4_create_datasets.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import pickle\n",
    "import pandas as pd\n",
    "\n",
    "from PIL import Image\n",
    "from pathlib import Path\n",
    "from torchvision import transforms\n",
    "from tqdm import tqdm\n",
    "\n",
    "from utils.ScrambleImage import ScrambleImage, channels_first, channels_last"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Create datasets"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "save_folder = Path(\"data/\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Load `pickle` file with the images that are in the training dataset of regular ImageNet"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "with open(save_folder / 'imgn_pict.pkl', 'rb') as f:\n",
    "    imgn_pict = pickle.load(f)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "4312"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(imgn_pict)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The classes we are going to analyze"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'n01440764',\n",
       " 'n02102040',\n",
       " 'n02979186',\n",
       " 'n03000684',\n",
       " 'n03028079',\n",
       " 'n03394916',\n",
       " 'n03417042',\n",
       " 'n03425413',\n",
       " 'n03445777',\n",
       " 'n03888257'}"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "imgn_folders = set(img_path.parent.stem for img_path in imgn_pict)\n",
    "imgn_folders"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Create a list of dictionaries that combine the path and the class of a file"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[{'class': 'n03394916',\n",
       "  'path': PosixPath('data/imagenette2/train/n03394916/n03394916_23108.JPEG')},\n",
       " {'class': 'n03394916',\n",
       "  'path': PosixPath('data/imagenette2/train/n03394916/n03394916_59626.JPEG')}]"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "imgn_class = [{\"class\":img_path.parent.stem, \"path\":img_path} for img_path in imgn_pict]\n",
    "imgn_class[:2]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "For convenience sake let's create a dataframe"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>class</th>\n",
       "      <th>path</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>n03394916</td>\n",
       "      <td>data/imagenette2/train/n03394916/n03394916_231...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>n03394916</td>\n",
       "      <td>data/imagenette2/train/n03394916/n03394916_596...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>n03394916</td>\n",
       "      <td>data/imagenette2/train/n03394916/n03394916_552...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>n03394916</td>\n",
       "      <td>data/imagenette2/train/n03394916/n03394916_329...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>n03394916</td>\n",
       "      <td>data/imagenette2/train/n03394916/n03394916_619...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "       class                                               path\n",
       "0  n03394916  data/imagenette2/train/n03394916/n03394916_231...\n",
       "1  n03394916  data/imagenette2/train/n03394916/n03394916_596...\n",
       "2  n03394916  data/imagenette2/train/n03394916/n03394916_552...\n",
       "3  n03394916  data/imagenette2/train/n03394916/n03394916_329...\n",
       "4  n03394916  data/imagenette2/train/n03394916/n03394916_619..."
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "data = pd.DataFrame(imgn_class)\n",
    "data[:5]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Randomly choose 3 images per class"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>class</th>\n",
       "      <th>path</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>n01440764</td>\n",
       "      <td>/home/malte/Dokumente/Masterarbeit/data/imagen...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>n01440764</td>\n",
       "      <td>/home/malte/Dokumente/Masterarbeit/data/imagen...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>n01440764</td>\n",
       "      <td>/home/malte/Dokumente/Masterarbeit/data/imagen...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>n02102040</td>\n",
       "      <td>/home/malte/Dokumente/Masterarbeit/data/imagen...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>n02102040</td>\n",
       "      <td>/home/malte/Dokumente/Masterarbeit/data/imagen...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "       class                                               path\n",
       "0  n01440764  /home/malte/Dokumente/Masterarbeit/data/imagen...\n",
       "1  n01440764  /home/malte/Dokumente/Masterarbeit/data/imagen...\n",
       "2  n01440764  /home/malte/Dokumente/Masterarbeit/data/imagen...\n",
       "3  n02102040  /home/malte/Dokumente/Masterarbeit/data/imagen...\n",
       "4  n02102040  /home/malte/Dokumente/Masterarbeit/data/imagen..."
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "grouped = data.groupby(\"class\")\n",
    "grouped = grouped.apply(lambda x: x.sample(n=3, random_state=1))\n",
    "\n",
    "grouped.reset_index(level=0, drop=True, inplace=True)\n",
    "grouped.reset_index(drop=True, inplace=True)\n",
    "grouped[:5]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "____________"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 1. Create scrambled image files"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Create top folder"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "folder_scr = save_folder / \"imagenette2_scr\"\n",
    "folder_scr.mkdir(exist_ok=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Create folder for each class"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 118,
   "metadata": {},
   "outputs": [],
   "source": [
    "for img_class in imgn_folders:\n",
    "    folder_scr_class = folder_scr / img_class\n",
    "    folder_scr_class.mkdir(exist_ok=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Resize, crop and normalize image accoring to standard ImageNet values"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "transform = transforms.Compose([\n",
    " transforms.Resize(256),\n",
    " transforms.CenterCrop(224),\n",
    "])\n",
    "\n",
    "transform_normalize = transforms.Normalize(\n",
    "     mean=[0.485, 0.456, 0.406],\n",
    "     std=[0.229, 0.224, 0.225]\n",
    " )"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "How many different variants of patches could be created?"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 120,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "We can use size '1' and have '50176.0' equal patches\n",
      "We can use size '2' and have '12544.0' equal patches\n",
      "We can use size '4' and have '3136.0' equal patches\n",
      "We can use size '7' and have '1024.0' equal patches\n",
      "We can use size '8' and have '784.0' equal patches\n",
      "We can use size '14' and have '256.0' equal patches\n",
      "We can use size '16' and have '196.0' equal patches\n",
      "We can use size '28' and have '64.0' equal patches\n",
      "We can use size '32' and have '49.0' equal patches\n",
      "We can use size '56' and have '16.0' equal patches\n",
      "We can use size '112' and have '4.0' equal patches\n"
     ]
    }
   ],
   "source": [
    "for i in range(1, 113):\n",
    "    if 224%i == 0:\n",
    "        devide = (224 / i)**2\n",
    "        print(f\"We can use size '{i}' and have '{devide}' equal patches\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Let's choose 4 distinct values for the block size"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 121,
   "metadata": {},
   "outputs": [],
   "source": [
    "block_sz = [28, 32, 56, 112]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 122,
   "metadata": {},
   "outputs": [],
   "source": [
    "var = list(range(0,3))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Create dataset"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Create new column for the indices to store the order of the scrambled files"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 123,
   "metadata": {},
   "outputs": [],
   "source": [
    "grouped[\"scrambled_indices\"] = None"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Loop over the dataframe index and create a scrambled image for each `Path` object. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 124,
   "metadata": {},
   "outputs": [],
   "source": [
    "for index, _class, path, indices in tqdm(grouped.itertuples()):\n",
    "    # load image\n",
    "    img = Image.open(path)\n",
    "    # apply transformation\n",
    "    trf_img = transform(img)\n",
    "    # convert to numpy array\n",
    "    trf_arr = np.array(trf_img)\n",
    "    # convert to channels first\n",
    "    trf_arr = channels_first(trf_arr)\n",
    "    \n",
    "    # also save unmodified image\n",
    "    save_path = folder_scr / f\"{_class}/{path.stem}_original.jpg\"\n",
    "    trf_img.save(save_path)\n",
    "    \n",
    "    for block_size in block_sz:\n",
    "        temp_ind = {}\n",
    "        \n",
    "        for variant in var:\n",
    "            scr = ScrambleImage(trf_arr, block_size)\n",
    "            scr.scramble()\n",
    "            ind, scr_arr = scr.export()\n",
    "            temp_ind[variant] = ind\n",
    "            \n",
    "            scr_arr = channels_last(scr_arr)\n",
    "            scr_img = Image.fromarray(scr_arr)\n",
    "            save_path = folder_scr / f\"{_class}/{path.stem}_{block_size}_{variant}.jpg\"\n",
    "            \n",
    "            scr_img.save(save_path)\n",
    "        \n",
    "        grouped.at[index, \"scrambled_indices\"] = temp_ind"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 125,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>class</th>\n",
       "      <th>path</th>\n",
       "      <th>scrambled_indices</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>n01440764</td>\n",
       "      <td>/home/malte/Dokumente/Masterarbeit/data/imagen...</td>\n",
       "      <td>{0: [2, 3, 1, 0], 1: [0, 3, 1, 2], 2: [3, 0, 1...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>n01440764</td>\n",
       "      <td>/home/malte/Dokumente/Masterarbeit/data/imagen...</td>\n",
       "      <td>{0: [1, 3, 0, 2], 1: [1, 3, 0, 2], 2: [2, 3, 1...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>n01440764</td>\n",
       "      <td>/home/malte/Dokumente/Masterarbeit/data/imagen...</td>\n",
       "      <td>{0: [0, 1, 2, 3], 1: [1, 3, 0, 2], 2: [2, 1, 0...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>n02102040</td>\n",
       "      <td>/home/malte/Dokumente/Masterarbeit/data/imagen...</td>\n",
       "      <td>{0: [2, 0, 1, 3], 1: [0, 2, 3, 1], 2: [2, 1, 3...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>n02102040</td>\n",
       "      <td>/home/malte/Dokumente/Masterarbeit/data/imagen...</td>\n",
       "      <td>{0: [1, 3, 2, 0], 1: [2, 3, 0, 1], 2: [3, 0, 2...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "       class                                               path  \\\n",
       "0  n01440764  /home/malte/Dokumente/Masterarbeit/data/imagen...   \n",
       "1  n01440764  /home/malte/Dokumente/Masterarbeit/data/imagen...   \n",
       "2  n01440764  /home/malte/Dokumente/Masterarbeit/data/imagen...   \n",
       "3  n02102040  /home/malte/Dokumente/Masterarbeit/data/imagen...   \n",
       "4  n02102040  /home/malte/Dokumente/Masterarbeit/data/imagen...   \n",
       "\n",
       "                                   scrambled_indices  \n",
       "0  {0: [2, 3, 1, 0], 1: [0, 3, 1, 2], 2: [3, 0, 1...  \n",
       "1  {0: [1, 3, 0, 2], 1: [1, 3, 0, 2], 2: [2, 3, 1...  \n",
       "2  {0: [0, 1, 2, 3], 1: [1, 3, 0, 2], 2: [2, 1, 0...  \n",
       "3  {0: [2, 0, 1, 3], 1: [0, 2, 3, 1], 2: [2, 1, 3...  \n",
       "4  {0: [1, 3, 2, 0], 1: [2, 3, 0, 1], 2: [3, 0, 2...  "
      ]
     },
     "execution_count": 125,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "grouped[:5]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Save dataset to disk"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 126,
   "metadata": {},
   "outputs": [],
   "source": [
    "grouped.to_csv(folder_scr / \"scramble.csv\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "____"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 2.Create patches dataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "from utils.PatchImage import PatchImage"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Load `scramble.csv` to get the sampled image paths"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>class</th>\n",
       "      <th>path</th>\n",
       "      <th>scrambled_indices</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>n01440764</td>\n",
       "      <td>/home/malte/Dokumente/Masterarbeit/data/imagen...</td>\n",
       "      <td>{0: array([2, 3, 1, 0]), 1: array([0, 3, 1, 2]...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>n01440764</td>\n",
       "      <td>/home/malte/Dokumente/Masterarbeit/data/imagen...</td>\n",
       "      <td>{0: array([1, 3, 0, 2]), 1: array([1, 3, 0, 2]...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>n01440764</td>\n",
       "      <td>/home/malte/Dokumente/Masterarbeit/data/imagen...</td>\n",
       "      <td>{0: array([0, 1, 2, 3]), 1: array([1, 3, 0, 2]...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>n02102040</td>\n",
       "      <td>/home/malte/Dokumente/Masterarbeit/data/imagen...</td>\n",
       "      <td>{0: array([2, 0, 1, 3]), 1: array([0, 2, 3, 1]...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>n02102040</td>\n",
       "      <td>/home/malte/Dokumente/Masterarbeit/data/imagen...</td>\n",
       "      <td>{0: array([1, 3, 2, 0]), 1: array([2, 3, 0, 1]...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "       class                                               path  \\\n",
       "0  n01440764  /home/malte/Dokumente/Masterarbeit/data/imagen...   \n",
       "1  n01440764  /home/malte/Dokumente/Masterarbeit/data/imagen...   \n",
       "2  n01440764  /home/malte/Dokumente/Masterarbeit/data/imagen...   \n",
       "3  n02102040  /home/malte/Dokumente/Masterarbeit/data/imagen...   \n",
       "4  n02102040  /home/malte/Dokumente/Masterarbeit/data/imagen...   \n",
       "\n",
       "                                   scrambled_indices  \n",
       "0  {0: array([2, 3, 1, 0]), 1: array([0, 3, 1, 2]...  \n",
       "1  {0: array([1, 3, 0, 2]), 1: array([1, 3, 0, 2]...  \n",
       "2  {0: array([0, 1, 2, 3]), 1: array([1, 3, 0, 2]...  \n",
       "3  {0: array([2, 0, 1, 3]), 1: array([0, 2, 3, 1]...  \n",
       "4  {0: array([1, 3, 2, 0]), 1: array([2, 3, 0, 1]...  "
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "grouped = pd.read_csv(folder_scr / \"scramble.csv\", index_col=0)\n",
    "grouped[:5]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "grouped[\"path\"] = grouped[\"path\"].apply(lambda x: Path(x))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Create top folder"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "folder_ptd = save_folder / \"imagenette2_ptd\"\n",
    "folder_ptd.mkdir(exist_ok=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Create folder for each class"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "for img_class in imgn_folders:\n",
    "    folder_ptd_class = folder_ptd / img_class\n",
    "    folder_ptd_class.mkdir(exist_ok=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Define the patch sizes that are applied to each image"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "patch_sizes = range(10, 80, 10)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Create the dataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "30it [00:16,  1.79it/s]\n"
     ]
    }
   ],
   "source": [
    "for index, _class, path, indices in tqdm(grouped.itertuples()):\n",
    "    # load image\n",
    "    img = Image.open(path)\n",
    "    # apply transformation\n",
    "    trf_img = transform(img)\n",
    "    \n",
    "    # also save unmodified image\n",
    "    save_path = folder_ptd / f\"{_class}/{path.stem}_original.jpg\"\n",
    "    trf_img.save(save_path)\n",
    "    \n",
    "    for patch_size in patch_sizes:\n",
    "        patch_img = PatchImage(trf_img, patch_size)\n",
    "        \n",
    "        for i, patch in enumerate(patch_img.patch()):\n",
    "            save_path = folder_ptd / f\"{_class}/{path.stem}_{patch_size}_{i}.jpg\"\n",
    "            patch.save(save_path)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "____"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 3.Create smoothed dataset"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "For this process to work, we need some additional files from [this](https://github.com/zhufeida/Benchmark_EPS) repository"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "! wget -q -c https://raw.githubusercontent.com/zhufeida/Benchmark_EPS/master/Resnet_tl/module.py -P utils/ "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "! wget -q https://github.com/zhufeida/Benchmark_EPS/raw/master/Resnet_tl/checkpoint_delta/model-345002.data-00000-of-00001 -P data/checkpoint_delta"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "! wget -q https://github.com/zhufeida/Benchmark_EPS/raw/master/Resnet_tl/checkpoint_delta/model-345002.index -P data/checkpoint_delta"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "! wget -q https://github.com/zhufeida/Benchmark_EPS/raw/master/Resnet_tl/checkpoint_delta/model-345002.meta -P data/checkpoint_delta"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Also, it might be easier to get `tensorlayer` from [the repository](https://github.com/zhufeida/Benchmark_EPS/tree/master/Resnet_tl) directly"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "import tensorflow as tf\n",
    "import numpy as np\n",
    "import imageio\n",
    "\n",
    "from utils.module import SRGAN_delta\n",
    "from PIL import Image\n",
    "from PIL import ImageChops\n",
    "from PIL import ImageOps"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### 1. Build the `tensorflow` graph"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:From /home/malte/anaconda3/envs/kerasCPU/lib/python3.7/site-packages/tensorflow/python/framework/op_def_library.py:263: colocate_with (from tensorflow.python.framework.ops) is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Colocations handled automatically by placer.\n"
     ]
    }
   ],
   "source": [
    "tf.reset_default_graph()\n",
    "test_input = tf.placeholder(tf.float32,[1, None, None,3], name='test')\n",
    "test_input_scale = test_input/127.5-1\n",
    "test_output = SRGAN_delta(test_input_scale,is_train=False,reuse=False)\n",
    "saver = tf.train.Saver()\n",
    "sess = tf.Session()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### 2. Load the checkpoint from the pretrained model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:From /home/malte/anaconda3/envs/kerasCPU/lib/python3.7/site-packages/tensorflow/python/training/saver.py:1266: checkpoint_exists (from tensorflow.python.training.checkpoint_management) is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Use standard file APIs to check for files with this prefix.\n",
      "INFO:tensorflow:Restoring parameters from checkpoint_delta/model-345002\n"
     ]
    }
   ],
   "source": [
    "saver.restore(sess, save_folder / 'checkpoint_delta/model-345002')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "outputs": [],
   "source": [
    "folder_smt = save_folder / \"imagenette2_smt\"\n",
    "folder_smt.mkdir(exist_ok=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Create folder for each class"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [],
   "source": [
    "for img_class in imgn_folders:\n",
    "    folder_smt_class = folder_smt / img_class\n",
    "    folder_smt_class.mkdir(exist_ok=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>class</th>\n",
       "      <th>path</th>\n",
       "      <th>scrambled_indices</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>n01440764</td>\n",
       "      <td>/home/malte/Dokumente/Masterarbeit/data/imagen...</td>\n",
       "      <td>{0: array([2, 3, 1, 0]), 1: array([0, 3, 1, 2]...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>n01440764</td>\n",
       "      <td>/home/malte/Dokumente/Masterarbeit/data/imagen...</td>\n",
       "      <td>{0: array([1, 3, 0, 2]), 1: array([1, 3, 0, 2]...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>n01440764</td>\n",
       "      <td>/home/malte/Dokumente/Masterarbeit/data/imagen...</td>\n",
       "      <td>{0: array([0, 1, 2, 3]), 1: array([1, 3, 0, 2]...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>n02102040</td>\n",
       "      <td>/home/malte/Dokumente/Masterarbeit/data/imagen...</td>\n",
       "      <td>{0: array([2, 0, 1, 3]), 1: array([0, 2, 3, 1]...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>n02102040</td>\n",
       "      <td>/home/malte/Dokumente/Masterarbeit/data/imagen...</td>\n",
       "      <td>{0: array([1, 3, 2, 0]), 1: array([2, 3, 0, 1]...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "       class                                               path  \\\n",
       "0  n01440764  /home/malte/Dokumente/Masterarbeit/data/imagen...   \n",
       "1  n01440764  /home/malte/Dokumente/Masterarbeit/data/imagen...   \n",
       "2  n01440764  /home/malte/Dokumente/Masterarbeit/data/imagen...   \n",
       "3  n02102040  /home/malte/Dokumente/Masterarbeit/data/imagen...   \n",
       "4  n02102040  /home/malte/Dokumente/Masterarbeit/data/imagen...   \n",
       "\n",
       "                                   scrambled_indices  \n",
       "0  {0: array([2, 3, 1, 0]), 1: array([0, 3, 1, 2]...  \n",
       "1  {0: array([1, 3, 0, 2]), 1: array([1, 3, 0, 2]...  \n",
       "2  {0: array([0, 1, 2, 3]), 1: array([1, 3, 0, 2]...  \n",
       "3  {0: array([2, 0, 1, 3]), 1: array([0, 2, 3, 1]...  \n",
       "4  {0: array([1, 3, 2, 0]), 1: array([2, 3, 0, 1]...  "
      ]
     },
     "execution_count": 46,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "grouped = pd.read_csv(folder_scr / \"scramble.csv\", index_col=0)\n",
    "grouped[:5]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {},
   "outputs": [],
   "source": [
    "grouped[\"path\"] = grouped[\"path\"].apply(lambda x: Path(x))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {},
   "outputs": [],
   "source": [
    "def smooth_image_array(img_arr):\n",
    "    im = np.float32(img_arr)\n",
    "    batch_images = [im]\n",
    "    test_output_eval = sess.run(test_output,feed_dict={test_input: batch_images}) \n",
    "    im_out = test_output_eval[0]\n",
    "    \n",
    "    im_out = (np.float64(im_out) + 1) / 2 * 255\n",
    "    im_out[im_out > 255] = 255\n",
    "    im_out[im_out < 0] = 0\n",
    "    \n",
    "    return im_out / 255"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[1, 2, 3, 4, 5]"
      ]
     },
     "execution_count": 49,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "smooth_op = list(range(1, 6))\n",
    "smooth_op"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {
    "collapsed": true,
    "jupyter": {
     "outputs_hidden": true
    }
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n",
      "Lossy conversion from float64 to uint8. Range [0, 1]. Convert image to uint8 prior to saving to suppress this warning.\n"
     ]
    }
   ],
   "source": [
    "for index, _class, path, indices in tqdm(grouped.itertuples()):\n",
    "    # load image    \n",
    "    im = imageio.imread(path)\n",
    "    # also save unmodified image\n",
    "    save_path = folder_smt / f\"{_class}/{path.stem}_original.jpg\"\n",
    "    imageio.imwrite(save_path, im)\n",
    "    \n",
    "    # repeat smoothening 'smooth_op' times\n",
    "    for op in smooth_op:\n",
    "        if op == 1:\n",
    "            smt_arr = im\n",
    "        else:\n",
    "            smt_arr = imageio.imread(save_path)\n",
    "        smt_arr = smooth_image_array(smt_arr)\n",
    "        save_path = folder_smt / f\"{_class}/{path.stem}_{op}.jpg\"\n",
    "        imageio.imwrite(save_path, smt_arr)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Also create difference dataset that consists of the parts that are discarded in the smoothening process."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 3.1 Create dataset that consists of the parts that are discarded in the smoothening process and their inverses"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [],
   "source": [
    "folder_smt_diff = save_folder / \"imagenette2_smtdiff\"\n",
    "folder_smt_diff.mkdir(exist_ok=True)\n",
    "\n",
    "folder_smt_diff_inv = save_folder / \"imagenette2_smtdiffinv\"\n",
    "folder_smt_diff_inv.mkdir(exist_ok=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Create folder for each class"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {},
   "outputs": [],
   "source": [
    "for img_class in imgn_folders:\n",
    "    folder_smt_diff_class = folder_smt_diff / img_class\n",
    "    folder_smt_diff_class.mkdir(exist_ok=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "metadata": {},
   "outputs": [],
   "source": [
    "for img_class in imgn_folders:\n",
    "    folder_smt_diff_inv_class = folder_smt_diff_inv / img_class\n",
    "    folder_smt_diff_inv_class.mkdir(exist_ok=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "metadata": {},
   "outputs": [],
   "source": [
    "smt_orig = sorted(folder_smt.rglob('*_original.jpg'))\n",
    "smt_one = sorted(folder_smt.rglob('*_1.jpg'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "metadata": {},
   "outputs": [],
   "source": [
    "for orig, one in tqdm(zip(smt_orig, smt_one):\n",
    "    orig_img = Image.open(orig)\n",
    "    one_img = Image.open(one)\n",
    "    \n",
    "    subfolder = orig.parent.stem\n",
    "    diff = ImageChops.difference(orig_img, one_img)\n",
    "    diff.save(folder_smt_diff / orig.parent.stem / f\"{orig.stem}_diff.jpg\")\n",
    "    invert_diff = ImageOps.invert(diff)\n",
    "    invert_diff.save(folder_smt_diff_inv / orig.parent.stem / f\"{orig.stem}_diff_inv.jpg\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Tranform images"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Up to now the images still have the full size and are not regularized with the ImageNet presets.\n",
    "\n",
    "Thus, the images are transformed."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 66,
   "metadata": {},
   "outputs": [],
   "source": [
    "def transform_img(img_folder):\n",
    "    for img_pth in img_folder.rglob('*.jpg'):\n",
    "        # load image\n",
    "        img = Image.open(img_pth)\n",
    "        # apply transformation\n",
    "        trf_img = transform(img)\n",
    "        trf_img.save(img_pth)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 67,
   "metadata": {},
   "outputs": [],
   "source": [
    "for img_folder in [folder_smt, folder_smt_diff, folder_smt_diff_inv]:\n",
    "    transform_img(img_folder)"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python [conda env:kerasCPU]",
   "language": "python",
   "name": "conda-env-kerasCPU-py"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.7"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}