run_model_cityscape.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 55,
   "id": "1d276f79",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\tomas\\AppData\\Local\\Temp\\ipykernel_22692\\1384538113.py:9: MatplotlibDeprecationWarning: The seaborn styles shipped by Matplotlib are deprecated since 3.6, as they no longer correspond to the styles shipped by seaborn. However, they will remain available as 'seaborn-v0_8-<style>'. Alternatively, directly use the seaborn API instead.\n",
      "  plt.style.use('seaborn-darkgrid')\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<style>.container { width:98% !important; }</style>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "import os\n",
    "import cv2\n",
    "import torch\n",
    "import numpy as np\n",
    "from PIL import Image\n",
    "from torchvision import transforms\n",
    "from sklearn.metrics import jaccard_score\n",
    "import matplotlib.pyplot as plt\n",
    "plt.style.use('seaborn-darkgrid')\n",
    "plt.rcParams.update({'font.size': 14})\n",
    "from IPython.display import display, HTML\n",
    "display(HTML(\"<style>.container { width:98% !important; }</style>\"))\n",
    "\n",
    "# Local imports\n",
    "from main_training import main\n",
    "from custom_model import initialize_model\n",
    "from dataloader import DataLoaderSegmentation\n",
    "from train import iou"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "e00017d2",
   "metadata": {},
   "outputs": [],
   "source": [
    "data_dir = r'..\\..\\data\\training_def'\n",
    "dest_dir = r'..\\..\\models\\deepLabV3_r3'\n",
    "city_scape = r'..\\..\\models\\deepLabV3_r2\\model_13_2_2_2_epoch_580.pth'\n",
    "num_classes = 7\n",
    "num_epochs = 25\n",
    "batch_size = 4"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "0584e3e0",
   "metadata": {},
   "source": [
    "### Training"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "849f75fa",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Initializing Datasets and Dataloaders...\n",
      "Initializing Model...\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\tomas\\miniconda3\\envs\\torch\\lib\\site-packages\\torchvision\\models\\_utils.py:208: UserWarning: The parameter 'pretrained' is deprecated since 0.13 and may be removed in the future, please use 'weights' instead.\n",
      "  warnings.warn(\n",
      "C:\\Users\\tomas\\miniconda3\\envs\\torch\\lib\\site-packages\\torchvision\\models\\_utils.py:223: UserWarning: Arguments other than a weight enum or `None` for 'weights' are deprecated since 0.13 and may be removed in the future. The current behavior is equivalent to passing `weights=DeepLabV3_ResNet101_Weights.COCO_WITH_VOC_LABELS_V1`. You can also use `weights=DeepLabV3_ResNet101_Weights.DEFAULT` to get the most up-to-date weights.\n",
      "  warnings.warn(msg)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Params to learn:\n",
      "\t module.classifier.0.convs.0.0.weight\n",
      "\t module.classifier.0.convs.0.1.weight\n",
      "\t module.classifier.0.convs.0.1.bias\n",
      "\t module.classifier.0.convs.1.0.weight\n",
      "\t module.classifier.0.convs.1.1.weight\n",
      "\t module.classifier.0.convs.1.1.bias\n",
      "\t module.classifier.0.convs.2.0.weight\n",
      "\t module.classifier.0.convs.2.1.weight\n",
      "\t module.classifier.0.convs.2.1.bias\n",
      "\t module.classifier.0.convs.3.0.weight\n",
      "\t module.classifier.0.convs.3.1.weight\n",
      "\t module.classifier.0.convs.3.1.bias\n",
      "\t module.classifier.0.convs.4.1.weight\n",
      "\t module.classifier.0.convs.4.2.weight\n",
      "\t module.classifier.0.convs.4.2.bias\n",
      "\t module.classifier.0.project.0.weight\n",
      "\t module.classifier.0.project.1.weight\n",
      "\t module.classifier.0.project.1.bias\n",
      "\t module.classifier.1.weight\n",
      "\t module.classifier.2.weight\n",
      "\t module.classifier.2.bias\n",
      "\t module.classifier.4.weight\n",
      "\t module.classifier.4.bias\n",
      "Train...\n",
      "Epoch 1/25\n",
      "----------\n",
      "train Loss: 1.3642 Acc: 0.1227\n",
      "val Loss: 1.6755 Acc: 1.0000\n",
      "\n",
      "Epoch 2/25\n",
      "----------\n",
      "train Loss: 0.4096 Acc: 0.9333\n",
      "val Loss: 0.9994 Acc: 1.0000\n",
      "\n",
      "Epoch 3/25\n",
      "----------\n",
      "train Loss: 0.0963 Acc: 1.0000\n",
      "val Loss: 0.5671 Acc: 1.0000\n",
      "\n",
      "Epoch 4/25\n",
      "----------\n",
      "train Loss: 0.0416 Acc: 1.0000\n",
      "val Loss: 0.3618 Acc: 1.0000\n",
      "\n",
      "Epoch 5/25\n",
      "----------\n",
      "train Loss: 0.0278 Acc: 1.0000\n",
      "val Loss: 0.2443 Acc: 1.0000\n",
      "\n",
      "Epoch 6/25\n",
      "----------\n",
      "train Loss: 0.0221 Acc: 1.0000\n",
      "val Loss: 0.1795 Acc: 1.0000\n",
      "\n",
      "Epoch 7/25\n",
      "----------\n",
      "train Loss: 0.0191 Acc: 1.0000\n",
      "val Loss: 0.1475 Acc: 1.0000\n",
      "\n",
      "Epoch 8/25\n",
      "----------\n",
      "train Loss: 0.0172 Acc: 1.0000\n",
      "val Loss: 0.1248 Acc: 1.0000\n",
      "\n",
      "Epoch 9/25\n",
      "----------\n",
      "train Loss: 0.0156 Acc: 1.0000\n",
      "val Loss: 0.1092 Acc: 1.0000\n",
      "\n",
      "Epoch 10/25\n",
      "----------\n",
      "train Loss: 0.0145 Acc: 1.0000\n",
      "val Loss: 0.0994 Acc: 1.0000\n",
      "\n",
      "Epoch 11/25\n",
      "----------\n",
      "train Loss: 0.0134 Acc: 1.0000\n",
      "val Loss: 0.0931 Acc: 1.0000\n",
      "\n",
      "Epoch 12/25\n",
      "----------\n",
      "train Loss: 0.0126 Acc: 1.0000\n",
      "val Loss: 0.0899 Acc: 1.0000\n",
      "\n",
      "Epoch 13/25\n",
      "----------\n",
      "train Loss: 0.0119 Acc: 1.0000\n",
      "val Loss: 0.0850 Acc: 1.0000\n",
      "\n",
      "Epoch 14/25\n",
      "----------\n",
      "train Loss: 0.0113 Acc: 1.0000\n",
      "val Loss: 0.0816 Acc: 1.0000\n",
      "\n",
      "Epoch 15/25\n",
      "----------\n",
      "train Loss: 0.0107 Acc: 1.0000\n",
      "val Loss: 0.0791 Acc: 1.0000\n",
      "\n",
      "Epoch 16/25\n",
      "----------\n",
      "train Loss: 0.0102 Acc: 1.0000\n",
      "val Loss: 0.0761 Acc: 1.0000\n",
      "\n",
      "Epoch 17/25\n",
      "----------\n",
      "train Loss: 0.0097 Acc: 1.0000\n",
      "val Loss: 0.0749 Acc: 1.0000\n",
      "\n",
      "Epoch 18/25\n",
      "----------\n",
      "train Loss: 0.0093 Acc: 1.0000\n",
      "val Loss: 0.0722 Acc: 1.0000\n",
      "\n",
      "Epoch 19/25\n",
      "----------\n",
      "train Loss: 0.0089 Acc: 1.0000\n",
      "val Loss: 0.0711 Acc: 1.0000\n",
      "\n",
      "Epoch 20/25\n",
      "----------\n",
      "train Loss: 0.0085 Acc: 1.0000\n",
      "val Loss: 0.0709 Acc: 1.0000\n",
      "\n",
      "Epoch 21/25\n",
      "----------\n",
      "train Loss: 0.0082 Acc: 1.0000\n",
      "val Loss: 0.0685 Acc: 1.0000\n",
      "\n",
      "Epoch 22/25\n",
      "----------\n",
      "train Loss: 0.0079 Acc: 1.0000\n",
      "val Loss: 0.0673 Acc: 1.0000\n",
      "\n",
      "Epoch 23/25\n",
      "----------\n",
      "train Loss: 0.0075 Acc: 1.0000\n",
      "val Loss: 0.0649 Acc: 1.0000\n",
      "\n",
      "Epoch 24/25\n",
      "----------\n",
      "train Loss: 0.0073 Acc: 1.0000\n",
      "val Loss: 0.0633 Acc: 1.0000\n",
      "\n",
      "Epoch 25/25\n",
      "----------\n",
      "train Loss: 0.0070 Acc: 1.0000\n",
      "Save current model : ..\\..\\models/deepLabV3_r2\\checkpoint_0025_DeepLabV3.pth\n",
      "val Loss: 0.0615 Acc: 1.0000\n",
      "Save current model : ..\\..\\models/deepLabV3_r2\\checkpoint_0025_DeepLabV3.pth\n",
      "\n",
      "Training complete in 11m 14s\n",
      "Best val Acc: 1.000000\n",
      "Save ...\n"
     ]
    }
   ],
   "source": [
    "state, model, hist = main(data_dir, dest_dir, num_classes, batch_size, num_epochs)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "de545018",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.legend.Legend at 0x277ff41f250>"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAA/wAAAKuCAYAAAAckBezAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjYuMiwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy8o6BhiAAAACXBIWXMAAA9hAAAPYQGoP6dpAACjS0lEQVR4nOzdd5hU5d3/8c+Ztr03mkuTYkQQxBZRiiCKoIAaowYLREVF84QkJsYkTyxJNMmj/iLRWIiIRsVOULFhixgQQYoiizTpyy7b6+zMnN8fuzOwwuIyO7Nnyvt1XVwuZ86c8x24H/J85r7v7zFM0zQFAAAAAABiis3qAgAAAAAAQOgR+AEAAAAAiEEEfgAAAAAAYhCBHwAAAACAGETgBwAAAAAgBhH4AQAAAACIQQR+AAAAAABiEIEfAAAAAIAYROAHAAAAACAGOawu4EjWrFmjH/7wh5o3b55OPfXUI547bdo0ffrpp0c8Z/78+YHrLFy4ULfeeuthz5syZYruueee4IoGAAAAACACRGzg37Ztm2666Sb5fL52nT9z5kxdfPHFhxzfvXu3HnjgAR1zzDE67rjjAseLiookSXfffbdcLler9xQWFnagcgAAAAAArBeRgf+dd97R7bffrsrKyna/54wzzjjkmNfr1RVXXKGEhATNmTNH6enpgdeKioqUk5OjSy65JCQ1AwAAAAAQSSJuD/91112nWbNmKS8vTxMnTuzQtebPn6/PP/9c119/vQYOHNjqtaKiIvXr169D1wcAAAAAIFJF3Az/li1bNHv2bF1zzTV65JFHgr5OWVmZHnroIfXs2VPXXnttq9f279+vkpISnXvuuZIkt9stSYcs7W9LSUl10HV1puzsFJWV1VpdBhAUxi+iGeMX0Yzxi2jG+EU0+67xm5eXdtTXjLjA/8Ybb7Q7eB/J448/rqqqKv3+978/5HobNmyQJO3Zs0dTp07Vhg0b5PP5NGjQIP3sZz/T6aef3uH7W80wJLvdJsOQTNPqaoCjw/hFNGP8IpoxfhHNGL+IZuEavxEX+EMR9uvq6vTCCy+oZ8+egVn8g/kb9q1cuVLTp0/XrFmztG3bNs2dO1czZszQnDlzNGbMmCPewzA6XGZY+euL9DqBw2H8IpoxfhHNGL+IZoxfRLNwjd+IC/yhsGjRIlVVVWn27Nmy2+2HvD548GDNnDlTU6dOVc+ePQPHx48fr4kTJ+qOO+7QqFGjZLMdvsVBdnaK7PaIa39wWDk5R7/sA4gUjF9EM8YvohnjF9GM8YtoFurxG5OB/+2335bT6dSECRMO+/rw4cM1fPjwQ453795d48aN08KFC7Vp0yb179//sO8vK6uN+G8ODaN5sOzfX82SJkQdxi+iGeMX0Yzxi2jG+EU0a8/4zc2NgT38HVVTU6Ply5drxIgRysjIOOr35+TkSJJqa4/c7CNa/hExzeipFfg2xi+iGeMX0Yzxi2jG+EU0C/X4jY516Udh9erVampq0plnntnmOTfeeKPGjRunhoaGQ17bvHmzJKmwsDBsNQIAAAAAEG4xF/i/+OILSdKgQYPaPCcvL0/bt2/XggULWh1fvny5PvroI40cOTIw0w8AAAAAQDSKyiX9O3bs0KpVq1RYWKihQ4e2em3r1q2Smvfjt+Xmm2/WRx99pHvvvVdFRUUaPHiwNm3apOeee075+fn63e9+F9b6AQAAACCUvF6PfD6f1WWgnWw2m+z28MfxqAz8K1as0G233aYpU6YcEvjLysokSenp6W2+Pzc3Vy+88IL+9re/6YMPPtDChQuVnZ2tKVOmaNasWSooKAhr/QAAAAAQCvX1taqtrZLH47a6FBwlh8OllJR0JSWlhO0ehmnS0uJolZRUW13CdzKM5i6OpaV0KUX0YfwimjF+Ec0Yv4hm8Th+6+trVVlZKpcrScnJqS2PJI/wx4lBkimv16u6uhq53fXKyMhVcnLKd47fvDy69AMAAABAXKitrZLLlaSsrDwZkf7ccLTidEoJCUkqLy9RbW2VkpPDM8sfc037AAAAACDWeb0eeTxuJSenEvajlGEYSk5OkcfjltfrCcs9CPwAAAAAEGX8Dfqal/EjWvkb94Wr4SKBHwAAAACiFrP70S28f38EfgAAAAAAYhCBHwAAAACAGETgBwAAAAAgBvFYPgAAAABAVJk79xE98cRj7Tr3mmuu1YwZ13fofqtWfaZbbpkZkmt1JgI/AAAAACCqjBw5Rj16HNPq2IMP3qeKigr99rd3tjret2+/Dt+vV6/e+u1v7wzJtToTgR8AAAAAEFWOPbafjj22dfh+7LGHJVVo/PgJIb9fdnZOWK4bbuzhBwAAAAAgBjHDDwAAAAAxxjRNNXh8VpfRpkSHTYYR3mfQ+/3hD7/XBx8s0Z13/kl//es9Ki8v16hRY/S7390lj8ej559/Vu+9946++Wabmprcys7O0amnnq7rrrtRWVnZkg6/h//iiyepT5+++uEPf6S5cx/Rxo0bZLfbNWzYybrhhpt1zDGFnfL5joTADwAAAAAxxDRN/fi5NVq7u8rqUto0pFu6HvvhkE4L/W63W//7v7fr0ksvV1pamgoKukqSfvvbX+njjz/UeedN1KRJk+V2u7Vs2SdatOhV7d27R/ff//cjXnfTpq91663/o/HjJ2j8+AnauLFICxe+pE2bNurZZ1+W3W7vjI/XJgI/AAAAAMSYzonR0cPr9Wry5KmtOux//fVG/ec/H+jiiy/V//zPLwLHL7nkh7r22iu1YsVyVVVVKj09o83r7ttXrDvu+JPOPntc4JjH06TXXluoVatW6OSTTwvHx2k3Aj8AAAAAxBDDMPTYD4ewpP9bzjhjZKvf9+vXX2+//aEMo3Vru/LyMqWmpkmS6urqjxj4ExISNGrUmFbHBg78nl57baH2798fosqDR+CPRaYp17Z3JPMEyehqdTUAAAAAOplhGEpyWrucPNLk5OQccszpdOndd9/SZ58t1+7du7R79y7t378/8GWEaR75S5OMjMxDlu27XC5Jks9n/RcuBP4YZK/cqvTXp0sFg6RL3rS6HAAAAACw3LeDeW1tjX7ykxtVVPSVBg8+UQMGHKfx4ydo4MDj9cILz+ittxZ/5zVttsh+8B2BPwaZzuTmH/Z9JXkbJVuCtQUBAAAAQIR54YXntGHDev3857dp8uSLWr0WCcvxQyGyv45AUHzJBfI5UyXTK3vlN1aXAwAAAAARp7KyUpJ07LH9Wh3/4ou1Wr16laTmZn/RjBn+WGQY8mb1lW3fGtnLN8mT1d/qigAAAAAgoowYcZZefPE53XHHbzVlysVKTU3Vhg3r9eabr8tut8vj8aimptrqMjuEGf4Y5c06VpJkL99scSUAAAAAEHlOOulk/f73f1RaWqqeeOJR/eMfc7Rhw3r9+Mc36O67/yxJWrbsE4ur7BjDNE3T6iKiTUlJ5H/Lk7zyb0pZ9mc1DLhY1WMfsLoc4KgYhpSbm6bS0mrxLxSiDeMX0Yzxi2gWb+O3qcmt/fv3KCenq5xOl9XlIEj+v8fc3K7q2jXniOM3Ly/tqK/PDH+M8mb2lSTZKzZZXAkAAAAAwAoE/hjVakl/PHzFCQAAAABohcAfo7yZvSTDJpu7Wra6fVaXAwAAAADoZAT+WGVPkDJ7Nv9YQeM+AAAAAIg3BP5Yltv8OD469QMAAABA/CHwx7LcfpIkezmN+wAAAAAg3hD4Y1nLDL+DTv0AAAAAEHcI/LEsMMO/xeJCAAAAAACdjcAfy1pm+G3VO6WmeouLAQAAAAB0JgJ/LEvOkS8hU4ZM2Su3Wl0NAAAAAKATEfhjmWHIm3WsJMlB4z4AAAAAiCsE/hjnzeorSbJX8Gg+AAAAAIgnDqsLQHgFAj8z/AAAAABixNy5j+iJJx5r17nXXHOtZsy4PqT337Fju445pjCk1wwHAn+M8y/pt5czww8AAAAgNowcOUY9ehzT6tiDD96niooK/fa3d7Y63rdvv5De+1e/mq3a2lo9+OAjIb1uOBD4Y5wns2UPf8VmyfRJBrs4AAAAAES3Y4/tp2OPbR3kH3vsYUkVGj9+Qljv/fHHH+nEE4eF9R6hQvqLcb70Y2TaHDI89bLV7LW6HAAAAABAJ2GGP9bZnfJm9JKjfJPsFZvkS+tmdUUAAAAA0GnWr/9C8+bN1bp1a9TY2KDu3XtowoQL9IMfXCa73R44b+PGDXr88Ue0ceMGVVZWKDc3X2eccaamT79W6ekZWrXqM91yy0xJ0urVqzRixHD9+tf/qwkTJln10b4TgT8OeDP7Ngf+8k1qOuYsq8sBAAAAEG6mKXnqra6ibY4kyTDCfpuPP/5Qv/nNL9WtW3ddfvmVSk5O0ooVy/X3vz+gdevW6A9/+LMMw9CuXTt1yy03KDc3Vz/4weVKS0vT+vVf6KWXFmj9+i/0yCNPqFev3vrtb+/UXXf9Tj179tKVV07XoEGDw/4ZOoLAHwe8WcdKW9+Sg8Z9AAAAQOwzTWW+PEXOvZ9ZXUmbmrqerIopL4c19Dc0NOiee+5Snz7H6h//+KdcLpck6aKLLtVjjz2sJ5+cq/fee1dnnz1OH374vmpqqnXffQ/qe98bJEmaNGmykpNT9PnnK1VaWqK8vHyNHz9Bd931O2VlZYe9V0AosIc/DngyWx7NV0HgBwAAAOJCJ8yeR7rPPluuiooKjR59turq6lRRURH4dfbZ4yRJH330niSpoKBAkvTwww/qs88+ldvtliTdfPNP9c9/Pq28vHxrPkQHMcMfB7xZLYG/fJPFlQAAAAAIO8Nonj2P8yX927d/I0l65JG/65FH/n7Yc/bs2SNJGjXqbJ1//gV6441F+vzzlUpISNDgwSfq9NNH6Nxzz1d6enpYaw0XAn8c8Ppn+Gv3ynDXyHSlWlwRAAAAgLAyDMmZbHUVlvJ6fZKkH/94po4//oTDnpOcnCJJstvtuu223+nqq3+spUs/0meffarVqz/XihXL9dRTT+gf//inunfv0Wm1hwqBPw6YiZnyJeXJVl8ie8UWefIju7EEAAAAAHRUt27NTyhzuRJ08smntnqtrq5Wy5f/Vzk5uZKkvXv3aOfOHRo+/BRdfPEPdfHFP5TH49Gzzz6lRx75u1555UXNmvU/nf0ROow9/HHCk9VHEsv6AQAAAMSHU045XcnJKXr++WdUWVnR6rUnn/ynfvvbX2nZsqWB3//P/9yoL7/8InCOw+EIrAw4+PF9NptNpmmG/wOEADP8ccKbeay0ezmN+wAAAADEhbS0NP30p7/Qn/50p6688oe64IIpys3N06pVK7RkyTs67rjjNWXKJZKkH/7wCr3//ju69daf6IILpqp79+7at2+fXn31JaWmpuqCC6YErpuVla1NmzbqlVde1JAhJ6pPn2Ot+ojficAfJ7xZzYPQwQw/AAAAgDhx3nkTVVDQRc88M18vvPCc3G63unTpoquumqHLLpumpKQkSVLPnr00Z85jevLJuXrrrTdUXl6m9PR0nXTSybrmmh+32r9/000/0cMPP6i//e3/NG3aNQR+WC/QqZ8ZfgAAAAAx6MUXFx32+LBhwzVs2PDvfP+xx/bTXXfd853nnXPOeTrnnPOOuj4rsIc/Tnj8nfortko+r8XVAAAAAADCjcAfJ3xpPWTaE2R4G2Wr3ml1OQAAAACAMCPwxwubXd7M3pLYxw8AAAAA8YDAH0c8mc3NJOwVWyyuBAAAAAAQbgT+OBJo3McMPwAAAADEPAJ/HPEGGvcR+AEAAAAg1hH444g3q3lJv6OcR/MBAAAAQKwj8McR/6P5bPWlMhoqrC0GAAAAQAiYVheADgnv3x+BP564UuRN6SJJslcwyw8AAABEK5utOcp5vV6LK0FHeL0eSQf+PkONwB9n/Mv67SzrBwAAAKKW3e6Qw+FSXV2NTJNZ/mhkmqbq6mrlcLhktzvCco/wXBURy5vVV9r5sRwVm9RodTEAAAAAgpaSkq7KylKVl5coOTmlJTQaVpeF72TK6/Worq5Wbne9MjJyw3YnAn+c8e/jZ4YfAAAAiG5JSSmSpNraKlVUlFpcDY6Ww+FSRkZu4O8xLPcI25URkQJL+tnDDwAAAES9pKQUJSWlyOv1yOfzWV0O2slms4VtGf/BCPxxxpvZEvgrt0neJsnutLYgAAAAAB1mtztkt1tdBSINTfvijC+1i0xHsgyfR/aq7VaXAwAAAAAIEwJ/vDFs8mT2kcSyfgAAAACIZQT+OOTN8jfu22RxJQAAAACAcCHwx6FA4z469QMAAABAzCLwxyF/4z5HBTP8AAAAABCrCPxxyHPwkn7TtLgaAAAAAEA4RHTgX7NmjY477jgtX768Xed/9tlnGjBgwGF/TZs2rdW5Xq9X8+bN03nnnafBgwdrzJgxuv/++9XQ0BCOjxJRvBm9ZcqQrbFSRkOZ1eUAAAAAAMLAYXUBbdm2bZtuuukm+Xy+dr+nqKhIkjR79mx16dKl1Wu5ubmtfn/HHXdowYIFGj9+vK688kqtX79ejzzyiL744gs9/vjjMgyj4x8iUjmT5EvrIXv1DjnKN6kpKcfqigAAAAAAIRaRgf+dd97R7bffrsrKyqN6X1FRkQzD0LRp05ScnNzmeWvXrtWCBQt06aWX6s477wwc79Gjh+677z4tXrxYEyZMCLr+aODN6it79Q7ZyzepqdupVpcDAAAAAAixiFvSf91112nWrFnKy8vTxIkTj+q9RUVF6tGjxxHDviS9/PLLkqTp06e3On7VVVcpISEh8Hos82S27OOv2GJxJQAAAACAcIi4wL9lyxbNnj1br7zyinr16tXu95mmqY0bN6pfv36Smvfo19fXH/bcNWvWKDMz85DrJyYmqn///lq7dm2w5UeNA4/mo1M/AAAAAMSiiFvS/8Ybb8jlch31+7755hvV1dWprq5OP/rRj7R69Wo1NTWpb9++mjVrVqsl+nv37lXXrl0Pe50uXbpo3bp1qq6uVlpaWpv3i/Qt/v762qrT19Kp31GxKeI/C+LPd41fIJIxfhHNGL+IZoxfRLNwjd+IC/zBhH3pQMO+tWvX6uqrr9b06dO1Z88ezZs3Tz/96U9VXl6uK664QpJUXV2t3r17H/Y6iYmJkqS6uro2A392dors9ohbHHFYOTltfGmRcKIkyV61Q7mZLsmR0HlFAe3U5vgFogDjF9GM8YtoxvhFNAv1+I24wB+snj17atasWRo5cqQGDx4cOH7hhRfq/PPP11//+ldNmjRJ6enpkpq3AByO/7jdbm/zXmVltRH/zaFhNA+W/furddiPaiYp25Umm7ta5ZvXyZszoNNrBNryneMXiGCMX0Qzxi+iGeMX0aw94zc39+i/DIiZwD9w4EANHDjwkOOpqamaOnWqHnroIa1cuVKjR49WSkqKGhoaDnsd//EjLeeXFDX/iJhmW7Ua8mb2lW3fatnKN8mTTeBH5Gl7/AKRj/GLaMb4RTRj/CKahXr8Rse69A7Kzc2VJNXW1kqSunfvrr179x723L179yorK0sJCbG/xN3fuM9RvtniSgAAAAAAoRYzgf+OO+7QmDFjtGvXrkNe27SpuRN9YWGhJGnIkCEqKyvTjh07Wp1XX1+vjRs3aujQoeEvOAJ4/J36Kwj8AAAAABBrYibwd+3aVbt27dKTTz7Z6vjmzZv18ssvq1+/fjrhhBMkSZMmTZIkPfbYY63OnT9/vtxut6ZOndo5RVvMm9lHEo/mAwAAAIBYFJV7+Hfs2KFVq1apsLAwMBs/bdo0vfHGG3ryySe1d+9enX766dq9e7eeeeYZORwO3XPPPTJaOu0NGzZMU6dO1YIFC1RZWakRI0Zo3bp1ev755zV69GiNHTvWyo/XafxL+u3lm5s3ikR6J0IAAAAAQLtFZeBfsWKFbrvtNk2ZMiUQ+JOSkvT000/r4Ycf1uLFi7VkyRKlp6dr5MiRuvnmmw95DN9dd92lwsJCvfTSS1qyZIm6dOmiG264Qddff33gi4FY583oKdOwy9ZUI1tdsXwpXawuCQAAAAAQIobZ1vPp0KaSkmqrS/hOhtH82IbS0iM/liTr6TPlqNyqigsXqKnHGZ1XIHAE7R2/QCRi/CKaMX4RzRi/iGbtGb95eUf/WL6Y2cOP4Hiz+kqicR8AAAAAxBoCf5zzZrYEfhr3AQAAAEBMIfDHOX/jPkc5M/wAAAAAEEsI/HHO4+/Uz5J+AAAAAIgpBP44F1jSX71Taqq3uBoAAAAAQKgQ+OOcmZQtX2KWJMlescXiagAAAAAAoULgx4F9/BU07gMAAACAWEHghzyBTv3s4wcAAACAWEHgx4F9/DTuAwAAAICYQeBHYEm/vZwl/QAAAAAQKwj8kDereYbfUbFZMn0WVwMAAAAACAUCP+RNL5Rpc8rwNMhWs8fqcgAAAAAAIUDgh2RzyJvRSxLL+gEAAAAgVhD4IemgZf0EfgAAAACICQR+SJK8mS2N++jUDwAAAAAxgcAPSZKnZYbfXk7gBwAAAIBYQOCHJMmb2RL4K1jSDwAAAACxgMAPSQf28Ntri2W4qy2uBgAAAADQUQR+SJLMhAx5k/MlSfaKLRZXAwAAAADoKAI/AryZfSTxaD4AAAAAiAUEfgR4s1o69dO4DwAAAACiHoEfAf7A76BxHwAAAABEPQI/AjyZPJoPAAAAAGIFgR8BgU79FVsln9fiagAAAAAAHUHgR4AvtbtMe4IMn1u26h1WlwMAAAAA6AACPw6w2QOd+h0s6wcAAACAqEbgRysef6f+CgI/AAAAAEQzAj9a8c/w28vp1A8AAAAA0YzAj1b8j+ajUz8AAAAARDcCP1rxB35HBTP8AAAAABDNCPxoxZPRvKTfVr9fRkO5xdUAAAAAAIJF4EdrrhR5U7tKkuwVWywuBgAAAAAQLAI/DuHN9O/jZ1k/AAAAAEQrAj8O4c3qK0lyEPgBAAAAIGoR+HEIj79TP0v6AQAAACBqEfhxCG9m8ww/S/oBAAAAIHoR+GNQrdujHzzxmf70xldBvd+/pN9e9Y3kbQplaQAAAACATkLgj0G7Khq0ZX+dXlq1M6j3+1K6ynQky/B5mkM/AAAAACDqEPhjUH5qgiSptMYtt8d39BcwjAP7+Ms3h7I0AAAAAEAnIfDHoIwkh1x2Q5JUUtMY1DW8mX0kSfYK9vEDAAAAQDQi8McgwzCUn9Y8y19cHWTgb5nhdzDDDwAAAABRicAfo/zL+vfVuIN6/4El/czwAwAAAEA0IvDHqIIOz/C3dOqv2CyZZsjqAgAAAAB0DgJ/jPIv6d8XbODP6CVThmyNlTLq94eyNAAAAABAJyDwxyj/DP++IJv2yZEkX/oxzT/SuA8AAAAAog6BP0blp7kkScXVwe3hlyRPZsuyfvbxAwAAAEDUIfDHqILUju3hlw506reXbwlJTQAAAACAzkPgj1H+PfxltW41eX1BXcPrn+FnST8AAAAARB0Cf4zKSnbKaTdkSiqtDW5Zv79Tv6N8cwgrAwAAAAB0BgJ/jLIZhgrSEyUF36nf07Kk31a9Q/I0hKw2AAAAAED4EfhjWNeM5sAf7D5+MylXvoQMGaZP9sptIawMAAAAABBuBP4Y1jUjSZK0rybITv2GIW9mH0l06gcAAACAaEPgj2H+Gf5gl/RLBzr1OyrYxw8AAAAA0YTAH8O6dHBJvyR5/J36adwHAAAAAFGFwB/DAjP8NR2f4bczww8AAAAAUYXAH8O6+Pfwd2RJf2CGf5NkmiGpCwAAAAAQfgT+GOaf4S+tdcvjCy6sezN6yjTssjXVyla7N5TlAQAAAADCiMAfw3JTE2S3GfKZ0v7aIDv1213yZvRs/rFiSwirAwAAAACEE4E/htlthvJSXJJCuKwfAAAAABAVCPwxLj8tQVJHG/cR+AEAAAAg2hD4Y1xBS+DvyKP5vJnNnfoddOoHAAAAgKhB4I9x+anNS/o7Evg9/kfzlRP4AQAAACBaEPhjXGBJf3WQTft00JL+ml1SU11I6gIAAAAAhBeBP8YVhGAPv5mYJV9itiTJQad+AAAAAIgKBP4YFwj8HVjSL0newLJ+GvcBAAAAQDQg8Mc4/5L+klq3vD4z6Ot4/Mv6adwHAAAAAFGBwB/jclJcshmS12eqvK4D+/gz/Y/mI/ADAAAAQDQg8Mc4h81QbkrHO/X7l/Q7WNIPAAAAAFGBwB8H/Mv6i2uCn+H3+Gf4K7dIpi8kdQEAAAAAwieiA/+aNWt03HHHafny5e06v76+Xvfff7/Gjx+vQYMG6eSTT9Z1112nNWvWHHLuwoULNWDAgMP++tWvfhXqj2Kp/NSON+7zpR8j0+aS4WmQrXp3qEoDAAAAAISJw+oC2rJt2zbddNNN8vnaN5tsmqZuuukmLV26VOeee66uuuoqlZWV6dlnn9UVV1yhxx57TKeffnrg/KKiIknS3XffLZfL1epahYWFofsgESA/FJ36bQ55M3rJUb5R9opN8qX3CFF1AAAAAIBwiMjA/8477+j2229XZWVlu9/z+uuva+nSpbr++us1e/bswPGLLrpIkyZN0t13363XX389cLyoqEg5OTm65JJLQlp7JAo8mq+mo4/m6ytH+UY5yjepqXBUCCoDAAAAAIRLxC3pv+666zRr1izl5eVp4sSJ7X7f0qVLJUmXXXZZq+Ndu3bVKaecok2bNqmsrCxwvKioSP369QtN0REuP7V5BUOHZvgleVoa99GpHwAAAAAiX8QF/i1btmj27Nl65ZVX1KtXr3a/79Zbb9VLL72kLl26HPLa/v37JUl2uz3w+5KSkkDgd7vdcruDb2gX6QpC0LRPap7hlyR7BYEfAAAAACJdxC3pf+ONNw7ZU98eWVlZysrKOuT4ypUrtXr1ag0cOFAZGRmSpA0bNkiS9uzZo6lTp2rDhg3y+XwaNGiQfvazn7Xa698WwzjqEjuVvz7DOGhJf3WjTJmyBVm8zx/4yzdH/OdHdDt4/ALRhvGLaMb4RTRj/CKahWv8RlzgDybst6W4uFi/+MUvJEk333xz4Li/Yd/KlSs1ffp0zZo1S9u2bdPcuXM1Y8YMzZkzR2PGjGnzutnZKbLbI25xxGHl5KQpLSNFhiF5fKaMxATltnwBcNRSh0iS7HXFyk01pcT0EFYKHConJ83qEoCgMX4RzRi/iGaMX0SzUI/fiAv8obJz505Nnz5du3bt0owZMzR27NjAa4MHD9bMmTM1depU9ezZM3B8/Pjxmjhxou644w6NGjVKNtvhQ31ZWW3Ef3NoGM2DZf/+apmmlJ3s0v5atzZ8s19Gl2AHkU1ZyQWy1xWrYvMaeQpODGXJQMC3xy8QTRi/iGaMX0Qzxi+iWXvGb27u0ee4mAz8a9eu1Y033qiSkhJNnz5dt956a6vXhw8fruHDhx/yvu7du2vcuHFauHChNm3apP79+7d5j2j5R8Q0m38VpCVof61bxdWNGlgQ/LdG3qy+stcVy1a+SWb+iaErFDgM//gFohHjF9GM8YtoxvhFNAv1+I2OdelH4d1339WVV16p0tJS3XrrrfrlL395VO/PycmRJNXW1oajPMv4O/UXV3ewcV/mgX38AAAAAIDIFVOB/6233tItt9wir9erBx54QDNmzDjseTfeeKPGjRunhoaGQ17bvLk5yBYWFoa11s4WaNxX07FH8/k79TvKN3W4JgAAAABA+MRM4N+wYYN+8YtfyOFw6PHHH9e5557b5rl5eXnavn27FixY0Or48uXL9dFHH2nkyJGBmf5YkZ96oFN/R3iyjpUk2Su2dLgmAAAAAED4ROUe/h07dmjVqlUqLCzU0KFDJUl/+ctf1NjYqFGjRmnv3r1auHDhIe8bN26ckpOTdfPNN+ujjz7Svffeq6KiIg0ePFibNm3Sc889p/z8fP3ud7/r7I8UdvktM/zFHQz83kx/4N8q+TySLSqHEAAAAADEvKhMaytWrNBtt92mKVOmaOjQofJ4PFq+fLkk6YMPPtAHH3xw2PctWbJEycnJys3N1QsvvKC//e1v+uCDD7Rw4UJlZ2drypQpmjVrlgoKCjrx03SO/LTmPfwdXdLvS+sm05Eow9MgW9UO+TJ7h6I8AAAAAECIGaZJD8ujVVJSbXUJ38kwmh/bUFra/FiHnRX1mjJ3hVx2Qx//ZISMDjxXMOu5c+TYv16V58+Tu9fY734DcJS+PX6BaML4RTRj/CKaMX4RzdozfvPyjv5pazGzhx9H5t/D7/aaqqz3dOhagX38dOoHAAAAgIhF4I8TLodN2clOSVJxiDr12yvo1A8AAAAAkYrAH0dC1anfm+l/NB8z/AAAAAAQqQj8cSRknfoDS/qZ4QcAAACASEXgjyP5qaHp1O/J7CNJsjWUyWgo73BdAAAAAIDQI/DHEf8Mf0eX9MuZLG9qd0k07gMAAACASEXgjyMF/iX9Ne4OX8vfuM/Bsn4AAAAAiEgE/jhSEKoZfkmeTDr1AwAAAEAkI/DHkYO79Jum2aFrHWjct6XDdQEAAAAAQo/AH0fyWpr2NXh8qm70dOhagcDPDD8AAAAARCQCfxxJdNqVkeiQFIJH87V06rdXfiN5O94TAAAAAAAQWgT+OHOgU3/HQrovpYt8zhQZprc59AMAAAAAIgqBP84c6NTfwcZ9hnHQsn4ezQcAAAAAkYbAH2dC2ak/sKyfR/MBAAAAQMQh8MeZgzv1d5R/ht/BDD8AAAAARBwCf5zJT2vu1L+vo0v6JXky+0pihh8AAAAAIhGBP84cmOHveGf9A3v4t0im2eHrAQAAAABCh8AfZ/xd+ourG2V2MKR7M3rJlCFbY6WM+tJQlAcAAAAACBECf5zxz/DXNXlV6/Z27GKORPnSC5t/ZFk/AAAAAEQUAn+cSXbZlZbgkNQ8y99RnqyWffw07gMAAACAiELgj0OBR/OFoHGfN7NlH385gR8AAAAAIgmBPw4FOvWH5NF8fSTRqR8AAAAAIg2BPw6Fo1O/gyX9AAAAABBRCPxxKNCpPwRL+j0tS/ptVTskT0OHrwcAAAAACA0CfxwqSD3waL6OMpNy5EvIkCFT9sqtHb4eAAAAACA0CPxxKJR7+GUY8ma2dOqncR8AAAAARAwCfxzKD2GXfumgffw07gMAAACAiEHgj0P+x/LVNHpV6/Z0+HqerJYZfhr3AQAAAEDEIPDHoRSXQykuuySpJBSd+lsa97GkHwAAAAAiB4E/ToWyU7+3ZYbfUb5JMs0OXw8AAAAA0HEE/jjl79QfisZ93vSeMm0OGZ462Wr3dPh6AAAAAICOI/DHKX+n/lA8mk92p7zpPZt/LN/S8esBAAAAADqMwB+n8lPD06nfXkGnfgAAAACIBAT+OBV4NF8ImvZJkjezjyQezQcAAAAAkYLAH6f8j+YL1Qy/J4tO/QAAAAAQSQj8cerADH+ol/QT+AEAAAAgEhD445S/S39lg0cNTd4OX8+/pN9es1ty13b4egAAAACAjiHwx6nUBLuSnM1//ftqOr6P30zMki8pR5LkqKRTPwAAAABYjcAfpwzDCHTqL65uCMk1PZn+ffw07gMAAAAAqxH441jIO/Vn9ZVE4z4AAAAAiAQE/jgW6k79NO4DAAAAgMhB4I9j/hn+4lB16s9snuF3sKQfAAAAACxH4I9jBakuSaF7NJ/Hv6S/Yotk+kJyTQAAAABAcAj8cSywhz8EXfolyZd2jEybS4a3UbbqXSG5JgAAAAAgOAT+OHagS39oZvhls8ub2VsSnfoBAAAAwGoE/jjmn+GvqG9Soyc0S/D9nfodNO4DAAAAAEsFHfiLior04osvBn7/9NNP6/TTT9eIESM0b968UNSGMMtIdCjB0TwESkLUqd+T2dKpnxl+AAAAALBUUIF/1apVuuiii/T4449Lkr766iv94Q9/kM/nU1JSku6991698cYbIS0UoWcYRuDRfCHr1B9o3McMPwAAAABYKajA/+ijjyozM1N//vOfJUn//ve/JUnz58/Xm2++qeHDh+tf//pX6KpE2OT7O/WHaIbfm+Wf4SfwAwAAAICVggr8n3/+uaZNm6bBgwdLkj7++GP17NlTAwYMkN1u17nnnqsNGzaEtFCER6BTf3VoOvV7M/tIkux1+2Q0VoXkmgAAAACAoxdU4G9oaFBubq4kqbS0VF9//bVOO+20wOt2u12maYamQoSVv1P/vhAt6TddafKmFEhiWT8AAAAAWCmowN+tWzdt3bpVkvT+++/LMAyNGDEi8Pqnn36qrl27hqZChFV+iPfwS5I3k2X9AAAAAGA1RzBvOuuss/T000+rrq5Ob7zxhtLT03XmmWdq3759evjhh7V48WLddNNNoa4VYRCY4Q/RHn6pZR//rqVylG9S6K4KAAAAADgaQQX+n/70p9q+fbueeeYZpaWl6Z577lFCQoJ27typZ599VmeeeaZmzJgR6loRBgVpzU37QjvD37KPv4JH8wEAAACAVYIK/ImJiXr44YdVXl6ulJQUuVzNoXHAgAF69tlnNXTo0JAWifDxP5avrK5JTV6fnPagdnm04gl06t/S4WsBAAAAAILToXSXlZUVCPuStGfPHmVmZna0JnSizCSnnHZDklRSE6JO/f7AX7lV8nlCck0AAAAAwNEJOvA/99xz+v3vfx/4/S233KJJkyZpwoQJuvHGG+V2hyY8IrwMwwh5p35faleZjkQZvibZq7aH5JoAAAAAgKMTVOB/5ZVX9Pvf/16rVq2SJH3wwQd6++23NWTIEF1wwQV6//339cQTT4S0UISPv1N/yBr3GTZ5MvtKolM/AAAAAFglqD38zz33nI4//nj961//kiS99dZbcjgceuihh5SdnS2bzabXXntN119/fUiLRXjkp4ahcV/WsXKWfil7xWZJ40J2XQAAAABA+wQ1w//1119r6tSpSkxMlCR98sknOv7445WdnS1JGjp0qLZvZyl3tPA37gttp37/DD+d+gEAAADACkEFfsMw5HQ6JUlbtmxRcXGxTjvttMDr9fX1SkpKCk2FCLuCwJL+0PVd8GY1B35HBUv6AQAAAMAKQQX+Xr166dNPP5UkLVq0SIZh6Mwzz5QkeTwevfHGG+rZs2foqkRYhbppnyR5Mv2P5mOGHwAAAACsEFTgnzp1ql577TVNnDhRjz76qHr37q3hw4fr66+/1kUXXaS1a9fqsssuC3WtCJOQN+2T5M3sI0myNZTLqC8L2XUBAAAAAO0TVOC/4oor9Ktf/UqGYej73/++/v73v0tqnt3ftWuXfvrTn2ry5MmhrBNh5A/8pTVueby+0FzUmSRvWg9JamncBwAAAADoTEF16Zekq6++WldffXWrY/3799cnn3wil8vV0brQibKTnbLbDHl9pkpr3eqSnhiS63oz+8pevVOO8k3ydD05JNcEAAAAALRP0IFfkmpqavTxxx9r586dcrlc6tatm0aMGBGq2tBJbIah/FSX9lQ1qri6MWSB35PVV64dH7KPHwAAAAAsEHTgf+utt/S73/1OVVVVMk1TUnP3/pSUFN15552aMGFCyIpE+OWnJmhPVWOIO/W3NO6r2BKyawIAAAAA2ieowL9mzRr97Gc/U1pamm655Rb169dPPp9PGzdu1NNPP61bb71VPXr00ODBg0NdL8Ik8Gi+EHbq92Y2P5qPGX4AAAAA6HxBBf6HH35YmZmZWrhwoXJycgLHzznnHF122WW68MIL9fjjj+tvf/tbyApFeIWlU39WS+Cv2i55GyV7QsiuDQAAAAA4sqC69H/++ee69NJLW4V9v5ycHF166aVauXJlh4tbs2aNjjvuOC1fvrzd73nllVc0efJknXjiiRoxYoTuuOMOVVZWHnKe1+vVvHnzdN5552nw4MEaM2aM7r//fjU0NHS47miUH4YZfl9ygXzOVBmmV/bKb0J2XQAAAADAdwsq8NfV1R027PtlZ2eruro66KIkadu2bbrpppvk87X/MXGPPPKIfvWrXykzM1O/+MUvNGnSJL3wwgu68sorDwnyd9xxh/70pz+pX79+uu2223TGGWfokUce0U033RToSRBPClKbn6xQXB26PfwyjAOz/DyaDwAAAAA6VVBL+nv06KFly5bp8ssvP+zry5YtU7du3YIu6p133tHtt99+2Jn5tuzdu1cPPvigzjrrLD3yyCOy2Zq/yzj++OP1s5/9TE899ZSuvfZaSdLatWu1YMECXXrppbrzzjtbfa777rtPixcvjrumg+FY0i81N+5z7lsjezmBHwAAAAA6U1Az/BMnTtQ777yjOXPmyO0+MCPsdrs1Z84cvfvuu0EH5uuuu06zZs1SXl6eJk6c2O73LVq0SE1NTbr66qsDYd9fa/fu3fXyyy8Hjvl/nj59eqtrXHXVVUpISGh1brzIT20O/KU1jfL6QrfCwd+4z0HjPgAAAADoVEHN8F977bX68MMPNWfOHP3zn/9UYWGhDMPQN998o7q6On3ve9/TddddF1RBW7Zs0ezZs3XNNdfokUceaff71qxZI0kaMmTIIa+dcMIJevPNN1VdXa20tDStWbNGmZmZ6tWrV6vzEhMT1b9/f61duzao2qNZTopLdkPymtL+Wndgxr+jPFl06gcAAAAAKwQV+F0ul+bPn6+5c+fqzTff1LZt22SapgoLCzV+/HjNmDFDiYmJQRX0xhtvyOVyHfX79u7dq/T0dKWmph7yWpcuXSRJu3bt0sCBA7V371517dr1sNfp0qWL1q1bF/hyoC2GcdQldip/fe2t02E3lJuaoOLqRpXUNqogPTSB35d1rCTJXrFFhszI/4NDRDja8QtEEsYvohnjF9GM8YtoFq7xG1Tgl5pnw2+66SbddNNNoawnqLAvSdXV1UpOTj7sa/4vH+rq6gLn9u7d+zvPbSvwZ2enyG4PajdEp8vJaftLi2/rnpWk4upG1cum3Nz2v++IMk+QDLts7irlOiulzGNCc13EhaMZv0CkYfwimjF+Ec0Yv4hmoR6/QQf+I3nuuef08ccfa86cOeG4fJu+q7u+3W7/znP9xw8+99vKymoj/ptDw2geLPv3V6u9Dx3ITmweDpt2V6q066ErJYKVkTdIzn1rVP3le2ocMDVk10XsCmb8ApGC8YtoxvhFNGP8Ipq1Z/wGMykblsD/1VdfacmSJeG4dJtSUlJUXl5+2Nfq6+slKbDcPyUl5ZDH9Pn5jx9pOb+kqPlHxDTbX6t/335xdWNIP19Tt9Pk3LdGjl3L1dCfwI/2O5rxC0Qaxi+iGeMX0Yzxi2gW6vEbHevS26FHjx6qrKwMLNs/2N69e2Wz2VRQUCBJ6t69u/bu3XvY6+zdu1dZWVlKSAjNHvZo4u/Uv686tI/ma+p2miTJuWd5SK8LAAAAAGhbzAT+wYMHS9JhO+yvW7dO/fr1C8zwDxkyRGVlZdqxY0er8+rr67Vx40YNHTo0/AVHoINn+EOpqevJMmXIUb5JRl1pSK8NAAAAADi8mAn85513npxOpx5//PFW+/Nfe+017d69W1OnHlhKPmnSJEnSY4891uoa8+fPl9vtbnVuPCloCfz7akIb+M3ETHlzBkpilh8AAAAAOktY9vCH244dO7Rq1SoVFhYGZuO7d++umTNn6sEHH9T06dN13nnnaevWrXrqqad0wgkn6Ic//GHg/cOGDdPUqVO1YMECVVZWasSIEVq3bp2ef/55jR49WmPHjrXqo1kqP7X5CQn7atzymaZsIexM2NTtVDn2fyXnrmVy9z0/ZNcFAAAAABxeVAb+FStW6LbbbtOUKVNaLb+fNWuWcnJy9PTTT+vOO+9Ubm6uLr30Ut1yyy2Bx+353XXXXSosLNRLL72kJUuWqEuXLrrhhht0/fXXy4j0Fvxhkpviks2QvD5TZXVNyk0J7hGJh+PudpqS1s2Ta/dy1YbsqgAAAACAthjmdz3LTtLZZ599VBetqKhQXV2dvvrqq6ALi2QlJdVWl/CdDKP5sQ2lpUf3WJIJjyxTSY1bT14xVN/rErpnQBq1+5Q7b5hMGdo/Y53MxMyQXRuxJ9jxC0QCxi+iGeMX0Yzxi2jWnvGblxemx/Lt2rXrqC8cr7Pk0S4/NUElNW7tq24MaeA3U/LlyewrR8VmOfd+Jnev+Nw2AQAAAACdpV2Bf8OGDeGuAxEiPy1BX+6tDnmnfqllH3/FZjl3LyPwAwAAAECYxUyXfoTGgcZ94Qn8kuTcTad+AAAAAAg3Aj9a8T+aLzwz/KdJkhwl6yQ3rfsAAAAAIJwI/GjFH/j31bhDfm1fWnd503rI8HnkLF4Z8usDAAAAAA4g8KOV/NSWwB+GGX6JZf0AAAAA0FkI/GglPzDD36h2PLHxqB0I/MtCfm0AAAAAwAEEfrSS19K0r8lrqqK+KeTX9+/jdxavljwNIb8+AAAAAKBZUIH/3HPP1aOPPqri4uJQ1wOLOe02ZSc7JYWncZ83o7e8yfkyvI1y7lsT8usDAAAAAJoFFfgNw9B9992nMWPG6Nprr9XixYvldoe+yRuscaBTfxj+Tg2DffwAAAAA0Akcwbxp8eLFWrt2rV555RW98cYb+s9//qOMjAydf/75mjJlik444YRQ14lOVJCWoK+Ka7SvJnyN+xI3LSLwAwAAAEAYBRX4JWnw4MEaPHiwfv3rX+u9997TK6+8ohdeeEHPPvus+vbtq4suukgXXHCBcnJyQlkvOkGnderfs0LyNkl2Z1juAwAAAADxrMNN+5xOp8aPH69//OMf+uijj3ThhRdq06ZN+vOf/6xRo0Zp9uzZ2rBhQyhqRSc5uFN/OHizB8iXkCHDUydH6RdhuQcAAAAAxLugZ/j9TNPU0qVLtXDhQr3//vuqqalRdna2LrzwQknSwoUL9fbbb+v//u//NH78+A4XjPDLT2vu1B+uGX4ZNjV1PVUJ296Wc/dyeQqGhuc+AAAAABDHgg78X331lRYuXKjXXntN+/fvl81m05lnnqmLLrpIo0ePlsPRfOmZM2dqypQp+vOf/0zgjxKBJf014WvE2NT9tEDgrx86M2z3AQAAAIB4FVTgnzRpkjZt2iTTNNWrVy9deeWVmjJlivLy8g45NyMjQyeeeKI+/vjjDheLznGgS3+jTNOUYRghv8eBffyfSqZPMjq8uwQAAAAAcJCgAv/OnTs1ZcoUXXTRRTrppJO+8/wJEyZoypQpwdwKFshrmeFv9PhU2eBRZlLom+p5co+Xz5kiW2Ol7Ps3yJv7vZDfAwAAAADiWVCBf+nSpUpOTm73+WPHjg3mNrBIgsOmrCSnyuubtK+6MSyBXzaHPF2Hy7X9Qzl3LyPwAwAAAECIBRX4/WH/s88+01tvvaUdO3bIbrerV69emjBhgo4//viQFonOl5+W0Bz4axrVPz81LPdo6nqaXNs/lGv3cjUMnh6WewAAAABAvAoq8Pt8Pv3yl7/Ua6+9JtM0W732xBNP6KqrrtIvf/nLkBQIa+SnulS0L4yd+iW5u52qFEnO3csl05TC0CsAAAAAAOJVUIF/7ty5WrRokcaNG6drr71Wffr0kdfr1ebNm/XII49o3rx56t+/P/v2o1i+v3FfGDv1ewqGyLQnyFZfKnvFFnmz+obtXgAAAAAQb4Jqjf7iiy/qjDPO0IMPPqjBgwcrNTVVGRkZGjZsmP7xj39o+PDhmj9/fqhrRSfyd+oP5wy/7Alq6jJMkuTcvSx89wEAAACAOBRU4N+zZ4/GjBlz2NcMw9C5556rrVu3dqgwWCs/9cCj+cKpqWvL4/l2Lw/rfQAAAAAg3gQV+Hv06HHEQL93717l5+cHXRSs1ykz/JKaup0micAPAAAAAKEWVOC/4YYb9Nxzz+n1118/5LUVK1bo6aef1qxZszpcHKzj38O/r6bxkMaModTUZZhMm0P2ml2yVe0I230AAAAAIN4E1bTv008/VX5+vn7+85/roYceUr9+/eRyufTNN99o7dq1SkhI0IsvvqgXX3wx8B7DMPTkk0+GrHCEV36qS5JU3+RTTaNXaYlBDZXv5kyWJ2+wnMWr5Ny9XI3px4TnPgAAAAAQZ4JKcS+88ELg582bN2vz5s2tXm9oaNCnn37a6pjBI9eiSqLTroxEhyobPCquaQxf4JfU1O3UlsC/TI0DLw7bfQAAAAAgngSV4jZs2BDqOhCB8tMSVNng0b7qRh2bmxK2+zR1O036/GH28QMAAABACAW1hx/xwd+pP+yN+7qeLFOGHJVbZastDuu9AAAAACBeBL1O2+1267HHHtPixYu1c+dOuVwude3aVeeee65mzJghl8sVyjphgfy05r/DcD+az0xIlyf3eDlLv5Bz96dq7DcprPcDAAAAgHgQ1Ay/2+3WlVdeqQcffFC7du1Sr169VFBQoO3bt+v//b//pyuuuEJutzvUtaKTFRzUqT/cmrqdKkly7lkW9nsBAAAAQDwIKvA/+uijWr16tWbOnKnly5fr1Vdf1aJFi7Rs2TLdcMMNWrdunebNmxfiUtHZDizpD/+XN4HAv4vADwAAAAChEFTgf/311zV27Fj9z//8T6ul+wkJCfrJT36isWPHatGiRSErEtbIb5nhL+7EGX5HWZGMhvKw3w8AAAAAYl1QgX/nzp36/ve/3+brp59+unbs2BF0UYgMBZ3UtE+SzKQcebL6SZKcuz/9jrMBAAAAAN8lqMCfnJyssrKyNl8vKyujaV8MyGtp2lfr9qqm0RP2+wWW9fN4PgAAAADosKAC/9ChQ/Xcc8+pvPzQpddlZWVasGCBhg4d2uHiYK0Ul0OpCXZJUklNZ+zjP02S5NxD4AcAAACAjgrqsXwzZ87U5ZdfrkmTJmnatGnq27evDMPQ119/raeeekoVFRW69tprQ10rLJCfmqCaxjoVVzeod05yWO/V1O0USZKjZJ0Md41MV2pY7wcAAAAAsSyowH/iiSfqr3/9q/73f/9X999/vwzDkCSZpqnU1FTdc889Gj58eEgLhTUK0hK0ZX9dp3Tq96V2kze9p+xV38ix9zM1FY4K+z0BAAAAIFYFFfjdbrcmTJigs846S5988om2b98u0zRVWFioM844Q6mpzMzGis7s1C817+O3V30j165lBH4AAAAA6ICgAv/kyZP1gx/8QFdffbXOOeecUNeECNKZnfolyd3tVCVueJ59/AAAAADQQUE17duxY4eSk8O7nxuRIb+lU/++TpzhlyRH8WrJU98p9wQAAACAWBRU4B84cKBWrlwZ6loQgfxL+jtjD78k+dJ7yptSIMPXJGfx551yTwAAAACIRUEt6b/mmmv0m9/8Rt98841GjRql3NxcORyHXmry5MkdrQ8Wy29Z0l/cSUv6ZRhq6naa7F8vlHP3cjV1/37n3BcAAAAAYkxQgX/27NmSpNWrV2v16tWSFOjULzV36zcMg8AfAwpaZvirGz2qc3uV7LKH/Z5N3U5TYkvgBwAAAAAEJ6jA/8c//rFVwEfsSk1wKMVlV63bq301jeqVHf7eDf59/M69n0let2R3hf2eAAAAABBrggr8U6dOPeLrHo9Hu3btCqogRJ781ARtLavTvurOCfzerH7yJWbL1lAmR8k6ebqcFPZ7AgAAAECsCapp33HHHafXXnutzddfffXV7/xSANGjszv1N+/jP0WS5Ny9rHPuCQAAAAAxpl0z/MXFxfrvf/8b+L1pmlqxYoU8Hs8h5/p8Pi1atEg+ny90VcJS/sZ9ndWpX2rex5+w5U05dy9X/bCbOu2+AAAAABAr2hX4s7Ky9OCDD2r37t2Smhv0Pf/883r++ecPOdc0TUnSBRdcEMIyYaXAo/k6a4ZfB+3j37NC8nklW/ibBQIAAABALGlX4He5XJozZ442bNgg0zT161//Wj/4wQ80dOjQQ8612WzKzc3VaaedFvJiYQ1/4O+0R/NJ8uR8Tz5Xmmzuajn2fyVP3qBOuzcAAAAAxIJ2N+077rjjdNxxx0mSVqxYoYsuukhDhgwJW2GIHAUWBH7Z7GrqerISvnlPzt3LCPwAAAAAcJSCatr3pz/9ibAfRwoCe/g7MfDroGX9NO4DAAAAgKMW1GP5JGn79u16+eWXVVpaKq/Xe8jrhmHoj3/8Y4eKQ2Twd+mvbPCoocmrRGfn7Kdv6ta8LcS5e7lkmpJhdMp9AQAAACAWBBX4ly5dquuvv/6wXfr9CPyxIy3BoUSHTQ0en0pq3DomK6lT7uvJO0GmI1G2hnLZy7+WN7t/p9wXAAAAAGJBUIH/wQcfVFJSkn7729/qhBNOkMvlCnVdiCCGYSg/LUHby+u1r6ax0wK/7C41FZwk166lcu5eTuAHAAAAgKMQVOD/6quvNHPmTB69F0f8gb9TG/epeR9/c+BfpoZB0zr13gAAAAAQzYJq2peYmKjMzMwQl4JIVpDavIqj0wN/92/t4wcAAAAAtEtQgf/000/Xhx9+GOpaEMH8j+br9E79BUNl2pyy1+6VrWp7p94bAAAAAKJZUIH/F7/4hb766ivdddddWr16tXbu3Kndu3cf8guxI98f+GvcnXtjR5I8BSdK4vF8AAAAAHA0gtrDP3HiRHk8Hj3zzDN65plnDnuOYRhav359h4pD5MhPtWaGX5Kaup4q554Vcu1ersbjLu30+wMAAABANAoq8I8fP14Gz0SPKwdm+Ds/8Lu7narkVXOa9/EDAAAAANolqMB/zz33hLoORLiClhn+sromuT0+uRxB7QYJiqfrcJmGTfaqb2Sr2S1fardOuzcAAAAARKuwpLaGhgb28MeYjCSHElpCfklt587ym640eXIHSZKcuz/t1HsDAAAAQLRqV+CfMmWKPvroo1bHmpqatGTJEpWXlx9y/ttvv62zzz47NBUiIhiGoXyLHs0nSU3dDno8HwAAAADgO7Ur8H/11VeqrKxsdaympkazZs3Shg0bwlIYIk9gH391J3fql9TU7VRJBH4AAAAAaK8OLek3TTNUdSAKWNqpv9spkiRH+UYZ9fs7/f4AAAAAEG06r/Maop6VnfrNxCx5sgdIYpYfAAAAANqDwI9288/wW7GHX2IfPwAAAAAcDQI/2q0grblp376azt/DL7GPHwAAAACOhsPqAr6tvLxcc+bM0Xvvvaf9+/erV69euvLKK3XxxRcf8X3Tpk3Tp58e+ZFt8+fP16mnNofGhQsX6tZbbz3seVOmTNE999wT3AeIYQea9lk1w9/8d+co/VJGY5XMhHRL6gAAAACAaBBRgb+urk4zZszQxo0bdfnll6tPnz568803dfvtt6u0tFQzZ85s870zZ8487JcCu3fv1gMPPKBjjjlGxx13XOB4UVGRJOnuu++Wy+Vq9Z7CwsIQfaLYUtAS+PfXutXk9clp79wFIr6UAnkyestRuVXOPSvk7sWjHwEAAACgLe0O/M8//7w++eSTwO/dbrcMw9DcuXP173//u9W5O3fuDKqYp59+Wl9++aXuu+8+nX/++ZKkSy+9VNdee63mzJmjCy+8UF27dj3se88444xDjnm9Xl1xxRVKSEjQnDlzlJ5+YEa4qKhIOTk5uuSSS4KqNR5lJjnltBtq8poqrXWra3pip9fQ1O3U5sC/exmBHwAAAACOoN2Bf8WKFVqxYsUhxz/++OPDnm8YxlEX8+qrr6qgoCAQ9v3X+fGPf6z//Oc/WrRoka677rp2X2/+/Pn6/PPPdcstt2jgwIGtXisqKlK/fv2OusZ4ZjMM5aUmaHdlg/ZVN1oU+E9T0lfPsY8fAAAAAL5DuwL//Pnzw12HqqurtWXLFo0bN+6Q14YMGSJJWrt2bbuvV1ZWpoceekg9e/bUtdde2+q1/fv3q6SkROeee66k5tUKkg5Z2o9DFaS6tLuywcJO/S37+EvWSk11kjPZkjoAAAAAINK1K/Cfcsop4a5DxcXFMk3zsEv2k5KSlJGRcVRbBR5//HFVVVXp97///SFBfsOGDZKkPXv2aOrUqdqwYYN8Pp8GDRqkn/3sZzr99NO/8/pBLGDoVP76Ql2nv3FfSY3bkj8DM72HvKndZK/ZLVfxKjUdM6Lzi0DYhWv8Ap2B8YtoxvhFNGP8IpqFa/xGTNO+6upqSVJy8uFnbBMTE1VfX9+ua9XV1emFF15Qz549A7P4B/M37Fu5cqWmT5+uWbNmadu2bZo7d65mzJihOXPmaMyYMW1ePzs7RfZOblgXrJyctJBer1d+mrShRFUeU7m5ob12u/UeIa17XhkVn0tDz7OmBnSKUI9foDMxfhHNGL+IZoxfRLNQj9+ICfymabb67+Fet9naF7IXLVqkqqoqzZ49W3a7/ZDXBw8erJkzZ2rq1Knq2bNn4Pj48eM1ceJE3XHHHRo1alSb9ysrq434bw4No3mw7N9frTb+SIOS5mj+4Nv2Vau0tDp0Fz4KCTnDlKbn5d70kapOuNmSGhBe4Rq/QGdg/CKaMX4RzRi/iGbtGb/BTLhGTOBPSUmRJDU0NBz29YaGhjY79H/b22+/LafTqQkTJhz29eHDh2v48OGHHO/evbvGjRunhQsXatOmTerfv3+b94iWf0RMM7S15qc2L+nfV9No2Z9BU7fmLRfOvatkehole4I1hSDsQj1+gc7E+EU0Y/wimjF+Ec1CPX4jZl16jx49ZBiG9u7de8hrdXV1qqqqUpcuXb7zOjU1NVq+fLlGjBihjIyMo64jJydHklRbW3vU740H/j38+yxq2idJ3sw+8iXlyvA2ylG8xrI6AAAAACCSRUzgT0lJUd++fbVu3bpDXluzpjnUDRs27Duvs3r1ajU1NenMM89s85wbb7xR48aNO+xqgs2bN0uSCgsL21t6XClIbW6AWFrrlsdn0VenhhHo1u/i8XwAAAAAcFghDfy1tbVtLslvjwsuuEC7du3S66+/Hjhmmqbmzp0rl8vV5hL9g33xxReSpEGDBrV5Tl5enrZv364FCxa0Or58+XJ99NFHGjlyZGCmH61lJbtktxnymdL+WrdldbhbAr9zzzLLagAAAACASBb0Hv6PPvpIX3zxhW688UZJ0r333qv58+fLMAxddtlluv3224/6mldddZX+/e9/65e//KW++OIL9e7dW4sXL9Ynn3yiW2+9Vfn5+ZKkHTt2aNWqVSosLNTQoUNbXWPr1q2Smvfjt+Xmm2/WRx99pHvvvVdFRUUaPHiwNm3apOeee075+fn63e9+d9S1xwu7zVBeikt7qxu1r7pRBWnW7J9v6naaJMmx5zPJ55FsEdOOAgAAAAAiQlAz/O+//76uv/56LVy4UJL02Wef6YknnlD37t01dOhQPf3003rmmWeO+rqJiYl66qmnNHnyZC1cuFB/+MMfVF5ernvvvVczZswInLdixQrdeuuth8zQS1JZWZkkKT09vc375Obm6oUXXtDFF1+sjz/+WHfddZfeeustTZkyRS+88IJ69Ohx1LXHk8A+/hoL9/FnD5AvIUO2plo5Sr+0rA4AAAAAiFRBTYs+8cQTKiws1Pz58yVJr7/+umw2WyD033jjjXrppZd0+eWXH/W1s7Ozdffddx/xnKlTp2rq1KmHfe2xxx5r131yc3N15513HnV9UGBWv9jCxn2y2dXU9WQlbHtXzt3L5ckfYl0tAAAAABCBgprh/+qrr3TZZZepoKBAkrR06VL1798/sIz+zDPP1JYtW0JXJSKK/9F8lgZ+HVjW76RxHwAAAAAcIqjA7/F4lJqaKknatWuXtm/frlNPPTXwus/nk8PBnupYlZ/W3Kl/X7V1TfskBTr1O3cvl0yfpbUAAAAAQKQJKvD36NFD69evlyS99dZbMgyj1WPw3n//ffbBx7CCCNjDL0me3EEyHcmyNVbIXlZkaS0AAAAAEGmCCvzjx4/XggULdP311+vBBx9UQUGBTjvtNG3fvl3XXXedli5dqsmTJ4e4VEQK/5L+fRYv6ZfdqaauwyWxrB8AAAAAvi2owH/jjTfqsssu06effqquXbvq/vvvl8PhUEVFhT7++GP94Ac/0LRp00JdKyKEv0t/Sa1bXp9paS2tlvUDAAAAAAKC2mhvs9n0m9/8Rr/5zW9aHR84cKDef//9QDM/xKacFJdshuT1mSqvcyu3ZcbfCq338ZuSYVhWCwAAAABEkqBm+A/H6/Xqv//9r77++mv5fDRQi2UOm6HclObGfZZ36s8/UaY9Qfa6fbJXbrW0FgAAAACIJEEFftM09X//93+64YYbJDWH/SuuuEIzZ87Utddeq4svvlg1NTUhLRSRxd+4r7jG2k79ciSqqeBESSzrBwAAAICDBRX4582bp8cee0y1tbWSmjv1r169WmeffbZuuOEGff311/rHP/4R0kIRWfz7+C1v3CepqdtpkiTn7mUWVwIAAAAAkSOoPfwLFy7U6aefrn/+85+SpHfffVcul0v33nuvUlJSVFlZqXfeeUc///nPQ1osIkfEdOoXjfsAAAAA4HCCmuH/5ptvdO6558poaZC2bNkyDRkyRCkpKZKam/ft2bMndFUi4gRm+GsiIPAXnCTTsMtevVO2qp1WlwMAAAAAESGowO9yuQKN+b766iuVlZXptNNOC7xeVVWltLS00FSIiJSf2ty0LxJm+OVKkSfvBEmScw+z/AAAAAAgBRn4+/Tpo/fff1+StGDBAhmGoVGjRkmSampq9PLLL6tv374hKxKRJ2Ka9rVgWT8AAAAAtBZU4J82bZr+85//6KSTTtJzzz2nIUOG6Pjjj9e6det07rnnasuWLbrmmmtCXSsiSMFBTft8pmlxNQc37iPwAwAAAIAUZNO+CRMmyOFw6OWXX1aXLl108803S5ISExOVmpqq2267TaNHjw5poYgsuSkuGZI8PlPldU3KSXFZWk9T15NlypCjYrOMuhKZyXmW1gMAAAAAVgsq8EvSOeeco3POOafVsX79+unNN9/scFGIfA67TTkpLpXWurWvptHywG8mZsqbc5wc+9fLuXu53MdOtLQeAAAAALBa0IFfknbs2KG3335bO3fulMvlUteuXXXOOeeoW7duoaoPESw/LaE58Fc36rgC65s0urudKsf+9XLtXkbgBwAAABD3gg788+bN01//+ld5PJ5Wx//6179q9uzZmj59eoeLQ2TLT3VpvaTi6ghq3LfuCfbxAwAAAICCDPwfffSR7rnnHvXp00czZ85U//795fV69fXXX+vRRx/VX/7yFw0YMEBnnHFGqOtFBAk07quJgEfz6UCnfvv+DTIaymUmZllcEQAAAABYJ6jAP3fuXPXs2VMvvviikpOTA8ePP/54nXPOOZoyZYqeeOIJAn+My0890Kk/EpjJefJk9pWjYrOcez6Tu/c4q0sCAAAAAMsE9Vi+L774QlOmTGkV9v2Sk5M1ZcoUrVu3rsPFIbL5Z/iLIyTwSwc/nm+ZxZUAAAAAgLWCCvwej0cpKSltvp6cnKyGhoagi0J0yI+wJf3SgWX9BH4AAAAA8S6owN+7d2+99957bb6+ZMkS9ezZM+iiEB3y05ofxbevulGmaVpcTTP/DL+j5AsZ7hqLqwEAAAAA6wQV+C+66CL997//1a9//WsVFxcHjhcXF+u2227Tp59+qilTpoSsSESmvJTmGX6311Rlvec7zu4cvrRu8qYdI8P0yrF3pdXlAAAAAIBlgmrad8UVV+iTTz7Ryy+/rFdeeUWpqakyDEPV1dUyTVMjR47UVVddFepaEWFcDpuyk50qq2tScU2jMpOdVpckqXlZv71oh5y7l6upcKTV5QAAAACAJYIK/DabTQ899JAWLlyoN998Uzt27JBpmho2bJjGjx+vCy+8UDZbUIsHEGXyUxNUVtekfdWNGpCfanU5kpoDf2LRi3LuXm51KQAAAABgmaAC//3336+zzjpLkydP1uTJk0NcEqJJQVqCNuyriczGfcWfS54GyZFocUUAAAAA0PmCmoafP3++Vq5kfzQOdOqPpEfzeTN6y5tcIMPnlnPfaqvLAQAAAABLBBX4k5OTZbfbQ10LolB+6oFO/RHDMA7M8u/i8XwAAAAA4lNQgX/27Nl69NFH9dxzz6m4uFg+ny/UdSFKBGb4a9wWV9JaIPCzjx8AAABAnApqD/+8efPkdrt1xx136I477jjsOYZhaP369R0qDpGvoCXwR9QMvw4K/Hs/k7xNkj0yniAAAAAAAJ0lqMCfmZmpzMzMEJeCaJSfeiDwm6YpwzAsrqiZN7u/fAmZsjVWyFGyTp4uw6wuCQAAAAA6VVCB/6mnngp1HYhSeS17+Bs8PlU3epSeGCEz6YZNTd1OVcLWt+TcvZzADwAAACDuBLWH/3BKS0vl8XhCdTlEiUSnXZlJzSE/kjr1Swct69/DPn4AAAAA8eeoAv+LL76oyZMnH7ZJ37333quRI0fqueeeC1lxiA4HOvVHWuO+0yRJzt2fSj6vxdUAAAAAQOdqd+D/y1/+ot/85jfatGmTtm3bdsjrDodDlZWVuuOOO3T33XeHskZEuAOd+iNrht+T+z35nKmyuatk37/B6nIAAAAAoFO1K/C///77mjt3rk455RS9+eab6tOnzyHn/OlPf9I777yjE044Qf/617+0dOnSkBeLyBSpnfplc8jTdbgkybV7mcXFAAAAAEDnalfgf+aZZ9S9e3fNnTtXPXr0aPO8rl27au7cucrKytLTTz8dsiIR2Q7u1B9p3P5l/ezjBwAAABBn2hX4161bpylTpsjp/O4O7GlpaZo8ebLWrFnT4eIQHfLTWvbwR9iSfumgxn27l0umaXE1AAAAANB52hX4a2trVVBQ0O6L9u7dW9XV1UEXhehyYIY/spr2SZInf7BMe4Js9ftlr9hsdTkAAAAA0GnaFfjz8vJUUlLS7ouWlZUpJycn6KIQXfx7+IurG2VG2iy6PUFNXYZJkpzs4wcAAAAQR9oV+AcNGqQlS5a0+6Jvv/22evXqFWxNiDL+Lv11TV7VuiPv8XcHHs/HPn4AAAAA8aNdgX/KlCn68ssvNW/evO88d968eVq/fr2mTJnS0doQJZKcdqUnOiQ1z/JHmgOBfxn7+AEAAADEDUd7Tho9erTOOecc3XvvvVqzZo1+9KMfaciQIXI4mt/e1NSkzz//XP/617/09ttv6+STT9b5558f1sIRWfJTE1TV4NG+mkb1zU2xupxWmgqGybQ5ZK/ZI1v1DvnSC60uCQAAAADCrl2BX5L+9Kc/yWazafHixXrzzTdlt9uVmZkpr9erqqoq+Xw+maapcePG6Y9//GPgywDEh/w0lzaV1kbko/nkTJInf4ice1fKuXu5Ggn8AAAAAOJAu1N5SkqKHnjgAX388cd69dVXtW7dOu3bt092u129evXSySefrIkTJ2r48OHhrBcRKpI79UvNj+drDvzL1DjwEqvLAQAAAICwO+pp+BEjRmjEiBHhqAVRLNCpvyYCZ/glNXU9VdJDNO4DAAAAEDfa1bQP+C75Bz2aLxI1dT1ZpmGTo3KbbLV7rS4HAAAAAMKOwI+QKAgs6Y/MwG8mpMuTe7wkHs8HAAAAID4Q+BES/hn+fRG6pF9q3scvEfgBAAAAxAcCP0IiP80lSapp9KrW7bG4msMj8AMAAACIJwR+hESKy6EUl12SVBKpnfq7niJJcpQVyagvs7gaAAAAAAgvAj9CJj/CO/WbSTnyZPWXJCVsfdPiagAAAAAgvAj8CBn/o/kitXGfJDUc9wNJUvJnf5O8kbkSAQAAAABCgcCPkPF36o/UR/NJUv2gq+RNzpe9eqcSv3re6nIAAAAAIGwI/AgZf+O+SO7UL2eS6ofdJElKXvn/JE+DxQUBAAAAQHgQ+BEy+an+Jf2RvVS+/vgr5E3pInvNHiWuf8bqcgAAAAAgLAj8CBl/076InuGXJEei6obfIklKXjlHaqq3uCAAAAAACD0CP0ImPwqa9vk1HPdDedN6yF63T0lfzLe6HAAAAAAIOQI/QsbftK+ywaOGJq/F1XwHu0t1w38iSUpe9XfJXWtxQQAAAAAQWgR+hExqgl3JTrskaV9NZO/jl6SGARfLm95TtoYyJa17wupyAAAAACCkCPwIGcMwAp36i6ujoPu93anaU34qSUr+/GEZ7mqLCwIAAACA0CHwI6SipVO/X2O/KfJkHStbY6WS1jxudTkAAAAAEDIEfoRU1HTq97PZVXfybElS0urHZDRUWFsPAAAAAIQIgR8h5Q/8xVHQqd+v8diJ8mQPkM1dpaTVj1pdDgAAAACEBIEfIVWQ2ryHPxoezRdg2FR7ys8kSUlr58qoL7O4IAAAAADoOAI/QqogLVFSdHTpP5i7z3lqyh0kW1Otkj9/yOpyAAAAAKDDCPwIKX+X/qia4Zckw1DdqT+XJCWtmyejdp/FBQEAAABAxxD4EVL+Lv3l9U1q9PgsrubouHueraaCoTI8DczyAwAAAIh6ERf4y8vLddddd2n06NEaPHiwLrjgAr344ovteu9nn32mAQMGHPbXtGnTWp3r9Xo1b948nXfeeRo8eLDGjBmj+++/Xw0NUfD8+AiWnuhQgqN5WJVES6d+P8NQ7Skts/xfPCVbzR6LCwIAAACA4DmsLuBgdXV1mjFjhjZu3KjLL79cffr00Ztvvqnbb79dpaWlmjlz5hHfX1RUJEmaPXu2unTp0uq13NzcVr+/4447tGDBAo0fP15XXnml1q9fr0ceeURffPGFHn/8cRmGEdoPFycMw1BBWoK2l9eruLpRPTKTrC7pqDQdc5aaup4i555PlbxyjmpG/sHqkgAAAAAgKBEV+J9++ml9+eWXuu+++3T++edLki699FJde+21mjNnji688EJ17dq1zfcXFRXJMAxNmzZNycnJbZ63du1aLViwQJdeeqnuvPPOwPEePXrovvvu0+LFizVhwoTQfbA4k5/q0vbyeu2Lthl+qXmW/9SfK/PVHyhx/TOqG3qDfOk9rK4KAAAAAI5aRC3pf/XVV1VQUBAI+1LzjPGPf/xjNTU1adGiRUd8f1FRkXr06HHEsC9JL7/8siRp+vTprY5fddVVSkhICLyO4OSnNe/j31cdXZ36/Zq6f1/u7t+X4WtS8sr/Z3U5AAAAABCUiAn81dXV2rJli4YMGXLIa/5ja9eubfP9pmlq48aN6tevn6TmPfr19fWHPXfNmjXKzMxUr169Wh1PTExU//79j3gffLeCQOCPwhn+FrWn/kKSlPjV87JVbrO2GAAAAAAIQsQs6S8uLpZpmoddsp+UlKSMjAzt3Lmzzfd/8803qqurU11dnX70ox9p9erVampqUt++fTVr1qxWS/T37t3b5taALl26aN26daqurlZaWlqb94v0Lf7++qyo0x/4i2saI/7PqS3ebifLXThSru0fKmXFA6oZ94DVJcUVK8cv0FGMX0Qzxi+iGeMX0Sxc4zdiAn91dbUktbkcPzExsc0Ze+lAw761a9fq6quv1vTp07Vnzx7NmzdPP/3pT1VeXq4rrrgicK/evXu3eR+puYFgW4E/OztFdnvELI44opyctr+0CJdju2VKksrqPcrN7fz7h8w5/ys9PkaJG19W4rhfSbn9rK4o7lgxfoFQYfwimjF+Ec0Yv4hmoR6/ERP4TdNs9d/DvW6ztR2ye/bsqVmzZmnkyJEaPHhw4PiFF16o888/X3/96181adIkpaenf+d9JMlut7d5r7Ky2oj/5tAwmgfL/v3VauOjhk2i6ZMk7S6vU2lpdefePJQS+yut1zglbHtHDW/dpZrxf7e6orhh5fgFOorxi2jG+EU0Y/wimrVn/AYzmRoxgT8lJUWS1NDQcNjXGxoajtihf+DAgRo4cOAhx1NTUzV16lQ99NBDWrlypUaPHq2UlJQj3kfSEZfzS4qaf0RMs/NrzU91SZL21zXJ7fHJGSWrIQ6n9pSfK2HbO0r4+t+qO+lmeXMOHWMIHyvGLxAqjF9EM8YvohnjF9Es1OM3YpJYjx49ZBiG9u7de8hrdXV1qqqqUpcuXYK6dm5uriSptrZWktS9e/fD3kdq3t+flZWlhISEoO4FKTPJKae9eQlESU10dur38+Ydr8a+E2TIVMqK+6wuBwAAAADaLWICf0pKivr27at169Yd8tqaNWskScOGDWvz/XfccYfGjBmjXbt2HfLapk2bJEmFhYWSmrv+l5WVaceOHa3Oq6+v18aNGzV06NCgPweaH6WYnxr9nfr9ak/+mUwZStj8huwlX1pdDgAAAAC0S8QEfkm64IILtGvXLr3++uuBY6Zpau7cuXK5XK067X9b165dtWvXLj355JOtjm/evFkvv/yy+vXrpxNOOEGSNGnSJEnSY4891urc+fPny+12a+rUqaH6SHEr8Gi+mugP/N6cAWrsd4EkKeXTv1pcDQAAAAC0T8Ts4Zekq666Sv/+97/1y1/+Ul988YV69+6txYsX65NPPtGtt96q/Px8SdKOHTu0atUqFRYWBmbjp02bpjfeeENPPvmk9u7dq9NPP127d+/WM888I4fDoXvuuUdGS6e9YcOGaerUqVqwYIEqKys1YsQIrVu3Ts8//7xGjx6tsWPHWvZnECvy/Y/mi4EZfkmqO3m2EjYtUsK2d+Qo/lyeAlaBAAAAAIhsERX4ExMT9dRTT+m+++7TwoULVVtbq969e+vee+/V5MmTA+etWLFCt912m6ZMmRII/ElJSXr66af18MMPa/HixVqyZInS09M1cuRI3XzzzYc8hu+uu+5SYWGhXnrpJS1ZskRdunTRDTfcoOuvvz7wxQCC51/SHyuB35vVV439pyqx6EWlfPp/qpz0tNUlAQAAAMARGWZbz6dDm0pKIv9Rc4bR/NiG0lJrHkvy/Oe79Jf3Nmt0v1z9+YLvdX4BYWCr3Kbsf42UYXpVPvVVeboOt7qkmGX1+AU6gvGLaMb4RTRj/CKatWf85uUd/WP5ImoPP2JHLDXt8/Nl9FLDcT+QJKUs/4vF1QAAAADAkRH4ERb5MdS072B1J/1Eps0p166lcu76xOpyAAAAAKBNBH6EhT/wl9a45fH6LK4mdHzpPdTwvcslSSnL/yrWiwEAAACIVAR+hEV2slMOmyFTUmmt2+pyQqrupFky7Qly7vlUzh0fWV0OAAAAABwWgR9hYTMM5ae6JMVOp34/X2pX1Q+aJqllLz+z/AAAAAAiEIEfYXNgH39szfBLUt3QG2U6EuXct1qub5ZYXQ4AAAAAHILAj7CJxU79fmZKvupPuFqSlMxefgAAAAARiMCPsInVTv1+dUNvlM+ZImfpF3JtfdPqcgAAAACgFQI/wiYQ+GNwhl+SzKRs1Q+eIcnfsT92nkYAAAAAIPoR+BE2BS2Bv7g69vbw+9WfeJ18rnQ5yoqUsOk1q8sBAAAAgAACP8KmoKVLf6wu6ZckMzFT9SdeK0lKXnGf5PNaXBEAAAAANCPwI2z8S/pLaxrl9cVuU7v6IT+WLyFTjvJNSvj6FavLAQAAAABJBH6EUXayS3ZD8prS/trYXdZvutJUN3SmJCnl0/slb5PFFQEAAAAAgR9hZLcZyk2N7U79fvUnXCNfUo7sVd8oseglq8sBAAAAAAI/wis/NbY79Qe4UlQ39EZJUvJnD0je2F3RAAAAACA6EPgRVgVpzY37imtiPwDXD7pS3uR82at3KvGrBVaXAwAAACDOEfgRVv7GfTE/wy9JziTVnTRLkpT82f+TPA0WFwQAAAAgnhH4EVYF8RT4JTV873J5U7vKXrtXSV/+y+pyAAAAAMQxAj/Cyr+HvzhOAr8ciao76SeSpOSVc6SmeosLAgAAABCvCPwIq8CS/hjv0n+whuN+IG/aMbLVlyjpiyetLgcAAABAnCLwI6zyU5ub9u2rcctnmhZX00nsLtWe/D+SpORVD0nuWmvrAQAAABCXCPwIq9zUBDlshrw+U9vK6qwup9M0DrhInoxesjWUKWndE1aXAwAAACAOEfgRVg6boVN6ZkqS3ttYam0xncnmUN3JP5UkJX/+sIzGKosLAgAAABBvCPwIu7H98yRJ724ssbiSztXYb7I8Wf1ka6xU0prHrS4HAAAAQJwh8CPsRh6bI4fN0ObSOm3dHz/L+mWzq+7k2ZKkpDWPyWgot7ggAAAAAPGEwI+wS0906rReWZLicJb/2PPlyTlONne1klY/anU5AAAAAOIIgR+d4uz+uZKkd4viK/DLsKn2lJ9JkpLXzJVRv9/iggAAAADECwI/OsXIvrly2Axt2V+nzaXx9Zg6d+/xasobLMNTp+TPH7a6HAAAAABxgsCPTpGW6Ags618SZ8v6ZRiqa5nlT1o3T0btPosLAgAAABAPCPzoNOMG+Lv1x9Hj+Vq4e45RU8EwGZ4GJa/6u9XlAAAAAIgDBH50mrP65shpN7Q1Dpf1yzBUe+rPJUlJXz4tW81uiwsCAAAAEOsI/Og0qQkOndazpVt/vDXvk9TU40y5u54qw9uo5JVzrC4HAAAAQIwj8KNTjQ0s6y+RaZoWV9PJDEN1LbP8ieufla1qp8UFAQAAAIhlBH50qrP65shlN7StrF6b99dZXU6na+p+utw9RsjwNSn5swesLgcAAABADCPwo1OlJjh0Wq9sSfG5rF+Sak/9hSQpccMLslVus7YYAAAAADGLwI9ON3ZArqTmwB93y/olebqcpMbC0TJMr1JWPGB1OQAAAABiFIEfne7MPs3L+r8pr9emeOvW38K/lz9h48uyl2+yuBoAAAAAsYjAj06XmuDQ93u3LOvfWGpxNdbw5A9RY+/xMkyf0t+6QUZjldUlAQAAAIgxBH5Y4uz+Ld3643RZvyTVjPi9fEl5cuz/SumLfyx5G60uCQAAAEAMIfDDEmf2zZbLbmh7eb2+LonPZf2+9GNUOekp+Zwpcu36RGlLZkumz+qyAAAAAMQIAj8skeI6eFl/fHbrlyRP3iBVnfeYTJtDiV8vVMrSu60uCQAAAECMIPDDMmNblvUv2Vgat8v6JanpmLNUPeY+SVLymkeVtPpRiysCAAAAEAsI/LDMiL7ZSnDYtL28XhvjdFm/X+OAqao5/XZJUurSO5Ww8VVrCwIAAAAQ9Qj8sEyrZf1F8bus369+6EzVDZ4uSUpb8lM5dy61uCIAAAAA0YzAD0uN7Z8rqXkffzwv65ckGYZqR/xeDX0nyvA1Kf2NGbKXrre6KgAAAABRisAPS43ok6MEh007Kxq0cV98L+uXJBk2VY99QO5up8nWVKOMRdNkq9phdVUAAAAAohCBH5ZKdtl1Rsuy/nfiuFt/K45EVU2YK0/OQNnripWx6EcyGsqtrgoAAABAlCHww3JjBzR363+3iGX9fmZChionPiVvajc5KjYr4/WrpaZ6q8sCAAAAEEUI/LDciD7N3fp3VTZow74aq8uJGL7Urqqc+JR8CRly7l2p9Ldvknweq8sCAAAAECUI/LBcktOuEX383fpLLa4msnhzBqhywhMy7QlK2Pa2Uj+8XWIVBAAAAIB2IPAjIozt37Ksn279h/B0O0VV4x6UKUNJ6/+l5M/+n9UlAQAAAIgCBH5EhDP6ZCvRYdPuygZ9Vcyy/m9z952gmrPuliSlfPpXJa5/1uKKAAAAAEQ6Aj8iQvOy/hxJzc37cKiGE65S7Uk3S5JSP/iVXNuWWFwRAAAAgEhG4EfEGDsgV5K0hGX9bao79VY1DLxEhulV+lvXy7F3ldUlAQAAAIhQBH5EjDN6tyzrr2rUepb1H55hqHrUn+UuHCXD06CM16+SvWKL1VUBAAAAiEAEfkSMRKddZ/ZlWf93sjtVOf4RNeUPka2hXBmLfiSjdp/VVQEAAACIMAR+RJSxA5q79bOs/zu4UlR5/pPypveUvWq7Ml6/SoabVREAAAAADiDwI6J8v1eWkpw27alq1Pq91VaXE9HM5FxVTHpavqQcOUvWKf3N6ySv2+qyAAAAAEQIAj8iSqLTrjNbuvW/U1RqcTWRz5fZW5XnPynTkSTXjo+U9t7PJdNndVkAAAAAIgCBHxHHv6z/XZb1t4un4ERVnfuITMOuxI0vK+W/f7K6JAAAAAARgMCPiHN6rywlO+0qrm7UF3tY1t8e7p5jVD36L5Kk5M8fVtKauRZXBAAAAMBqBH5EnOZu/dmSmmf50T6Nx/1Ataf+UpKU8vHv5dr0msUVAQAAALASgR8RaWx/f7f+UvlY1t9udSfNUv2gq2TIVPo7t8i5679WlwQAAADAIgR+RKTTe2ezrD8YhqGaM+9UY59zZfjcSn9jhuz7v7K6KgAAAAAWIPAjIiU4bDrr2OZu/e8Wsaz/qNjsqhr3oJq6niybu0oZi6bJVr3b6qoAAAAAdDICPyLW2P65kqQlG0tY1n+0HEmqnPBPebL6yV67VxmLfiSjodzqqgAAAAB0IgI/ItZpvbKV4rJrX41b63ZXWV1O1DETs1Q56Wl5U7rIUb5RGW/MkDz1VpcFAAAAoJMQ+BGxEhw2ndW3ZVn/xlKLq4lOvrTuqpz0lHyuNDn3fKr0d26RfF6rywIAAADQCSIu8JeXl+uuu+7S6NGjNXjwYF1wwQV68cUX2/Xe+vp63X///Ro/frwGDRqkk08+Wdddd53WrFlzyLkLFy7UgAEDDvvrV7/6Vag/FoI0doC/Wz/L+oPlzTlOVRPmyrS5lLBlsVI//p3EnyUAAAAQ8xxWF3Cwuro6zZgxQxs3btTll1+uPn366M0339Ttt9+u0tJSzZw5s833mqapm266SUuXLtW5556rq666SmVlZXr22Wd1xRVX6LHHHtPpp58eOL+oqEiSdPfdd8vlcrW6VmFhYXg+II7aaT2zlOKyq6RlWf+Q7hlWlxSVmrp/X9Vj/5/S3r5RSeuelDelq+pPmmV1WQAAAADCKKIC/9NPP60vv/xS9913n84//3xJ0qWXXqprr71Wc+bM0YUXXqiuXbse9r2vv/66li5dquuvv16zZ88OHL/ooos0adIk3X333Xr99dcDx4uKipSTk6NLLrkkvB8KHeJy2DTy2By9sX6f3ikqIfB3QGO/SbLVFSv1498rddk98qUUqHEg4x8AAACIVRG1pP/VV19VQUFBIOxLkmEY+vGPf6ympiYtWrSozfcuXbpUknTZZZe1Ot61a1edcsop2rRpk8rKygLHi4qK1K9fvxB/AoTD2P7+Zf2lLOvvoPohP1bd0OaVMmnv/VzOb963uCIAAAAA4RIxgb+6ulpbtmzRkCFDDnnNf2zt2rVtvv/WW2/VSy+9pC5duhzy2v79+yVJdrs98PuSkpJA4He73XK73R3+DAiPU3tmKTXBrtJat9bsolt/R9We/ms19J8iw/Qq483r5ShebXVJAAAAAMIgYgJ/cXGxTNM87JL9pKQkZWRkaOfOnW2+PysrS4MGDZJhGK2Or1y5UqtXr9bAgQOVkdG8HHzDhg2SpD179mjq1Kk68cQTNXjwYF188cX673//G8JPhVBwOWwa2dKtf8nGEouriQGGTdVj/k/uHmfK8NQp4/WrZKvcZnVVAAAAAEIsYvbwV1dXS5KSk5MP+3piYqLq64/uGeLFxcX6xS9+IUm6+eabA8f9DftWrlyp6dOna9asWdq2bZvmzp2rGTNmaM6cORozZswRr/2t7xUijr++SK+zvcYNyNPr6/dpycZSzR7dV3ZbjHwwqzhcqp7wqDJevliO0i+VuegKVVy0UGZyrtWVSYq98Yv4wvhFNGP8IpoxfhHNwjV+Iybwmy17s8029mibpimbrf0LEnbu3Knp06dr165dmjFjhsaOHRt4bfDgwZo5c6amTp2qnj17Bo6PHz9eEydO1B133KFRo0a1eb/s7BTZ7RGzOOKIcnLSrC4hJCZkpui3i4tUWuvWtpomndonx+qSYkCadNUr0tyxsld8o5w3r5Guek1KSLW6sIBYGb+IT4xfRDPGL6IZ4xfRLNTjN2ICf0pKiiSpoaHhsK83NDS02aH/29auXasbb7xRJSUlmj59um699dZWrw8fPlzDhw8/5H3du3fXuHHjtHDhQm3atEn9+/c/7PXLymoj/ptDw2geLPv3V8fMI9dH9c3Roi+L9eKn29U33fXdb0A7JMt2/tPKfOlC2XZ/Ls/j41Uz6l55Cg7tpdGZYnH8In4wfhHNGL+IZoxfRLP2jN/c3KP/MiBiAn+PHj1kGIb27t17yGt1dXWqqqo6bEO+b3v33Xf185//XA0NDbr11ls1Y8aMo6ojJ6d55ri2tvaI50XLPyKmGT21fpez++dp0ZfFeu/rUv2MZf0h483so8rz5ynjtSvlKPlCGS9MVMMJV6n21FtlJqRbWlssjV/EH8YvohnjF9GM8YtoFurxGzHr0lNSUtS3b1+tW7fukNfWrFkjSRo2bNgRr/HWW2/plltukdfr1QMPPNBm2L/xxhs1bty4w64m2Lx5sySpsLDwaD8CwuyUnplKS3Bof61bq3dVWl1OTPF0OUlll72vhn6TZchU0rp5ynpmtBK+XsT/YgIAAABRKmICvyRdcMEF2rVrl15//fXAMdM0NXfuXLlcLk2YMKHN927YsEG/+MUv5HA49Pjjj+vcc89t89y8vDxt375dCxYsaHV8+fLl+uijjzRy5MjATD8ih9Nu06hjm/9e3i2iW3+omSn5qj5njioueFaejN6y1xUr/e0blPHaNLr4AwAAAFHIMNvqkmeBhoYGXXTRRfrmm280bdo09e7dW4sXL9Ynn3zSann+jh07tGrVKhUWFmro0KGSpBkzZujjjz/WqFGj2vxiYNy4cUpOTlZpaakuueQSFRcXa/LkyRo8eLA2bdqk5557TtnZ2XrmmWfUo0ePNussKakO/YcPMcNo3uNRWhpbe5g+2Vqmn7z8hbKTnXrj+tNY1h8ungYlr3pIySvnyPC5ZdoTVDf8J6ober1kTwj77WN1/CI+MH4RzRi/iGaMX0Sz9ozfvLyj38MfUYFfksrKynTffffpvffeU21trXr37q2rr75akydPDpzz8ssv67bbbtOUKVN0zz33yOPx6MQTT1RTU9MRr71kyZJAkC8tLdXf/vY3ffDBB9q/f7+ys7M1atQozZo1SwUFBUe8DoHfOh6vT+P/sUxVDR794weDddIxmVaXFNPsFVuU+uGv5dr5sSTJk3Wsakb+UU3dvx/W+8bq+EV8YPwimjF+Ec0Yv4hmcRP4owGB31p3vVWkf39RrIuGdNWvxvazupzYZ5pK+PpVpX58h2z1pZKkhgEXq+aM38pMCs/Wl1gev4h9jF9EM8YvohnjF9EsXIE/ovbwA+0xdkCeJOn9r0vl8fGvedgZhhr7T1HZ5R+o/vhpMmUosehFZf/rLCWuf0YyfVZXCAAAAOAwCPyIOicfk6mMRIfK6pr0+c4Kq8uJG2ZipmpG/UkVF70qT873ZGusVNr7tyrz5amy7//K6vIAAAAAfAuBH1HHYbdp1LG5kqQlG0strib+eLqcpPIfvKGaM/5XpiNZzr2fKev585TyyR+kpjqrywMAAADQgsCPqDR2QHPgf28jy/otYXOo/sRrVXb5B2rsc64Mn0fJnz+s7GdGy7X1HaurAwAAACACP6LU8JZl/eX1TVq1o8LqcuKWL62bqs57XJUTnpA3tbvsNbuU8cY1Sl/8Y9mqd1tdHgAAABDXCPyISg67TaP7Nc/yv7uxxOJq4O49TmWXv6+6oTfItDmUsOVNZT8zSkmrH5V8HqvLAwAAAOISgR9Ra2x/f7f+/SzrjwTOZNV+/3aV/2CxmroMl+GpU+rSO5X1/AQ59q6yujoAAAAg7hD4EbVOKmxe1l9R36SVLOuPGN6c41Qx9WVVj/6LfAmZcuxfr8yXLlTqB7fJaKiwujwAAAAgbhD4EbUcNkNj+rcs6y9iWX9EMWxq+N5lKrviQzUMvESGTCV9+ZSynxmlhI2vSCYrMgAAAIBwI/Ajqh1Y1l8qj9dncTX4NjMpR9Vn36+Kyc/Lk3WsbPWlSn/nZmX8+zLZK7ZYXR4AAAAQ0wj8iGrDjslUZpJTlQ0erdxRaXU5aENT9++r/NK3VXvqrTLtCXLt/FhZz45V8qf3SZ4Gq8sDAAAAYhKBH1HNYTM0pqVb/zt0649sdpfqht+issuWyF04UobPrZQV9ynruXFy7viP1dUBAAAAMYfAj6g3dkBz4P+AZf1RwZfRS5UTn1bVOQ/Lm1wgR+VWZf77MqW9PUtG7T6rywMAAABiBoEfUW9oj0xltSzrX0G3/uhgGGrsN0nll7+vuhOukSlDiV+/quxnRinxi/mSyRc3AAAAQEcR+BH1Du7Wv6So1OJqcDTMhHTVnnWXKi55TU15J8jmrlLah79WxosXStuXEfwBAACADiDwIyYEuvVvYll/NPLkD1HFxa+p+sw75XOmyln8ufTP8cp64mSlvn+rXNvelTz1VpcJAAAARBWH1QUAoTC0R4ayk50qq2vSp9sr9P3e2VaXhKNls6th8HS5+05QyrJ7lbjlTdnripW0/hklrX9GpiNR7mNGyt1rnBp7nS0zOc/qigEAAICIRuBHTLDbDI3ul6uX1uzRu0UlBP4o5kvpopqx9ysx8++qXPuOXFvflmvrO7LX7FbC1reUsPUtpcqQp8swNfYaJ3evcfJm95cMw+rSAQAAgIhC4EfMGDcgTy+t2aMPN+9Xk9cnp50dK1HNkaCmwpFyHzNSOvNu2UvXK2Hb23Jte1fOfWvk3LtSzr0rpWX3yJveU429m8N/U9dTJLvT6uoBAAAAyxH4ETNO7N56Wf8ZzPLHDsOQN+941eUdr7qTfypbzR65ti2Ra9vbcu1cKnvVN0pe87iS1zwuX0KG3IWj5e59jtyFo2QmpFtdPQAAAGAJAj9iht1m6Oz+eXph9W69W1RC4I9hvtSuahj0IzUM+pHkrpVr50dK2PqOXNvela2hTIlfv6rEr1+VaXOoqdtpzfv+e4+TL73Q6tIBAACATkPgR0w5u3+uXli9Wx9uYll/3HClyN3nPLn7nCf5vHIUr1LC1ual/47yr+Xa+bFcOz9W6sf/K0/OQDX2Okfu3uPkyR8iGYwPAAAAxC4CP2LKid0zlJPi0v5atz79pkJn9GGWP67Y7PJ0PVmerier9vu3y16xRa5t78q19W0593wqx/4NcuzfoJSVf5M3OV/uXmfL3escuXuMkJxJVlcPAAAAhBSBHzHFbjN0dr9cPb96t97ZWELgj3PezD6qP/E61Z94nYyGcrm+eU+ure/Itf0D2ev2KWn9s0pa/2zzI/96nCV373Fq7DWWR/4BAAAgJhD4EXPGDsjT86t368NNpXJ7+snlYNk2JDMxS40DLlLjgIskb6Ocu5YpYds7LY/826WEbW8rYdvbzY/8Kxiqxt7n8Mg/AAAARDUCP2LOkO7pyk1xqbTWreXflOvMvjlWl4RIY29+5F9T4UjpzLtk3/9Vy77/d5of+Ve8Ss7iVS2P/CuUu+doNXUZrqaCYc2N//gCAAAAAFGAwI+YYzMMnd0/Vws+360lG0sI/Dgyw5A393uqy/2e6k7+n8M88m+7ktY9qaR1T0qSfInZauoyTJ6CoWoqGCZP/hAe/QcAAICIROBHTBrbP08LPt+tDzbtl9vjY1k/2u1wj/xz7vqvnHtXyVH6pWwNZUrY9q4Str0rSTJlyJvVT00FQ+Xp0vwlgDe7v2Tjn1cAAABYi/+PFDFpcPd05aW6VFLj1rJvynUWs/wIxsGP/JMkT4McpV/KWfy5HMWfy1n8uexV2+Uo3yhH+UZpwwJJkulIVlP+YHm6DGv+IqBgqHwpXSz8IAAAAIhHBH7EJJthaEy/5mX97xaVEPgRGo5EebqcJE+XkwKHjLoSOYtXy1G8quWLgNWyNdXItXuZXLuXBc7zpnZr3gJQMLR5S0DeIMnBowABAAAQPgR+xKxxA5qX9X+0eb8aPT4lsKwfYWAm58nde5zcvcc1H/B5ZS/fJGfxqpZVAKtkL9soe81u2Wt2S5tfa36fzSFPzvdavgAYKk/BMHkzetMQEAAAACFD4EfMOqFbuvJTXdpX49aybeUaeSyz/OgENru8OQPkzRkgfe8ySZLhrpFj35rANgBH8eey1+2Ts2StnCVrlfRFS0PAhEx5Ck5UU4F/K8CJMhOzrPw0AAAAiGIEfsSs5m79eXp21S69u7GEwA/LmK5UNfU4Q009zlC9JJmmbNW7WsJ/y1aAknWyNVbItf0DubZ/EHivJ7OPPP4vALoMkyd7oGR3WvVRAAAAEEUI/IhpZ/fP1bOrduk/LOtHJDEM+dJ7qDG9hxr7TWo+5nXLsf+r5lUAe5u3Azgqt8pRsUWOii1KLHpRkmTaE+TN6CVvZp/mXxm95Wn52UzKYUsAAAAAAgj8iGmtl/WXaeSxuVaXBBye3SVP/hB58oeo4YSrJUlGQ3nz7P/eVXLua2kI2FgpR1mRHGVFh1zC50qXN7P3gS8DDvpSwHSldvIHAgAAgNUI/IhpNsPQ2AF5emblLr1TVELgR1QxE7Pk7jlG7p5jWg74ZKvaIXvLrL+9covsFVtlr9giW/Uu2dxVsu1bI+e+NYdcy5ucf+DLgIyDvwwolOwJnfzJAAAA0BkI/Ih5Y/s3B/7/bC5TQ5NXiU671SUBwTFs8mX0lC+jp5p6jm79mqdB9spvWr4E2NLypUDLlwH1pbLX7ZO9bp+0e3mrt5mGTb60HvJm9pYno/XKAF9qN8nG/70AAABEKwI/Yt6grmkqSEtQcXWj/rutXKP7McuPGORIPPB0gG8xGqtkr9wa+CKg+VfLlwFNNbJXbZe9artc+rDV+wL9Ag7uGZDZR57MPjKTcukXAAAAEOEI/Ih5hmHo7P65emblLi3ZWELgR9wxE9ID/QFav2DKqCuR43BfBlRuk+FtPEK/gDR5M3rLl9pVvuS8ll/5Lb/8P+dKjsRO+pQAAAD4NgI/4sK4ln38H23ez7J+wM8wZKbkqyklX03dTm39ms8rW82uwEoAe8UWOVp6BtiqdsjmrpatZK1UsvaIt/AlZHzrC4Fv/9z8XzMxm+0DAAAAIUbgR1w4vkuauqQlaG91oz7ZVq4xzPIDR2azy5deKF96oZoKR7Z+zdsoe+V22Su3yla7T7a6YtnqSlp+7Qv8bHgbZWuslK2xUirfdMTbmYZNvqTc5vB/yJcDBfKlHPTlgDOV7QQAAADtQOBHXGhe1p+nf63cqXeLSgj8QEfYE+TN7idvdr+2zzFNGe6qQ74EOPDzvpYvC0pk1O+XYfoONBb8DqYj8fCrBRKzZLrSZCZkyHSlyZeQLtOV3nzMlSoZthD+IQAAAEQ+Aj/ixrgBufrXyp36eMt+1bm9SnaxfBgIG8OQmZAhb0KGvFnHHvlcn0e2+v2BLwKMw3054D/mrpbhaQg0GmwvU0bLlwHph34ZkJAunyv9oNcyWr4kaP7ywNdyDv0IAABAtCHwI258r0uauqUnaHdVoy6d95l+fHqhzj++ixw2lgYDlrI55EspkC+l4LvPbaqXrb6kZXXAt74QaKyQ0Vgtw10lo7FKNndV8+99bhlqXnEgd1XQZZo2V8uXA/4vDtJlJqQ1f1lw0BcGZkK6lJMrZ51Ppj1Rpj1BciQ0v9+RINkTZDpajttdko3/KQYAAOFhmKZpWl1EtCkpqba6hO9kGFJubppKS6vF3/ABn35Trt+/WaSSGrckqTArSded3lPjBubJxp7giMH4RUh5Glq+AKiW0Vgpw/3/27v36CjKO27g35m9JiEEIeGWKAQk0KIx3qAC2iMCViAInlYsWKAEECqCrQeQQ0urqNUW01PFtjTQUy6+GMEIiII9BXteqEpFyv0taJFbIJQYLiG33Z153j/msjN7CUES9pLv55ycnXkuM8/sPj7y/PaZ2WrIDZe0wICv2hIciLLvq4aEluuIQnLoQQCPHgTQXs3ggMMD4XBrQQOHB8Lh1bfdUcp6LGX1bVea9udMhXClQrjSAIerxa6JEhfHX0pk7L+UyJrSf7Oy0q/+uJzwXz1O+BNbvV/BO3vP4K//OokLdX4AQK+sNDwxoDvu69keEif+Mcf+S3FFqJB8l/VgwEUtcBApeKAHCGTfJbhFPfz1dZCUBkiBekD1QQo0AEqDlqYGYn1V2ooDVzAAEP01DcKZEtw2AgaRyjtT+WsLCY7jLyUy9l9KZJzwxxFO+JNDjS+ANZ+XY/WuU6jxKQCAW7qkY8bA7ujX7YYYt651Y/+lRNak/qsqgOKDpNRDUhqAgB4IUHxaUCCgpxtpZn6DtmLBVrbBUrYhpGyDdg5/LSR/jfaq+lv0+oXT20igIFW/ncEbXIXgTLHf5qC/CqdRxquvfrDuewGHh8GFFsDxlxIZ+y8lspaa8PPGQWq10txOTLmnG35Q0BWrdp1C6e5yHDhTjSfX7cddN2ZgxqBc5HdtG+tmElEykh2AnALhSmnBmwWiUHx6AMAIAtRACoTsh+b7awFr0MCsY9kXKgBowYpAPVD3dYtfipBd4UECI3hgCxIYed6QwIIbkF0QDpfl1R3y6gqWk52Aww0huwCHC0J2214hOfiTkUREFFc44adWLyPFhZn35uKxO7Lx150nULbvDHadvIiiNXswqEd7TB/YHb07tol1M4mImofDDeFwQ3jbNd8xhQCUekj+OkuQoCZkXw8MWFcpBOq1evpKBHNFQsDIs6x2MOpYVihIql/b919uvmu5BgJSMCAgO0MCCsEAgj1oEAweGA9y1J7J4NYf9OjVPzNP8LMzn/XgDj4Q0nx2g9sszwdDEhERl/R/A1zSn9zOXKrHsk+O4/2DZ6Ho792QvCw8MbAburdPjW3jWgn2X0pk7L8tTFWCz0II1NmDBEbQIDRIYGwH6u2BBcUPqH7t9gjVr+/7QtID0fNb+PaI5qI9GNIeBBBOjx6EsPx6hP4gSE9qKup9KiA5ICRZW7kgOwBJ1o4lOQBZSxeyEzDKSA4IObgNSYaQg9tavnXbUrex4xurLPQgCBzavratBzS4soLA8ZcSG+/hjyOc8LcOx6tq8eePj+Nvh88BAGQJGP7tTph6Tzd0zeDvcbck9l9KZOy/rYgQgOoHFD8k1ae/+rXbJtRASDChkaCC4gvWMwIUis/+/AbV2PZZnv/gM+uE1dVvsWgNzJUV5ooKd+R9PcBx5bIefeWFsaLCkmcGMLTbN4RlG7IDArIZvDADG5ZtrYykP39CDgY2zKCJHAx8yJZ6vF2kSTj+UiLjhD+OcMLfuhz532X86Z/HsP1oFQDAKUsYk98Fk/vfiMw2nhi3Ljmx/1IiY/+luKAGtACBagkIBCIEBoyHP+oPgpSUBrRJkVFTXQshFEiqCghF/1MhqcFtqAFI5raWLgm9vKqXN+rqx5Esdc3yapTjC0XbNwIkik8PrLSugIaVkKzBBNmSI+n5km1f22wsDY3kSRHK2vPM80lylOdbhN7C4tZvd4nynIwot8EY6Vc8ttOFDpk34OvztRCSI+T9MrYZOKH4xAl/HOGEv3Xaf/oS/vjPY/jsxAUAgMcpY+ztXfGju29EuxT+nnVzYv+lRMb+S4ksYfqvGtBWVCgN9pUS5mqJBvvqCiO4oep1FL8e4PDpAQVjhUYwzxZgMAIOwhqQ0AMUUM2AhhbkUM2AhfYntHqqAkAAqiXwYRxHKLF+R1sNERIAEMYtJfrKC2G5BcVccWG7NcVSJ2wlh54mW25PMYITUQMZzsjBC0u98EBHlICHQ3+4qOzmr5gkIE744wgn/K3bZyfO4487jmH/Ga0fpLkdGH9nDn54ZzbaePhgpObA/kuJjP2XEhn7bwxZAwWqal8hIUTICgoVMH/jQ38VIfuWNCnS74E0Uj78mNZy4XmSUCy3tARvcbnyMzKakt9Yevg5GTzRCEkO+ZUR7WGiWoDBAUhOLcAhO/XbTazp11JGSxOyQ3++hvEsDiPdsu1wWtKM47iCgRLJoQU8jOPLTgjJaf4qitDzYDwPJMHxZ/mI4sTdN92Au37YDjuOVuGP/zyGL87V4M+fHEfpv8sxsd+N+EFBV3hdjKoSERHRVTCX6TsB/Z8Rof/mZwymcZIEZLZPQ+W5CxCqdcVF8NYT++oKbeWFpK+8CK7SsN6qEh540Y5jX8khWVZ5mLekCCX44E8jIGIJVgSf36EHLKI+JDQQ4Tkhfvsx1YD9vRCqebtOa6CtsDACEdZXLWhgDx4EAwVCdkJNz0H1fS8CruR8ODcn/ETfgCRJuLdnBwzs0R5bj1TiT/88hhPn6/Da//0K/+fzckz+zk0YfWtnuByJH20kIiIiShiyrH0DLGu3W7aaoIlQQ4IJ9uCApPj0YEVAD1hor+azONRAyPY3KROpvKWMUPQHmipaW/WVIVq6P1hXDQAiAEnRXs00o16ET1ELwOi36HyDt68uvwiBrFuu/XOIQ5zwE10DWZIwtHcW7u+ViQ8OncWyT47jzKUG/Gbrl1j92UlMHdAND32rExwyHxBDRERERC1EkvWf1ww+UDrZgxtQowQK9MBEMFAQsAQkQvMVqGlZCGT2jfVVtRhO+ImagVOWMOqWzvhen45Yv/8M/rLzJE5fasBzW45gxb9O4okB3TE4LxMynwxLRERERPTN6cENOOxBjaQNcFwjrjcmakZup4xHb8/G+qK78dS9uWjrdeJYVR3mb/p/mLD63/jn0SrwOZlERERERHQ9cMJP1AK8Lgcm9LsRG6b0w5Tv3IRUlwOH/3cZT797AFPf2ovPT16IdROJiIiIiCjJccJP1ILaeJx4YmB3bJjSD4/flQOPU8be05cw/e19mLluHw6eucRv/ImIiIiIqEXwHn6i66Bdqguzv9sD4+7MxvJPT2D9/grsPH4BO4/vQarLga4ZXmRneJHdTn/NSEF2hhddMrzwOBmXIyIiIiKiq8cJP9F1lNXGg2eH9MKP7s5ByScnsOXQWdT6FXxZWYMvK2ui1HHrQQA9ENDOi65tteBAhzQ3HwRIREREREQRccJPFAPZGSn41fd6Y/6QXjhzqR7lF+tRfqEe5RfrcPpicL/Wr+DcZR/OXfZhT/mlsON4nLI5+TdejcBA1wwvUt2OGFwdERERERHFA074iWLI45TRvX0qurdPDcsTQuBiXQDlF+u0AIDl7/SFOpytbkBDQMVXVbX4qqo24vHbp7qCtwtYVghkZ3iR1cYDh8zVAUREREREyYoTfqI4JUkS2qW60C7Vhb5d2oblBxQVFdUNwUDAhXp9dYC2SuBifQBVtX5U1fpx4Ex1WH2nLKFLW4/tNoF2qS60cTuQ5nYi1e1AmkfbTnM7kOp28PYBIiIiIqIEwgk/UYJyOmTktEtBTruUiPnV9QEtAHCpHuUXgqsETut/AVXg5IV6nLxQDxxv2jmNiX+aOxgISPPor9Z0jzXfsq3nu/kgQiIiIiKiFscJP1GSSvc60dvbBr07tQnLU1SBc5cbbEGA8ov1qK4PoMYXQI1P0f4aArjsU6Co2k8HGunnrrFtLodkWznQxhY40AIG7dumIODzwyXLcDlkeJwSXA4Zboe273ZKcOv7bocMV5R9pyxB4soEIiIiImqFOOEnaoUcsoTObb3o3NaLO29svKwQAj5FaIGABgW1PgWXzaCAlmZsa3laoMAaNKjxafVq/QoAwK8IXKjz40Kdv8WvVQLgdspwOSQzWOAJ2Xc7Zbit+w5JT9P2XQ4JTlmCU9YCCA5Z33dIcEjaq1OWg+khZaLmyZbjOuzpDFIQERER0bWKuwn/+fPnsWTJEmzbtg1ff/01unfvjgkTJuD73/9+k+q/++67WLFiBY4dO4Y2bdpg6NChePrpp5GRkWErpygKVq1ahdLSUpSXlyMzMxOFhYWYMWMGvF5vS1waUUKSJAkepwSP040Izxa8KooqUOdXcNkaEIgUSGgIQHI5cOmy9mBCvyLgU1T4FRU+RcAXUG37fkU103yKMFckAIAA0BBQ0RAAAOXaLuA6kiWYwQB78EALDDgk/TV02wgohKZJEhwywspHPJ6RLkmQLflOo74lTTbKSNC2JQmybGxrrw5JgiRpda37smzJs9XTjidJUli6rB9HspQhIiIiosjiasJfW1uLoqIiHDlyBOPGjUOPHj2wZcsWLFiwAJWVlZg+fXqj9ZcuXYri4mLcc889mDNnDk6dOoVVq1Zh9+7dKC0ttU3kn3vuOZSWluLBBx/EhAkTcOjQISxduhQHDhzAsmXL+I9IohbgkCW08TjRxtP40CNJQGZmOiorqyFEo0UjUlQ9CBAlIOA3AwYCDYpq2Y8eUAiowvxTVIGAIhBQVW07NE8VCCgqFGGUC8mz1lMEIl2iKqC1RUmcIEUsSNCCI3JIMMEILsgSIgYKjEBF9KCEFkwIC2Y04TiyDKSmuOFr8EOCdlxJrytBfzWCHkYetPpaWX3blibZygb3G6kvATIk2761vu38elnoaaH1rec33vPwtOC5jB8Aka3t1g4fPDfs7bWWR8i1W89JRERETRdXE/7Vq1fj4MGDKC4uxogRIwAAY8eOxdSpU7FkyRI8/PDD6NKlS8S6FRUVeP3113Hfffdh6dKlkGXtoWB9+/bFM888g1WrVmHq1KkAgH379qG0tBRjx47F888/bx4jJycHxcXF2Lx5M4YPH97CV0tELUX7BtoBr8sR66Y0iRItGKAHBKx5RuBAEcKsp6gI2dfyA2p4mnEs1VI+oCLCMQUCln3VUtfaXiEEFAGoQkAVgKofRwitTaqqBTSMY6h6WUXVyqjCUj6kjNqEYI8AoAhAUYS+R8nOmPzDEtCBniaFBjgQOVihlY8cUNAPbQnW2AMSZr7tePYAjLWNUkjbrnQ8a7uNfVt7oV+7pZ1GoMY4R+ixrfW11gbbC/M9C543LdWNujqfeeyw96SR9sGaZ2zbjhH+vkQ8h7FtOUfoe2Q9DhB6ntDjRj8XEH6+aOeyHUvvkNHOabzXkq2c/ZoQ0hbJTJMs24204UrXc7XHZFCNKOnE1YR//fr16NSpkznZB7SBZ8qUKdi+fTvee+89TJs2LWLd9957D36/H5MmTTIn+wAwcuRIFBcXo6yszJzwl5WVAQAmT55sO8bEiRPxxhtvoKysjBN+IrpujCXybm0vxq2JH0JowQItiAA9uCCgqkZAwJKu2gMFYYEHS3qkIEPwONoxzEBGhACGCutxIp1fwJviRk1NA1Q9X8sDgGA5YXkVCE9ThXYuiNA6gEB4WeMYqh77MOobQRcRUt9oW2h9IRAlzdIOIOz8AMKvDcF65uu19gtoQR4Iod+kw0APUXNrNNihBwWiBRq0PClk33JsS31rmcbyrlTfGtSQHRJURYSlh7bZSIh0HcGyUnhbQoI2xpGtASczPULgyDy2tS2RAkWW40QOFIUcOyzAE61ceDts5REsaH+P7Z+pPd/+/lo1mhd2nZacCOmR3sPQ9iFivfDP1mibBKBjugfD+mQl7c9Px82Ev7q6GkePHsXQoUPD8m677TYA2jfz0ezdu9dW1urWW2/Fli1bUF1djfT0dOzduxft2rVD9+7dbeW8Xi/y8vIaPQ8REV0f5jelDil+/mfVBJJ0bbektAZmMEefuKshQQMjoGEPigQDBiJKoMGsHyGQYtSHAFRzO9iO0DpaO+2Bi2Cb7YEPEaGM9Xhh12O9BoSUtQSGjHLW6zLaFdpO6/WZASEEr9EI9gCRr9Pafq/Xhbo6n35OS/tD24rg52dtq+1coccPORdsn1Gw/WZwSNiPEbovopwn0nWFfqbW9yqYf6XzBf+jbuycsKZZPiOjuvnayPmNumHvwXVgbUP4QHa9WkF0/dx4Qwr6dk6PdTNaRNz8G+rs2bMQQkRcsp+SkoKMjAycOnUqav2Kigq0bdsWbdqE/wRZ586dAQDl5eXo06cPKioqot4a0LlzZ+zfv98MDhAREVHzCl0OT/GDAavE0NTARJODDUaZazhmpEAHzHPZz2lNM85tb5u1rv289rr2PEkC2mak4sKFGjPIZuZbzh1ss/09MNsYck2hwSFhyQgNgNnfo5AgkQhvS2Ofh70ObJ+J/T0JaYflvJHaYU23lQ3Jt7bLlh/lM4v0uYSfy/7e2a8jeNxobYyUHhaSivD+RWq3kd6xjRt5WWlIVnEz4a+urgYApKZGfgy41+tFXV1do/UbqwtoDwU0yubm5l6xbGMT/nhf8RG6tIookbD/UiJj/6VExv6bGMLvtecHBmj9tkOHdHzdxsmAFSWclhp/42bCH4xCRf6vUwhhuze/sWNE43AE741t7DyhZUO1b58Gh6PxtsSLDh24SoESF/svJTL2X0pk7L+UyNh/KZE1d/+Nmwl/Wpq2jKK+vj5ifn19fdRl+Eb98+fPR8wzVgYYy/3T0tIaPQ+ARr/dr6qqifvItxnh/JpL8ijxsP9SImP/pUTG/kuJjP2XEllT+m9m5tUHA+Jmwp+TkwNJklBRURGWV1tbi0uXLpn34kerf/DgQdTW1oYt7a+oqIAsy+jUqRMAIDs7G2fOnIl4nIqKCtxwww3weDyNtjdRBhHr/UtEiYb9lxIZ+y8lMvZfSmTsv5TImrv/xs269LS0NPTs2RP79+8PyzOewH/HHXdErZ+fnw8g8pP89+/fj169epnf8N92222oqqrCyZMnbeXq6upw5MgR3H777d/4OoiIiIiIiIjiQdxM+AFg1KhRKC8vx/vvv2+mCSGwfPlyuN1uDB8+PGrdhx56CC6XC8uWLbPdn79p0yacPn0ajzzyiJlWWFgIACgpKbEdY+XKlfD5fLayRERERERERIkobpb0A8DEiROxceNGzJs3DwcOHEBubi42b96Mjz/+GHPnzkXHjh0BACdPnsTu3btx0003md/GZ2dnY/r06Xj99dcxefJkPPTQQ/jqq6+watUq3HrrrXjsscfM89xxxx145JFHUFpaiosXL2LQoEHYv38/3n77bdx///0YMmRITK6fiIiIiIiIqLlI4kqPtr/OqqqqUFxcjG3btqGmpga5ubmYNGkSRo8ebZYpKyvD/PnzMWbMGLz88su2+mvWrMHq1atx/PhxZGZm4oEHHsCsWbOQkZFhKxcIBFBSUoJ33nkHFRUV6Ny5MwoLC/HEE0+YP80Xzblz1c12vS2Fv6NLiYz9lxIZ+y8lMvZfSmTsv5TImtJ/s7Ku/qF9cTfhTwSc8BO1LPZfSmTsv5TI2H8pkbH/UiJrqQl/XN3DT0RERERERETNgxN+IiIiIiIioiTECT8RERERERFREuKEn4iIiIiIiCgJccJPRERERERElIQ44SciIiIiIiJKQpzwExERERERESUhTviJiIiIiIiIkhAn/ERERERERERJiBN+IiIiIiIioiTECT8RERERERFREuKEn4iIiIiIiCgJccJPRERERERElIQ44SciIiIiIiJKQpzwExERERERESUhTviJiIiIiIiIkhAn/ERERERERERJiBN+IiIiIiIioiQkCSFErBtBRERERERERM2L3/ATERERERERJSFO+ImIiIiIiIiSECf8REREREREREmIE34iIiIiIiKiJMQJf5I5f/48Fi1ahPvvvx/5+fkYNWoU1q1bF+tmETXJggUL0Lt374h/ZWVlsW4eUZi9e/fiW9/6Fnbu3BmWd/r0acydOxeDBg1CQUEBHn30UWzdujUGrSSKrLH+O2nSpKjjcaTyRNfL4cOHMWvWLHznO9/BLbfcgsGDB+PFF19EdXW1rRzHYIpHTe2/zTkGO5vzAii2amtrUVRUhCNHjmDcuHHo0aMHtmzZggULFqCyshLTp0+PdROJGnX48GFkZ2dj9uzZYXl33HFHDFpEFN2xY8fw5JNPQlXVsLxz587h8ccfx4ULF/CjH/0InTp1wrp16/CTn/wEixcvRmFhYQxaTBTUWP8FtPH4lltuwYQJE8Lyevbs2dLNI4ro6NGjeOyxx+BwODB+/Hh06dIFe/bswerVq/Hpp5+itLQUqampHIMpLjW1/wLNPAYLShpLly4VeXl5YtOmTWaaqqqiqKhI9O3bV5w+fTqGrSNqnKIoIj8/X8yePTvWTSG6or/97W/i7rvvFnl5eSIvL098+umntvyFCxeK3r17i88//9xMq6+vF6NGjRL9+/cXNTU117vJRKYr9d+zZ8+KvLw88Zvf/CZGLSSKbPLkyaJv377i8OHDtvQVK1aIvLw8UVJSIoTgGEzxqan9t7nHYC7pTyLr169Hp06dMGLECDNNkiRMmTIFfr8f7733XgxbR9S4Y8eOob6+Hr169Yp1U4gaNW3aNMycORNZWVkYOXJkWL6iKNi4cSMKCgpsK1M8Hg8mTJiA8+fP4x//+Md1bDFR0JX6L6B9swSA4zHFFZ/Ph127duHOO+9EXl6eLW/06NEAgM8++4xjMMWlpvZfoPnHYE74k0R1dTWOHj2K2267LSzPSNu3b9/1bhZRk/3nP/8BAHMQrKurg6IosWwSUURHjx7Fz372M7z77rvo3r17WP4XX3yB2tpaFBQUhOUZ4/HevXtbuJVEkV2p/wLh43FtbW3Upf9E14vT6cSmTZuwaNGisLzKykoAgCzLHIMpLjW1/wLNPwbzHv4kcfbsWQgh0KVLl7C8lJQUZGRk4NSpUzFoGVHTGNHM7du349e//jXKy8vhcrlw3333Yf78+bjxxhtj3EIizQcffAC32x01/+zZswAQcTzu3LkzAHA8ppi5Uv8Fgv/YfPfddzF16lRUVlYiJSUFw4YNw7PPPov27dtfj6YS2ciyHPXfAn/5y18AAP379+cYTHGpqf0XaP4xmBP+JGE82dF40EMor9eLurq669kkoqtiTPj37NmDGTNm4IYbbsDu3buxcuVK/Pvf/8batWuRk5MT41YS4YqTpcbGY6/XCwAcjylmrtR/AeDIkSMAgEOHDmHu3LnweDz4+OOP8fbbb2Pv3r1Yu3Yt2rZt29JNJWqS9evXY+3atejSpQt+8IMf4KOPPgLAMZgSQ2j/BZp/DOaEP0kIIWyvkfKNZSJE8Wj48OH49re/jWnTppn/Qx4yZAgKCgrw1FNP4Xe/+x1effXVGLeS6MqijcPWPI7HFM8ee+wx1NTUYMqUKWZf/d73vofc3Fy8/PLLWL58OX7605/GuJVE2jegCxYsQGpqKl577TWkpaVxDKaEEan/As0/BrO3Jwmjg9TX10fMr6+vR3p6+vVsEtFVGTVqFGbNmmVO9g3Dhg1Dly5dsGPHjhi1jOjqGONxpG+QjDGa4zHFs/Hjx2PatGlhk6Lx48fD4XBg+/btMWoZUdAbb7yBZ599FqmpqSgpKUF+fj4AjsGUGKL1X6D5x2B+w58kcnJyIEkSKioqwvJqa2tx6dIl874lokTToUMHc8k/Ubwzbj2JNB4baRyPKRG53W60bdsWNTU1sW4KtWJ+vx8LFy5EWVkZOnXqhD//+c/o06ePmc8xmOLZlfpvY77pGMxv+JNEWloaevbsif3794flGU8itf40CVE8qaqqQmFhIWbOnBmW5/f7cfz4cXTr1i0GLSO6ej169EB6enrEX0bheEzx7vDhwxgxYgReeOGFsLyvv/4a58+f53hMMaMoCp555hmUlZWhd+/eWLt2bdhkiWMwxaum9N+WGIM54U8io0aNQnl5Od5//30zTQiB5cuXw+12Y/jw4TFsHVF07du3h6Io+Oijj3DgwAFb3tKlS1FdXY0xY8bEqHVEV8fpdGL48OHYtWsXdu/ebaY3NDRg5cqVyMzMxH333RfDFhJF161bN5w7dw7r16/H6dOnbXnGc1Q4HlOs/P73v8eHH36I/Px8vPnmm+jUqVNYGY7BFK+a0n9bYgzmkv4kMnHiRGzcuBHz5s3DgQMHkJubi82bN+Pjjz/G3Llz0bFjx1g3kSiqX/7yl5gyZQomTZqE8ePHo2PHjti5cyc+/PBD9OvXDxMnTox1E4ma7KmnnsK2bdswdepU/PjHP0aHDh2wbt06HDlyBMXFxfB4PLFuIlFEXq8Xv/jFLzBnzhw8+uijGDduHNLT07F161Z88sknKCwsxEMPPRTrZlIrdPr0aSxfvhySJGHo0KHYtm1bWJnMzEwMHDiQYzDFnavpv809BnPCn0S8Xi9WrVqF4uJibNiwATU1NcjNzcUrr7yC0aNHx7p5RI3q378/3nrrLSxZsgRr1qxBbW0tcnJyMHv2bBQVFcHlcsW6iURNlpWVhTVr1uDVV1/FypUr4ff70bt3byxduhTf/e53Y908okYVFhYiMzMTS5cuRUlJCRRFQW5uLhYuXIgf/vCHsW4etVL/+te/EAgEACDqr/b069cPAwcO5BhMcedq+m9zj8GSaOy3K4iIiIiIiIgoIfEefiIiIiIiIqIkxAk/ERERERERURLihJ+IiIiIiIgoCXHCT0RERERERJSEOOEnIiIiIiIiSkKc8BMRERERERElIU74iYiIiIiIiJIQJ/xERERERERESYgTfiIiolbu9ddfR+/eva/49+KLL8a6qWZbd+7cGeumEBERxT1nrBtARERE8WHs2LG48847o+b37NnzOraGiIiIrhUn/ERERAQAKCgowMMPPxzrZhAREVEz4ZJ+IiIiIiIioiTECT8RERFdlcGDB2PSpEnYsWMHxowZg/z8fNx///1YvHgx6uvrw8pv3LgRY8eORUFBAQoKCjB27Fhs2LAhrJwQAmvWrMEjjzyCgoICDBgwANOnT8fBgwfDyp4/fx4LFy7EgAEDkJ+fj9GjR2PTpk0tcr1ERESJihN+IiIiAgDU1taiqqoq6p8Qwiz73//+F9OnT8fNN9+MefPmIT8/HyUlJSgqKoKqqma5RYsWYc6cOfD7/Zg5cyZmzpwJn8+HuXPn4oUXXrCdf968efjVr36FlJQUPP3005g0aRIOHjyIxx9/HIcOHbKVnT9/Po4cOYInn3wSM2fORGVlJZ555hls3769Zd8kIiKiBMJ7+ImIiAiANjlftGhR1PzPPvsMbdu2BQD873//w9NPP40ZM2YAAMaPH4+XXnoJK1aswIYNGzBmzBjs2rULq1evxj333IOSkhK4XC4AwMSJE1FUVIRVq1Zh2LBh6NevHz799FNs2LABI0eOxOLFiyFJEgDggQcewMiRI/GnP/0Jr732mtmW22+/HcuWLYMsa99d5OfnY+LEifjggw9w7733tsj7Q0RElGg44SciIiIAQFFREQYNGhQ1PzU11dxOT09HUVGRLX/69OlYsWIFPvzwQ4wZMwabN28GAMycOdOc7AOAy+XCrFmzMH78eHzwwQfo168f/v73vwMApkyZYk72Ae2XAdatW4eOHTvazjV69Ghzsg9oDxwEtEAEERERaTjhJyIiIgDAzTffjAEDBjSpbLdu3eB2u21p7du3R0ZGBk6ePAkAOHHiBACgV69eYfXz8vIAAKdOnbK9Rvrpv759+4alZWVl2fa9Xi8AwOfzNan9RERErQHv4SciIqKrFjrZNyiKAofDAQC2e/4jlbMex+/3X9X5rd/uExERUWT8vyURERFdtRMnToRN6M+ePYvLly+je/fuAICbbroJAPDFF1+E1f/yyy8BAF27dgUA5OTkAAC++uqrsLLFxcV46aWXmq3tRERErQUn/ERERHTVKisrw35a7w9/+AMAYOTIkQCABx98EACwZMkSBAIBs1wgEMCSJUtsZYYMGQIAWLFihe2YJ06cwF//+lfzNgEiIiJqOt7DT0RERACAPXv2mMvxI0lLSzMn5i6XCz//+c+xb98+3HzzzdixYwe2bt2KoUOHYtiwYQCA/v37Y+zYsSgtLcWjjz6KESNGAADef/99HDx4EOPGjcPdd98NALj33nsxcuRIvPPOO6ioqMDgwYNx+fJlvPnmm/B4PJgzZ04LXz0REVHy4YSfiIiIAAClpaUoLS2Nmp+dnW1O+Dt27IjnnnsOr7zyCtauXYvs7GzMnTsXkyZNstV5/vnnkZ+fj7feeguvvfYaHA4H+vTpg8WLF6OwsNBW9re//S3y8/Oxbt06vPLKK8jIyMBdd92F2bNnIzc3t9mvl4iIKNlJorEn6hARERGFGDx4MABg27ZtMW4JERERNYb38BMRERERERElIU74iYiIiIiIiJIQJ/xERERERERESYj38BMRERERERElIX7DT0RERERERJSEOOEnIiIiIiIiSkKc8BMRERERERElIU74iYiIiIiIiJIQJ/xERERERERESYgTfiIiIiIiIqIkxAk/ERERERERURLihJ+IiIiIiIgoCXHCT0RERERERJSE/j8/jYvo+n3d3gAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<Figure size 1200x800 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig, ax = plt.subplots(figsize = (12, 8))\n",
    "ax.plot(hist['train_loss'], label = 'Train')\n",
    "ax.plot(hist['val_loss'], label = 'Test')\n",
    "ax.set_xlabel('Epoch')\n",
    "ax.set_ylabel('Cross Entropy Loss')\n",
    "ax.legend(frameon = True)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "a8ce2040",
   "metadata": {},
   "source": [
    "### Inference"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "4a9244d8",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\tomas\\miniconda3\\envs\\torch\\lib\\site-packages\\torchvision\\models\\_utils.py:208: UserWarning: The parameter 'pretrained' is deprecated since 0.13 and may be removed in the future, please use 'weights' instead.\n",
      "  warnings.warn(\n",
      "C:\\Users\\tomas\\miniconda3\\envs\\torch\\lib\\site-packages\\torchvision\\models\\_utils.py:223: UserWarning: Arguments other than a weight enum or `None` for 'weights' are deprecated since 0.13 and may be removed in the future. The current behavior is equivalent to passing `weights=None`.\n",
      "  warnings.warn(msg)\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "DeepLabV3(\n",
       "  (backbone): IntermediateLayerGetter(\n",
       "    (conv1): Conv2d(3, 64, kernel_size=(7, 7), stride=(2, 2), padding=(3, 3), bias=False)\n",
       "    (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    (relu): ReLU(inplace=True)\n",
       "    (maxpool): MaxPool2d(kernel_size=3, stride=2, padding=1, dilation=1, ceil_mode=False)\n",
       "    (layer1): Sequential(\n",
       "      (0): Bottleneck(\n",
       "        (conv1): Conv2d(64, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "        (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (relu): ReLU(inplace=True)\n",
       "        (downsample): Sequential(\n",
       "          (0): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "          (1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        )\n",
       "      )\n",
       "      (1): Bottleneck(\n",
       "        (conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "        (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (relu): ReLU(inplace=True)\n",
       "      )\n",
       "      (2): Bottleneck(\n",
       "        (conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "        (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (relu): ReLU(inplace=True)\n",
       "      )\n",
       "    )\n",
       "    (layer2): Sequential(\n",
       "      (0): Bottleneck(\n",
       "        (conv1): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)\n",
       "        (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (relu): ReLU(inplace=True)\n",
       "        (downsample): Sequential(\n",
       "          (0): Conv2d(256, 512, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
       "          (1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        )\n",
       "      )\n",
       "      (1): Bottleneck(\n",
       "        (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "        (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (relu): ReLU(inplace=True)\n",
       "      )\n",
       "      (2): Bottleneck(\n",
       "        (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "        (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (relu): ReLU(inplace=True)\n",
       "      )\n",
       "      (3): Bottleneck(\n",
       "        (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "        (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (relu): ReLU(inplace=True)\n",
       "      )\n",
       "    )\n",
       "    (layer3): Sequential(\n",
       "      (0): Bottleneck(\n",
       "        (conv1): Conv2d(512, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "        (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (relu): ReLU(inplace=True)\n",
       "        (downsample): Sequential(\n",
       "          (0): Conv2d(512, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "          (1): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        )\n",
       "      )\n",
       "      (1): Bottleneck(\n",
       "        (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(2, 2), dilation=(2, 2), bias=False)\n",
       "        (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (relu): ReLU(inplace=True)\n",
       "      )\n",
       "      (2): Bottleneck(\n",
       "        (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(2, 2), dilation=(2, 2), bias=False)\n",
       "        (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (relu): ReLU(inplace=True)\n",
       "      )\n",
       "      (3): Bottleneck(\n",
       "        (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(2, 2), dilation=(2, 2), bias=False)\n",
       "        (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (relu): ReLU(inplace=True)\n",
       "      )\n",
       "      (4): Bottleneck(\n",
       "        (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(2, 2), dilation=(2, 2), bias=False)\n",
       "        (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (relu): ReLU(inplace=True)\n",
       "      )\n",
       "      (5): Bottleneck(\n",
       "        (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(2, 2), dilation=(2, 2), bias=False)\n",
       "        (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (relu): ReLU(inplace=True)\n",
       "      )\n",
       "      (6): Bottleneck(\n",
       "        (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(2, 2), dilation=(2, 2), bias=False)\n",
       "        (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (relu): ReLU(inplace=True)\n",
       "      )\n",
       "      (7): Bottleneck(\n",
       "        (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(2, 2), dilation=(2, 2), bias=False)\n",
       "        (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (relu): ReLU(inplace=True)\n",
       "      )\n",
       "      (8): Bottleneck(\n",
       "        (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(2, 2), dilation=(2, 2), bias=False)\n",
       "        (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (relu): ReLU(inplace=True)\n",
       "      )\n",
       "      (9): Bottleneck(\n",
       "        (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(2, 2), dilation=(2, 2), bias=False)\n",
       "        (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (relu): ReLU(inplace=True)\n",
       "      )\n",
       "      (10): Bottleneck(\n",
       "        (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(2, 2), dilation=(2, 2), bias=False)\n",
       "        (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (relu): ReLU(inplace=True)\n",
       "      )\n",
       "      (11): Bottleneck(\n",
       "        (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(2, 2), dilation=(2, 2), bias=False)\n",
       "        (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (relu): ReLU(inplace=True)\n",
       "      )\n",
       "      (12): Bottleneck(\n",
       "        (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(2, 2), dilation=(2, 2), bias=False)\n",
       "        (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (relu): ReLU(inplace=True)\n",
       "      )\n",
       "      (13): Bottleneck(\n",
       "        (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(2, 2), dilation=(2, 2), bias=False)\n",
       "        (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (relu): ReLU(inplace=True)\n",
       "      )\n",
       "      (14): Bottleneck(\n",
       "        (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(2, 2), dilation=(2, 2), bias=False)\n",
       "        (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (relu): ReLU(inplace=True)\n",
       "      )\n",
       "      (15): Bottleneck(\n",
       "        (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(2, 2), dilation=(2, 2), bias=False)\n",
       "        (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (relu): ReLU(inplace=True)\n",
       "      )\n",
       "      (16): Bottleneck(\n",
       "        (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(2, 2), dilation=(2, 2), bias=False)\n",
       "        (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (relu): ReLU(inplace=True)\n",
       "      )\n",
       "      (17): Bottleneck(\n",
       "        (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(2, 2), dilation=(2, 2), bias=False)\n",
       "        (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (relu): ReLU(inplace=True)\n",
       "      )\n",
       "      (18): Bottleneck(\n",
       "        (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(2, 2), dilation=(2, 2), bias=False)\n",
       "        (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (relu): ReLU(inplace=True)\n",
       "      )\n",
       "      (19): Bottleneck(\n",
       "        (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(2, 2), dilation=(2, 2), bias=False)\n",
       "        (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (relu): ReLU(inplace=True)\n",
       "      )\n",
       "      (20): Bottleneck(\n",
       "        (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(2, 2), dilation=(2, 2), bias=False)\n",
       "        (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (relu): ReLU(inplace=True)\n",
       "      )\n",
       "      (21): Bottleneck(\n",
       "        (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(2, 2), dilation=(2, 2), bias=False)\n",
       "        (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (relu): ReLU(inplace=True)\n",
       "      )\n",
       "      (22): Bottleneck(\n",
       "        (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(2, 2), dilation=(2, 2), bias=False)\n",
       "        (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (relu): ReLU(inplace=True)\n",
       "      )\n",
       "    )\n",
       "    (layer4): Sequential(\n",
       "      (0): Bottleneck(\n",
       "        (conv1): Conv2d(1024, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(2, 2), dilation=(2, 2), bias=False)\n",
       "        (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (relu): ReLU(inplace=True)\n",
       "        (downsample): Sequential(\n",
       "          (0): Conv2d(1024, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "          (1): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        )\n",
       "      )\n",
       "      (1): Bottleneck(\n",
       "        (conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(4, 4), dilation=(4, 4), bias=False)\n",
       "        (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (relu): ReLU(inplace=True)\n",
       "      )\n",
       "      (2): Bottleneck(\n",
       "        (conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(4, 4), dilation=(4, 4), bias=False)\n",
       "        (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (relu): ReLU(inplace=True)\n",
       "      )\n",
       "    )\n",
       "  )\n",
       "  (classifier): DeepLabHead(\n",
       "    (0): ASPP(\n",
       "      (convs): ModuleList(\n",
       "        (0): Sequential(\n",
       "          (0): Conv2d(2048, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "          (1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "          (2): ReLU()\n",
       "        )\n",
       "        (1): ASPPConv(\n",
       "          (0): Conv2d(2048, 256, kernel_size=(3, 3), stride=(1, 1), padding=(12, 12), dilation=(12, 12), bias=False)\n",
       "          (1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "          (2): ReLU()\n",
       "        )\n",
       "        (2): ASPPConv(\n",
       "          (0): Conv2d(2048, 256, kernel_size=(3, 3), stride=(1, 1), padding=(24, 24), dilation=(24, 24), bias=False)\n",
       "          (1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "          (2): ReLU()\n",
       "        )\n",
       "        (3): ASPPConv(\n",
       "          (0): Conv2d(2048, 256, kernel_size=(3, 3), stride=(1, 1), padding=(36, 36), dilation=(36, 36), bias=False)\n",
       "          (1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "          (2): ReLU()\n",
       "        )\n",
       "        (4): ASPPPooling(\n",
       "          (0): AdaptiveAvgPool2d(output_size=1)\n",
       "          (1): Conv2d(2048, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "          (2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "          (3): ReLU()\n",
       "        )\n",
       "      )\n",
       "      (project): Sequential(\n",
       "        (0): Conv2d(1280, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        (2): ReLU()\n",
       "        (3): Dropout(p=0.5, inplace=False)\n",
       "      )\n",
       "    )\n",
       "    (1): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "    (2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    (3): ReLU()\n",
       "    (4): Conv2d(256, 7, kernel_size=(1, 1), stride=(1, 1))\n",
       "  )\n",
       ")"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "device = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")\n",
    "model = initialize_model(num_classes, use_pretrained = False)\n",
    "state_dict = torch.load(os.path.join(dest_dir, 'best_DeepLabV3_floodDetection.pth'), \n",
    "                        map_location = device)\n",
    "\n",
    "model = model.to(device)\n",
    "model.load_state_dict(state_dict, strict=False)\n",
    "model.eval()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "id": "c329f466",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "../../data/training_def/test/Images\\resized_CAM_BF_01_20220515235422_rtm.jpg\n",
      "../../data/training_def/test/Labels\\resized_CAM_BF_01_20220515235422_rtm_label.jpg\n",
      "\n",
      "../../data/training_def/test/Images\\resized_CAM_BF_01_20220517004222_rtm.jpg\n",
      "../../data/training_def/test/Labels\\resized_CAM_BF_01_20220517004222_rtm_label.jpg\n",
      "\n",
      "../../data/training_def/test/Images\\resized_CAM_BF_01_20221120192422_rtm.jpg\n",
      "../../data/training_def/test/Labels\\resized_CAM_BF_01_20221120192422_rtm_label.jpg\n",
      "\n",
      "../../data/training_def/test/Images\\resized_CAM_BF_01_20221109125422_rtm.jpg\n",
      "../../data/training_def/test/Labels\\resized_CAM_BF_01_20221109125422_rtm_label.jpg\n",
      "\n",
      "../../data/training_def/test/Images\\resized_CAM_BF_01_20221030232422_rtm.jpg\n",
      "../../data/training_def/test/Labels\\resized_CAM_BF_01_20221030232422_rtm_label.jpg\n",
      "\n",
      "../../data/training_def/test/Images\\resized_CAM_BF_01_20220519024221_rtm.jpg\n",
      "../../data/training_def/test/Labels\\resized_CAM_BF_01_20220519024221_rtm_label.jpg\n",
      "\n",
      "../../data/training_def/test/Images\\resized_CAM_BF_01_20220930152422_rtm.jpg\n",
      "../../data/training_def/test/Labels\\resized_CAM_BF_01_20220930152422_rtm_label.jpg\n",
      "\n",
      "../../data/training_def/test/Images\\resized_CAM_BF_01_20220809200622_rtm.jpg\n",
      "../../data/training_def/test/Labels\\resized_CAM_BF_01_20220809200622_rtm_label.jpg\n",
      "\n",
      "../../data/training_def/test/Images\\resized_CAM_BF_01_20220515221822_rtm.jpg\n",
      "../../data/training_def/test/Labels\\resized_CAM_BF_01_20220515221822_rtm_label.jpg\n",
      "\n",
      "../../data/training_def/test/Images\\resized_CAM_BF_01_20220515190022_rtm.jpg\n",
      "../../data/training_def/test/Labels\\resized_CAM_BF_01_20220515190022_rtm_label.jpg\n",
      "\n"
     ]
    }
   ],
   "source": [
    "pathout = r'../../models/deepLabV3_r2/outputs'\n",
    "\n",
    "path_test_images = r'../../data/training_def/test/Images'\n",
    "path_test_labels = r'../../data/training_def/test/Labels'\n",
    "images = [os.path.join(path_test_images, x) for x in os.listdir(path_test_images)]\n",
    "labels = [os.path.join(path_test_labels, x[:-4]+'_label.jpg') for x in os.listdir(path_test_images)]\n",
    "\n",
    "transforms_image =  transforms.Compose([transforms.ToTensor()])\n",
    "\n",
    "for im, lab in zip(images, labels):\n",
    "    print(im)\n",
    "    print(lab+'\\n')\n",
    "    ## prediction\n",
    "    image = Image.open(im)\n",
    "    image_np = np.asarray(image)\n",
    "    image = transforms_image(image)\n",
    "    image = image.unsqueeze(0)\n",
    "    image = image.to(device)\n",
    "    outputs = model(image)[\"out\"]\n",
    "    _, preds = torch.max(outputs, 1)\n",
    "    preds = preds.to(\"cpu\")\n",
    "    preds_np = preds.squeeze(0).cpu().numpy().astype(np.uint8)\n",
    "    preds_np_color = cv2.applyColorMap(preds_np * 50, cv2.COLORMAP_HSV)\n",
    "    image_np = cv2.cvtColor(image_np, cv2.COLOR_RGB2BGR)\n",
    "    \n",
    "    ## label\n",
    "    true_label = cv2.imread(lab)\n",
    "    true_label = np.clip(true_label, 0, num_classes)\n",
    "    true_label_color = cv2.applyColorMap(true_label * 50, cv2.COLORMAP_HSV)\n",
    "    \n",
    "#     label_aux = transforms_image(Image.fromarray(true_label)).squeeze(0)\n",
    "    iou_mean = jaccard_score(true_label.max(2).reshape(-1), preds_np.reshape(-1), average = 'macro')\n",
    "#     iou_mean = iou(preds, label_aux.max(0)[0], num_classes).mean()\n",
    "    \n",
    "    ## plot\n",
    "    fig, ax = plt.subplots(figsize = (20, 20), sharey = True, ncols = 2)\n",
    "    ax[0].imshow(image_np)\n",
    "    ax[0].imshow(true_label_color, alpha = 0.3)\n",
    "    ax[1].imshow(image_np)\n",
    "    ax[1].imshow(preds_np_color, alpha = 0.3)\n",
    "\n",
    "    ax[0].set_title('True labels')\n",
    "    ax[1].set_title(f'Predicted labels: Mean IoU = {iou_mean:0.2f}')\n",
    "    \n",
    "    fig.savefig(os.path.join(pathout, os.path.basename(im)[:-4]+'_prediction.jpg'), dpi = 200, bbox_inches = 'tight')\n",
    "    plt.close()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "id": "986de536",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(749, 1000)"
      ]
     },
     "execution_count": 60,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "preds_np.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 61,
   "id": "f75a12de",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[[3, 3, 3],\n",
       "        [3, 3, 3],\n",
       "        [3, 3, 3],\n",
       "        ...,\n",
       "        [1, 1, 1],\n",
       "        [1, 1, 1],\n",
       "        [2, 2, 2]],\n",
       "\n",
       "       [[3, 3, 3],\n",
       "        [3, 3, 3],\n",
       "        [3, 3, 3],\n",
       "        ...,\n",
       "        [1, 1, 1],\n",
       "        [1, 1, 1],\n",
       "        [1, 1, 1]],\n",
       "\n",
       "       [[3, 3, 3],\n",
       "        [3, 3, 3],\n",
       "        [3, 3, 3],\n",
       "        ...,\n",
       "        [1, 1, 1],\n",
       "        [1, 1, 1],\n",
       "        [1, 1, 1]],\n",
       "\n",
       "       ...,\n",
       "\n",
       "       [[1, 1, 1],\n",
       "        [1, 1, 1],\n",
       "        [1, 1, 1],\n",
       "        ...,\n",
       "        [3, 3, 3],\n",
       "        [3, 3, 3],\n",
       "        [3, 3, 3]],\n",
       "\n",
       "       [[1, 1, 1],\n",
       "        [1, 1, 1],\n",
       "        [1, 1, 1],\n",
       "        ...,\n",
       "        [3, 3, 3],\n",
       "        [3, 3, 3],\n",
       "        [3, 3, 3]],\n",
       "\n",
       "       [[1, 1, 1],\n",
       "        [1, 1, 1],\n",
       "        [1, 1, 1],\n",
       "        ...,\n",
       "        [3, 3, 3],\n",
       "        [3, 3, 3],\n",
       "        [3, 3, 3]]], dtype=uint8)"
      ]
     },
     "execution_count": 61,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "true_label"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "id": "2f2395d3",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(749, 1000)"
      ]
     },
     "execution_count": 50,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "true_label.max(2).shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "id": "da02a600",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.007990375611333977"
      ]
     },
     "execution_count": 54,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "jaccard_score(true_label.max(2).reshape(-1), preds_np.reshape(-1), average = 'macro')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "bc23441a",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.15"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}