",
      "original_label\n",
      "888\n",
      "974\n",
      "wrong\n",
      "['./test/fgsm_n04525038_13934_885_885.png']\n",
      "original_label\n",
      "886\n",
      "912\n",
      "wrong\n",
      "['./test/fgsm_n13133613_26355_998_998.png']\n",
      "original_label\n",
      "988\n",
      "473\n",
      "wrong\n",
      "['./test/fgsm_n02106166_1874_232_232.png']\n",
      "original_label\n",
      "233\n",
      "231\n",
      "wrong\n",
      "['./test/fgsm_n02101388_4928_215_215.png']\n",
      "original_label\n",
      "216\n",
      "213\n",
      "wrong\n",
      "['./test/fgsm_n03394916_36000_566_566.png']\n",
      "original_label\n",
      "567\n",
      "507\n",
      "wrong\n",
      "['./test/fgsm_n03782006_17741_664_664.png']\n",
      "original_label\n",
      "665\n",
      "782\n",
      "wrong\n",
      "['./test/fgsm_n09332890_14668_975_975.png']\n",
      "original_label\n",
      "976\n",
      "563\n",
      "wrong\n",
      "['./test/fgsm_n09332890_23380_975_975.png']\n",
      "original_label\n",
      "976\n",
      "448\n",
      "wrong\n",
      "['./test/fgsm_n10148035_19712_982_982.png']\n",
      "original_label\n",
      "983\n",
      "413\n",
      "wrong\n",
      "['./test/fgsm_n02948072_2134_470_470.png']\n",
      "original_label\n",
      "471\n",
      "989\n",
      "wrong\n",
      "['./test/fgsm_n09332890_30274_975_975.png']\n",
      "original_label\n",
      "976\n",
      "450\n",
      "wrong\n",
      "['./test/fgsm_n10148035_10827_982_982.png']\n",
      "original_label\n",
      "983\n",
      "841\n",
      "wrong\n",
      "['./test/fgsm_n12267677_9139_988_988.png']\n",
      "original_label\n",
      "989\n",
      "18\n",
      "wrong\n",
      "['./test/fgsm_n03584254_3186_605_605.png']\n",
      "original_label\n",
      "606\n",
      "917\n",
      "wrong\n",
      "['./test/fgsm_n07711569_4936_935_935.png']\n",
      "original_label\n",
      "936\n",
      "924\n",
      "wrong\n",
      "['./test/fgsm_n02106662_16931_235_235.png']\n",
      "original_label\n",
      "236\n",
      "899\n",
      "wrong\n",
      "['./test/fgsm_n04525038_3209_885_885.png']\n",
      "original_label\n",
      "886\n",
      "502\n",
      "wrong\n",
      "['./test/fgsm_n03876231_12569_696_696.png']\n",
      "original_label\n",
      "697\n",
      "750\n",
      "wrong\n",
      "['./test/fgsm_n02102177_4166_218_218.png']\n",
      "original_label\n",
      "219\n",
      "217\n",
      "wrong\n",
      "['./test/fgsm_n02101388_2773_215_215.png']\n",
      "original_label\n",
      "216\n",
      "243\n",
      "wrong\n",
      "['./test/fgsm_n02101388_8518_215_215.png']\n",
      "original_label\n",
      "216\n",
      "213\n",
      "wrong\n",
      "['./test/fgsm_n09468604_12333_979_979.png']\n",
      "original_label\n",
      "980\n",
      "973\n",
      "wrong\n",
      "['./test/fgsm_n02454379_5643_363_363.png']\n",
      "original_label\n",
      "364\n",
      "105\n",
      "wrong\n",
      "['./test/fgsm_n01631663_12945_27_27.png']\n",
      "original_label\n",
      "28\n",
      "43\n",
      "wrong\n",
      "['./test/fgsm_n13133613_52790_998_998.png']\n",
      "original_label\n",
      "999\n",
      "974\n",
      "wrong\n",
      "['./test/fgsm_n04254120_603_804_804.png']\n",
      "original_label\n",
      "805\n",
      "645\n",
      "wrong\n",
      "['./test/fgsm_n01828970_9816_92_92.png']\n",
      "original_label\n",
      "93\n",
      "96\n",
      "wrong\n",
      "['./test/fgsm_n01631663_4527_27_27.png']\n",
      "original_label\n",
      "28\n",
      "43\n",
      "wrong\n",
      "['./test/fgsm_n02101388_5722_215_215.png']\n",
      "original_label\n",
      "216\n",
      "853\n",
      "wrong\n",
      "['./test/fgsm_n01631663_5297_27_27.png']\n",
      "original_label\n",
      "28\n",
      "306\n",
      "wrong\n",
      "['./test/fgsm_n02102177_8153_218_218.png']\n",
      "original_label\n",
      "219\n",
      "220\n",
      "wrong\n",
      "['./test/fgsm_n13133613_17544_998_998.png']\n",
      "original_label\n",
      "999\n",
      "988\n",
      "wrong\n",
      "['./test/fgsm_n02106662_756_235_235.png']\n",
      "original_label\n",
      "236\n",
      "224\n",
      "wrong\n",
      "['./test/fgsm_n13133613_30576_998_998.png']\n",
      "original_label\n",
      "999\n",
      "988\n",
      "wrong\n",
      "['./test/fgsm_n02102177_3032_218_218.png']\n",
      "original_label\n",
      "219\n",
      "199\n",
      "wrong\n",
      "['./test/fgsm_n01664065_8558_33_33.png']\n",
      "original_label\n",
      "34\n",
      "35\n",
      "wrong\n",
      "['./test/fgsm_n04486054_10665_873_873.png']\n",
      "original_label\n",
      "874\n",
      "830\n",
      "wrong\n",
      "['./test/fgsm_n02948072_2585_470_470.png']\n",
      "original_label\n",
      "608\n",
      "502\n",
      "wrong\n",
      "['./test/fgsm_n04606251_7399_913_913.png']\n",
      "original_label\n",
      "914\n",
      "872\n",
      "wrong\n",
      "['./test/fgsm_n02939185_34856_469_469.png']\n",
      "original_label\n",
      "910\n",
      "470\n",
      "wrong\n",
      "['./test/fgsm_n01631663_7879_27_27.png']\n",
      "original_label\n",
      "28\n",
      "713\n",
      "wrong\n",
      "['./test/fgsm_n02102177_5486_218_218.png']\n",
      "original_label\n",
      "219\n",
      "236\n",
      "wrong\n",
      "['./test/fgsm_n07711569_4839_935_935.png']\n",
      "original_label\n",
      "936\n",
      "660\n",
      "wrong\n",
      "['./test/fgsm_n02106662_24524_235_235.png']\n",
      "original_label\n",
      "540\n",
      "175\n",
      "wrong\n",
      "['./test/fgsm_n02102177_6860_218_218.png']\n",
      "original_label\n",
      "219\n",
      "213\n",
      "wrong\n",
      "['./test/fgsm_n02102177_4245_218_218.png']\n",
      "original_label\n",
      "219\n",
      "248\n",
      "wrong\n",
      "['./test/fgsm_n02454379_4539_363_363.png']\n",
      "original_label\n",
      "364\n",
      "262\n",
      "wrong\n",
      "['./test/fgsm_n02102177_5635_218_218.png']\n",
      "original_label\n",
      "219\n",
      "213\n",
      "wrong\n",
      "['./test/fgsm_n02843684_21542_448_448.png']\n",
      "original_label\n",
      "449\n",
      "685\n",
      "wrong\n",
      "['./test/fgsm_n10148035_21586_982_982.png']\n",
      "original_label\n",
      "983\n",
      "645\n",
      "wrong\n",
      "['./test/fgsm_n02422699_861_352_352.png']\n",
      "original_label\n",
      "353\n",
      "350\n",
      "wrong\n",
      "['./test/fgsm_n04606251_13126_913_913.png']\n",
      "original_label\n",
      "914\n",
      "697\n",
      "wrong\n",
      "['./test/fgsm_n02445715_7341_361_361.png']\n",
      "original_label\n",
      "376\n",
      "292\n",
      "wrong\n",
      "['./test/fgsm_n02106662_20407_235_235.png']\n",
      "original_label\n",
      "236\n",
      "226\n",
      "wrong\n",
      "['./test/fgsm_n04367480_28152_840_840.png']\n",
      "original_label\n",
      "463\n",
      "841\n",
      "wrong\n",
      "['./test/fgsm_n01608432_6323_21_21.png']\n",
      "original_label\n",
      "22\n",
      "18\n",
      "wrong\n",
      "['./test/fgsm_n02106662_79892_235_235.png']\n",
      "original_label\n",
      "226\n",
      "175\n",
      "wrong\n",
      "['./test/fgsm_n01664065_30159_33_33.png']\n",
      "original_label\n",
      "34\n",
      "35\n",
      "wrong\n",
      "['./test/fgsm_n02101388_6690_215_215.png']\n",
      "original_label\n",
      "216\n",
      "213\n",
      "wrong\n",
      "['./test/fgsm_n02101388_20717_215_215.png']\n",
      "original_label\n",
      "216\n",
      "213\n",
      "wrong\n",
      "['./test/fgsm_n03404251_5627_568_568.png']\n",
      "original_label\n",
      "569\n",
      "786\n",
      "wrong\n",
      "['./test/fgsm_n01608432_13039_21_21.png']\n",
      "original_label\n",
      "22\n",
      "18\n",
      "wrong\n",
      "['./test/fgsm_n01608432_37968_21_21.png']\n",
      "original_label\n",
      "22\n",
      "18\n",
      "wrong\n",
      "['./test/fgsm_n03793489_10620_673_673.png']\n",
      "original_label\n",
      "674\n",
      "762\n",
      "wrong\n",
      "['./test/fgsm_n04254120_4824_804_804.png']\n",
      "original_label\n",
      "805\n",
      "645\n",
      "wrong\n",
      "['./test/fgsm_n01664065_4022_33_33.png']\n",
      "original_label\n",
      "34\n",
      "35\n",
      "wrong\n",
      "['./test/fgsm_n03110669_13998_513_513.png']\n",
      "original_label\n",
      "514\n",
      "855\n",
      "wrong\n",
      "['./test/fgsm_n03742115_1202_648_648.png']\n",
      "original_label\n",
      "761\n",
      "649\n",
      "wrong\n",
      "['./test/fgsm_n02102177_1524_218_218.png']\n",
      "original_label\n",
      "219\n",
      "213\n",
      "wrong\n",
      "['./test/fgsm_n02106662_2687_235_235.png']\n",
      "original_label\n",
      "236\n",
      "228\n",
      "wrong\n",
      "['./test/fgsm_n01631663_20999_27_27.png']\n",
      "original_label\n",
      "28\n",
      "27\n",
      "wrong\n",
      "['./test/fgsm_n02422699_12387_352_352.png']\n",
      "original_label\n",
      "353\n",
      "354\n",
      "wrong\n",
      "['./test/fgsm_n02102177_7931_218_218.png']\n",
      "original_label\n",
      "219\n",
      "218\n",
      "wrong\n",
      "['./test/fgsm_n09468604_66893_979_979.png']\n",
      "original_label\n",
      "980\n",
      "563\n",
      "wrong\n",
      "['./test/fgsm_n07711569_5892_935_935.png']\n",
      "original_label\n",
      "936\n",
      "568\n",
      "wrong\n",
      "['./test/fgsm_n12267677_334_988_988.png']\n",
      "original_label\n",
      "989\n",
      "680\n",
      "wrong\n",
      "['./test/fgsm_n02102177_1589_218_218.png']\n",
      "original_label\n",
      "219\n",
      "221\n",
      "wrong\n",
      "['./test/fgsm_n13133613_1376_998_998.png']\n",
      "original_label\n",
      "988\n",
      "656\n",
      "wrong\n",
      "['./test/fgsm_n02101388_10206_215_215.png']\n",
      "original_label\n",
      "216\n",
      "179\n",
      "wrong\n",
      "['./test/fgsm_n04525305_9797_886_886.png']\n",
      "original_label\n",
      "887\n",
      "599\n",
      "wrong\n",
      "['./test/fgsm_n02948072_18772_470_470.png']\n",
      "original_label\n",
      "471\n",
      "820\n",
      "wrong\n",
      "['./test/fgsm_n02219486_29218_310_310.png']\n",
      "original_label\n",
      "998\n",
      "989\n",
      "wrong\n",
      "['./test/fgsm_n02454379_4466_363_363.png']\n",
      "original_label\n",
      "364\n",
      "675\n",
      "wrong\n",
      "['./test/fgsm_n01631663_4208_27_27.png']\n",
      "original_label\n",
      "28\n",
      "43\n",
      "wrong\n",
      "['./test/fgsm_n02102177_3371_218_218.png']\n",
      "original_label\n",
      "219\n",
      "217\n",
      "wrong\n",
      "['./test/fgsm_n02948072_11450_470_470.png']\n",
      "original_label\n",
      "471\n",
      "888\n",
      "wrong\n",
      "['./test/fgsm_n02799071_6770_429_429.png']\n",
      "original_label\n",
      "430\n",
      "645\n",
      "wrong\n",
      "['./test/fgsm_n02106662_16968_235_235.png']\n",
      "original_label\n",
      "236\n",
      "225\n",
      "wrong\n",
      "['./test/fgsm_n02106662_12147_235_235.png']\n",
      "original_label\n",
      "236\n",
      "175\n",
      "wrong\n",
      "['./test/fgsm_n03110669_90528_513_513.png']\n",
      "original_label\n",
      "514\n",
      "651\n",
      "wrong\n",
      "['./test/fgsm_n02101388_9771_215_215.png']\n",
      "original_label\n",
      "216\n",
      "342\n",
      "wrong\n",
      "['./test/fgsm_n02106662_20546_235_235.png']\n",
      "original_label\n",
      "236\n",
      "251\n",
      "wrong\n",
      "['./test/fgsm_n02799071_25705_429_429.png']\n",
      "original_label\n",
      "430\n",
      "989\n",
      "wrong\n",
      "['./test/fgsm_n02799071_31326_429_429.png']\n",
      "original_label\n",
      "430\n",
      "982\n",
      "wrong\n",
      "['./test/fgsm_n09468604_19698_979_979.png']\n",
      "original_label\n",
      "980\n",
      "526\n",
      "wrong\n",
      "['./test/fgsm_n13133613_22014_998_998.png']\n",
      "original_label\n",
      "999\n",
      "530\n",
      "wrong\n",
      "['./test/fgsm_n01608432_10700_21_21.png']\n",
      "original_label\n",
      "22\n",
      "18\n",
      "wrong\n",
      "['./test/fgsm_n01644373_11474_31_31.png']\n",
      "original_label\n",
      "32\n",
      "670\n",
      "wrong\n",
      "['./test/fgsm_n01828970_4879_92_92.png']\n",
      "original_label\n",
      "92\n",
      "93\n",
      "wrong\n",
      "['./test/fgsm_n02948072_14855_470_470.png']\n",
      "original_label\n",
      "471\n",
      "929\n",
      "wrong\n",
      "['./test/fgsm_n04254120_1187_804_804.png']\n",
      "original_label\n",
      "805\n",
      "645\n",
      "wrong\n",
      "['./test/fgsm_n02102177_4170_218_218.png']\n",
      "original_label\n",
      "219\n",
      "384\n",
      "wrong\n",
      "['./test/fgsm_n01828970_22691_92_92.png']\n",
      "original_label\n",
      "93\n",
      "85\n",
      "wrong\n",
      "['./test/fgsm_n03782006_15770_664_664.png']\n",
      "original_label\n",
      "528\n",
      "665\n",
      "wrong\n",
      "['./test/fgsm_n02804610_19011_432_432.png']\n",
      "original_label\n",
      "433\n",
      "765\n",
      "wrong\n",
      "['./test/fgsm_n02101388_2887_215_215.png']\n",
      "original_label\n",
      "216\n",
      "218\n",
      "wrong\n",
      "['./test/fgsm_n03404251_4230_568_568.png']\n",
      "original_label\n",
      "569\n",
      "870\n",
      "wrong\n",
      "['./test/fgsm_n02219486_30703_310_310.png']\n",
      "original_label\n",
      "311\n",
      "108\n",
      "wrong\n",
      "['./test/fgsm_n02101388_4276_215_215.png']\n",
      "original_label\n",
      "216\n",
      "170\n",
      "wrong\n",
      "['./test/fgsm_n03924679_13361_713_713.png']\n",
      "original_label\n",
      "743\n",
      "621\n",
      "wrong\n",
      "['./test/fgsm_n01664065_29111_33_33.png']\n",
      "original_label\n",
      "34\n",
      "37\n",
      "wrong\n",
      "['./test/fgsm_n09332890_21158_975_975.png']\n",
      "original_label\n",
      "976\n",
      "913\n",
      "wrong\n",
      "['./test/fgsm_n02804610_9399_432_432.png']\n",
      "original_label\n",
      "433\n",
      "419\n",
      "wrong\n",
      "['./test/fgsm_n02101388_1145_215_215.png']\n",
      "original_label\n",
      "216\n",
      "170\n",
      "wrong\n",
      "['./test/fgsm_n02102177_5828_218_218.png']\n",
      "original_label\n",
      "219\n",
      "213\n",
      "wrong\n",
      "['./test/fgsm_n09332890_33293_975_975.png']\n",
      "original_label\n",
      "976\n",
      "978\n",
      "wrong\n",
      "['./test/fgsm_n01631663_2572_27_27.png']\n",
      "original_label\n",
      "28\n",
      "391\n",
      "wrong\n",
      "['./test/fgsm_n02219486_29150_310_310.png']\n",
      "original_label\n",
      "311\n",
      "306\n",
      "wrong\n",
      "['./test/fgsm_n04553703_1025_896_896.png']\n",
      "original_label\n",
      "877\n",
      "897\n",
      "wrong\n",
      "['./test/fgsm_n01608432_1097_21_21.png']\n",
      "original_label\n",
      "22\n",
      "132\n",
      "wrong\n",
      "['./test/fgsm_n01631663_5761_27_27.png']\n",
      "original_label\n",
      "28\n",
      "27\n",
      "wrong\n",
      "['./test/fgsm_n02106166_1918_232_232.png']\n",
      "original_label\n",
      "233\n",
      "236\n",
      "wrong\n",
      "['./test/fgsm_n01631663_2387_27_27.png']\n",
      "original_label\n",
      "28\n",
      "27\n",
      "wrong\n",
      "['./test/fgsm_n13133613_79556_998_998.png']\n",
      "original_label\n",
      "988\n",
      "846\n",
      "wrong\n",
      "['./test/fgsm_n02422699_22493_352_352.png']\n",
      "original_label\n",
      "353\n",
      "354\n",
      "wrong\n",
      "['./test/fgsm_n04525038_8603_885_885.png']\n",
      "original_label\n",
      "825\n",
      "974\n",
      "wrong\n",
      "['./test/fgsm_n13133613_55951_998_998.png']\n",
      "original_label\n",
      "336\n",
      "378\n",
      "wrong\n",
      "['./test/fgsm_n02102177_4519_218_218.png']\n",
      "original_label\n",
      "219\n",
      "217\n",
      "wrong\n",
      "['./test/fgsm_n10148035_29758_982_982.png']\n",
      "original_label\n",
      "983\n",
      "571\n",
      "wrong\n",
      "['./test/fgsm_n09468604_7567_979_979.png']\n",
      "original_label\n",
      "980\n",
      "971\n",
      "wrong\n",
      "['./test/fgsm_n03924679_6991_713_713.png']\n",
      "original_label\n",
      "714\n",
      "422\n",
      "wrong\n",
      "['./test/fgsm_n04525305_11312_886_886.png']\n",
      "original_label\n",
      "887\n",
      "645\n",
      "wrong\n",
      "['./test/fgsm_n01631663_8371_27_27.png']\n",
      "original_label\n",
      "28\n",
      "43\n",
      "wrong\n",
      "['./test/fgsm_n02797295_8850_428_428.png']\n",
      "original_label\n",
      "767\n",
      "429\n",
      "wrong\n",
      "['./test/fgsm_n02102177_6888_218_218.png']\n",
      "original_label\n",
      "219\n",
      "170\n",
      "wrong\n",
      "['./test/fgsm_n03680355_1454_630_630.png']\n",
      "original_label\n",
      "631\n",
      "474\n",
      "wrong\n",
      "['./test/fgsm_n02102177_4260_218_218.png']\n",
      "original_label\n",
      "219\n",
      "213\n",
      "wrong\n",
      "['./test/fgsm_n01608432_13976_21_21.png']\n",
      "original_label\n",
      "22\n",
      "24\n",
      "wrong\n",
      "['./test/fgsm_n02101388_9410_215_215.png']\n",
      "original_label\n",
      "216\n",
      "213\n",
      "wrong\n",
      "['./test/fgsm_n09332890_32827_975_975.png']\n",
      "original_label\n",
      "976\n",
      "571\n",
      "wrong\n",
      "['./test/fgsm_n01631663_12063_27_27.png']\n",
      "original_label\n",
      "28\n",
      "15\n",
      "wrong\n",
      "['./test/fgsm_n04525305_10609_886_886.png']\n",
      "original_label\n",
      "887\n",
      "645\n",
      "wrong\n",
      "['./test/fgsm_n02106166_11699_232_232.png']\n",
      "original_label\n",
      "265\n",
      "173\n",
      "wrong\n",
      "['./test/fgsm_n07711569_4888_935_935.png']\n",
      "original_label\n",
      "924\n",
      "869\n",
      "wrong\n",
      "['./test/fgsm_n03876231_19828_696_696.png']\n",
      "original_label\n",
      "697\n",
      "744\n",
      "wrong\n",
      "['./test/fgsm_n02939185_19353_469_469.png']\n",
      "original_label\n",
      "470\n",
      "464\n",
      "the final wrongs\n",
      "164\n",
      "total\n",
      "1000\n"
     ]
    }
   ],
   "source": [
    "#plt.imshow(data[\"img\"])\n",
    "list_image = []\n",
    "list_label = []\n",
    "counter = 0\n",
    "act_counter =0\n",
    "for data in testloader:\n",
    "    #net1 = net1.to(DEVICE)\n",
    "    #net2 = net2.to(DEVICE)\n",
    "    #####################actual image\n",
    "    img2 = data['img2'].to(DEVICE)\n",
    "    #path = data['path']\n",
    "    img2 = img2.view(-1,img2.shape[-3],img2.shape[-2],img2.shape[-1])\n",
    "\n",
    "    img2 = img2.float()\n",
    "    resnet2 = ResnetNetwork()\n",
    "    resnet2 = resnet2.to(DEVICE)\n",
    "    resnet2.eval()\n",
    "    img2 = Variable(img2.cuda(), requires_grad=True)\n",
    "    pred2 = resnet2(img2)\n",
    "    max_index2 = pred2.argmax(dim=1) \n",
    "    \n",
    "    \n",
    "    #print(\"check the original max index\")\n",
    "    #print(max_index2.item())\n",
    "    list_image.append(data['path'])\n",
    "    list_label.append(max_index2.item())\n",
    "    #print(\"check the label\")\n",
    "    #print(label.item())\n",
    "    \n",
    "    #########image\n",
    "    img1 = data['img'].to(DEVICE)\n",
    "    path = data['path']\n",
    "    img1 = img1.view(-1,img1.shape[-3],img1.shape[-2],img1.shape[-1])\n",
    "\n",
    "    img1 = img1.float()\n",
    "    \n",
    "    \n",
    "    #print(img1.shape)\n",
    "    label = data['label'].to(DEVICE)\n",
    "    mislabel = data['mislabel'].to(DEVICE)\n",
    "   \n",
    "    resnet =  ResnetNetwork()\n",
    "    resnet = resnet.to(DEVICE)\n",
    "    resnet.eval()\n",
    "    #print(\"check the image sizes\")\n",
    "    #print(img1.shape)\n",
    "    img1 = Variable(img1.cuda(), requires_grad=True)\n",
    "    pred = resnet(img1)\n",
    "    max_index = pred.argmax(dim=1)\n",
    "    val =nn.Softmax(dim=1)(pred)\n",
    "    \n",
    "    #print(\"check the max index of adv image\")\n",
    "    #print(max_index.item())\n",
    "    #print(\"prob\")\n",
    "    #print(val.shape)\n",
    "    values, index = torch.topk(val,1)\n",
    "    #print(values.item())\n",
    "    #print(index)\n",
    "    #print(\"check the label\")\n",
    "    #print(label.item())\n",
    "    #print(path)\n",
    "    #break\n",
    "    act_counter+=1\n",
    "    if(max_index.item()!=max_index2.item()):\n",
    "        print(\"wrong\")\n",
    "        print(path)\n",
    "        print(\"original_label\")\n",
    "        print(max_index2.item()+1)\n",
    "        print(max_index.item()+1)\n",
    "        counter +=1\n",
    "print(\"the final wrongs\")\n",
    "print(counter)\n",
    "print(\"total\")\n",
    "print(act_counter)\n",
    "\n",
    "\n",
    "   \n",
    "    \n",
    "  #print(\"the \")  \n",
    "    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'\\nheatmap = np.uint8(255 * heatmap)\\nheatmap = cv2.applyColorMap(heatmap, cv2.COLORMAP_JET)\\nimg = heatmap * 0.4 + img\\nname = path[0].split(\\'/\\')[-1].split(\\'.\\')[-2]\\nprint(\"check the name\")\\nprint(name)\\n\\ncv2.imwrite(\\'./test1/\\'+str(name)+\\'.jpg\\', img)\\n#plt.imshow(img)\\n#break\\n#########################################################################for label\\n   '"
      ]
     },
     "execution_count": 45,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "    '''\n",
    "    ##############################################################for max_index\n",
    "    pred[:,(max_index.item()-1)].backward()\n",
    "    gradients = resnet.get_gradient()\n",
    "    pooled_gradients = torch.mean(gradients, dim=[0, 2, 3])\n",
    "    activations = resnet.get_activations(img1).detach()\n",
    "    for i in range(activations.shape[1]):\n",
    "        activations[:, i, :, :] *= pooled_gradients[i]\n",
    "    heatmap = F.relu(torch.sum(activations, dim=1)).squeeze()\n",
    "    heatmap = np.maximum(heatmap.cpu(), 0)\n",
    "    heatmap /= torch.max(heatmap)\n",
    "    #plt.matshow(heatmap.squeeze())\n",
    "    heatmap = heatmap.numpy()\n",
    "    #print(np.unique(heatmap))\n",
    "    \n",
    "    img = io.imread(path[0])\n",
    "    if(img.shape.__len__() == 2):\n",
    "                img = np.repeat(img[:, :, np.newaxis], 3, axis=2) \n",
    "    heatmap = cv2.resize(heatmap, (img.shape[1], img.shape[0]))\n",
    "    '''\n",
    "    '''\n",
    "    p = np.percentile(heatmap, 50)\n",
    "    heatmap[heatmap>=p]=1\n",
    "    heatmap[heatmap<p]=0\n",
    "    heatmap = heatmap.astype(int)\n",
    "    x = copy.deepcopy(heatmap)\n",
    "    r_temp = img[:,:,0]*(x*(-1))\n",
    "    img[:,:,0] = img[:,:,0] +r_temp +(0*x)\n",
    "    g_temp = img[:,:,1]*(x*(-1))\n",
    "    img[:,:,1] = img[:,:,1] +g_temp +(0*x)\n",
    "    b_temp = img[:,:,2]*(x*(-1))\n",
    "    img[:,:,2] = img[:,:,2] +b_temp +(0*x)\n",
    "    '''\n",
    "    '''\n",
    "    heatmap = np.uint8(255 * heatmap)\n",
    "    heatmap = cv2.applyColorMap(heatmap, cv2.COLORMAP_JET)\n",
    "    img = heatmap * 0.4 + img\n",
    "    name = path[0].split('/')[-1].split('.')[-2]\n",
    "    print(\"check the name\")\n",
    "    print(name)\n",
    "    \n",
    "    cv2.imwrite('./test1/'+str(name)+'.jpg', img)\n",
    "    #plt.imshow(img)\n",
    "    #break\n",
    "    #########################################################################for label\n",
    "   '''"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "for data in testloader:\n",
    "    #net1 = net1.to(DEVICE)\n",
    "    #net2 = net2.to(DEVICE)\n",
    "    img1 = data['img'].to(DEVICE)\n",
    "    print(\"check the image\")\n",
    "    print(img1.shape)\n",
    "    path = data['path']\n",
    "\n",
    "    #img1 = img1.float()\n",
    "    img1 = img1.view(-1,img1.shape[-3],img1.shape[-2],img1.shape[-1])\n",
    "    print(img1.shape)\n",
    "    img1 = img1.float()\n",
    "    #print(img1.shape)\n",
    "    label = data['label'].to(DEVICE)\n",
    "    print(\"CHECK THE PATH\")\n",
    "    print(path[0])   \n",
    "    resnet = ResNet()\n",
    "    resnet = resnet.to(DEVICE)\n",
    "    resnet.eval()\n",
    "    print(\"check the image sizes\")\n",
    "    print(img1.shape)\n",
    "    img1 = Variable(img1.cuda(), requires_grad=True)\n",
    "    pred, acts = resnet(img1)\n",
    "    max_index = pred.argmax(dim=1) \n",
    "    print(\"check the max index\")\n",
    "    print(max_index)\n",
    "    pred[:,(max_index.item())].backward()\n",
    "    gradients = resnet.get_gradient()\n",
    "    pooled_gradients = torch.mean(gradients, dim=[0, 2, 3])\n",
    "    activations = resnet.get_activations(img1).detach()\n",
    "    for i in range(activations.shape[1]):\n",
    "        activations[:, i, :, :] *= pooled_gradients[i]\n",
    "    heatmap = F.relu(torch.sum(activations, dim=1)).squeeze()\n",
    "    heatmap = np.maximum(heatmap.cpu(), 0)\n",
    "    heatmap /= torch.max(heatmap)\n",
    "    #plt.matshow(heatmap.squeeze())\n",
    "    heatmap = heatmap.numpy()\n",
    "    #print(np.unique(heatmap))\n",
    "    \n",
    "    print(np.unique(heatmap))\n",
    "    \n",
    "    print(label.item())\n",
    "    img = io.imread(path[0])\n",
    "    heatmap = cv2.resize(heatmap, (img.shape[1], img.shape[0]))\n",
    "    p = np.percentile(heatmap, 70)\n",
    "    heatmap[heatmap>=p]=1\n",
    "    heatmap[heatmap<p]=0\n",
    "    heatmap = np.uint8(255 * heatmap)\n",
    "    heatmap = cv2.applyColorMap(heatmap, cv2.COLORMAP_JET)\n",
    "    superimposed_img = heatmap * 0.4 + img\n",
    "    name = path[0].split('/')[-1].split('.')[-2]\n",
    "    cv2.imwrite('./originallabel_camfiles/'+str(name)+'_'+str(max_index.item()+1)+'.jpg', superimposed_img)\n",
    "    "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#plt.imshow(data[\"img\"])\n",
    "for data in testloader:\n",
    "    #net1 = net1.to(DEVICE)\n",
    "    #net2 = net2.to(DEVICE)\n",
    "    img1 = data['img'].to(DEVICE)\n",
    "    print(\"check the image\")\n",
    "    print(img1.shape)\n",
    "    path = data['path']\n",
    "\n",
    "    #img1 = img1.float()\n",
    "    img1 = img1.view(-1,img1.shape[-3],img1.shape[-2],img1.shape[-1])\n",
    "    print(img1.shape)\n",
    "    img1 = img1.float()\n",
    "    #print(img1.shape)\n",
    "    label = data['label'].to(DEVICE)\n",
    "    mislabel = data['mislabel'].to(DEVICE)\n",
    "   \n",
    "    resnet = ResNet()\n",
    "    resnet = resnet.to(DEVICE)\n",
    "    resnet.eval()\n",
    "    print(\"check the image sizes\")\n",
    "    print(img1.shape)\n",
    "    img1 = Variable(img1.cuda(), requires_grad=True)\n",
    "    pred, acts = resnet(img1)\n",
    "    max_index = pred.argmax(dim=1) \n",
    "    print(\"check the max index\")\n",
    "    print(max_index)\n",
    "    ##############################################################for actual label\n",
    "    pred[:,(label.item())].backward()\n",
    "    gradients = resnet.get_gradient()\n",
    "    pooled_gradients = torch.mean(gradients, dim=[0, 2, 3])\n",
    "    activations = resnet.get_activations(img1).detach()\n",
    "    for i in range(activations.shape[1]):\n",
    "        activations[:, i, :, :] *= pooled_gradients[i]\n",
    "    heatmap = F.relu(torch.sum(activations, dim=1)).squeeze()\n",
    "    heatmap = np.maximum(heatmap.cpu(), 0)\n",
    "    heatmap /= torch.max(heatmap)\n",
    "    #plt.matshow(heatmap.squeeze())\n",
    "    heatmap = heatmap.numpy()\n",
    "    #print(np.unique(heatmap))\n",
    "    p = np.percentile(heatmap, 97)\n",
    "    heatmap[heatmap>=p]=1\n",
    "    heatmap[heatmap<p]=0\n",
    "    print(np.unique(heatmap))\n",
    "    print(\"CHECK THE PATH\")\n",
    "    print(path[0])\n",
    "    print(label.item())\n",
    "    img = io.imread(path[0])\n",
    "    heatmap = cv2.resize(heatmap, (img.shape[1], img.shape[0]))\n",
    "    heatmap = np.uint8(255 * heatmap)\n",
    "    heatmap = cv2.applyColorMap(heatmap, cv2.COLORMAP_JET)\n",
    "    superimposed_img = heatmap * 0.4 + img\n",
    "    name = path[0].split('/')[-1].split('.')[-2]\n",
    "    print(\"check the name\")\n",
    "    print(name)\n",
    "    \n",
    "    cv2.imwrite('./originallabel_camfiles/'+str(name)+'_'+str(label.item())+'.jpg', superimposed_img)\n",
    "    #########################################################################for label\n",
    "   "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "print(list_image)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 297,
   "metadata": {},
   "outputs": [],
   "source": [
    "with open('./actualimagefile.txt', 'w') as filehandle:\n",
    "    for i in range(len(list_image)):\n",
    "        filehandle.write('%s|%s\\n' % (list_image[i][0],list_label[i]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "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
}