24.3 ImageDataGenerator in Keras.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Keras图像增强技术详解"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "在深度学习中，数据短缺是我们经常面临的一个问题，虽然现在有不少公开数据集，但跟大公司掌握的海量数据集相比，数量上仍然偏少，而某些特定领域的数据采集更是非常困难。根据之前的学习可知，数据量少带来的最直接影响就是过拟合。那有没有办法在现有少量数据基础上，降低或解决过拟合问题呢？\n",
    "\n",
    "答案是有的，就是数据增强技术。我们可以对现有的数据，如图片数据进行平移、翻转、旋转、缩放、亮度增强等操作，以生成新的图片来参与训练或测试。这种操作可以将图片数量提升数倍，由此大大降低了过拟合的可能。本文将详解图像增强技术在Keras中的原理和应用。"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Keras中的ImageDataGenerator类"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "图像增强的官网地址是：https://keras.io/preprocessing/image/ ，API使用相对简单，功能也很强大。\n",
    "\n",
    "先介绍的是ImageDataGenerator类，这个类定义了图片该如何进行增强操作，其API及参数定义如下："
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "```\n",
    "keras.preprocessing.image.ImageDataGenerator(\n",
    "　　　　featurewise_center=False, #输入值按照均值为0进行处理\n",
    "　　　　samplewise_center=False, #每个样本的均值按0处理\n",
    "　　　　featurewise_std_normalization=False, #输入值按照标准正态化处理\n",
    "　　　　samplewise_std_normalization=False, #每个样本按照标准正态化处理 \n",
    "　　　　zca_whitening=False, # 是否开启增白\n",
    "　　　　zca_epsilon=1e-06, \n",
    "　　　　rotation_range=0, #图像随机旋转一定角度，最大旋转角度为设定值\n",
    "　　　　width_shift_range=0.0, #图像随机水平平移，最大平移值为设定值。若值为小于1的float值，则可认为是按比例平移，若大于1，则平移的是像素；若值为整型，平移的也是像素；假设像素为2.0，则移动范围为[-1,1]之间\n",
    "　　　　height_shift_range=0.0, #图像随机垂直平移，同上\n",
    "　　　　brightness_range=None, # 图像随机亮度增强，给定一个含两个float值的list，亮度值取自上下限值间\n",
    "　　　　shear_range=0.0, # 图像随机修剪\n",
    "　　　　zoom_range=0.0, # 图像随机变焦 \n",
    "　　　　channel_shift_range=0.0, \n",
    "　　　　fill_mode='nearest', #填充模式，默认为最近原则，比如一张图片向右平移，那么最左侧部分会被临近的图案覆盖\n",
    "　　　　cval=0.0, \n",
    "　　　　horizontal_flip=False, #图像随机水平翻转\n",
    "　　　　vertical_flip=False, #图像随机垂直翻转\n",
    "　　　　rescale=None, #缩放尺寸\n",
    "　　　　preprocessing_function=None, \n",
    "　　　　data_format=None, \n",
    "　　　　validation_split=0.0, \n",
    "　　　　dtype=None)\n",
    "```"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "下文将以mnist和花类的数据集进行图片操作，其中花类（17种花，共1360张图片）数据集可见百度网盘： https://pan.baidu.com/s/1YDA_VOBlJSQEijcCoGC60w 。让我们以直观地方式看看各参数能带来什么样的图片变化。"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "下载之后是个压缩包，请解压到本课件的根目录，解压后的文件路径为./17flowers/"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<img src='./image/17flowers.png' />"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 对MNIST的随机旋转"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "我们可用mnist数据集对图片进行随机旋转，旋转的最大角度由参数定义。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2023-01-04T02:42:44.395177Z",
     "start_time": "2023-01-04T02:42:44.389183Z"
    }
   },
   "outputs": [],
   "source": [
    "import tensorflow as tf\n",
    "from tensorflow.keras.preprocessing.image import ImageDataGenerator\n",
    "from tensorflow.keras.datasets import mnist\n",
    "import matplotlib.pyplot as plt\n",
    "from tensorflow.keras import backend as K\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2023-01-04T02:42:45.033529Z",
     "start_time": "2023-01-04T02:42:44.795415Z"
    }
   },
   "outputs": [],
   "source": [
    "# K.set_image_dim_ordering('th')\n",
    "K.set_image_data_format('channels_first')\n",
    "\n",
    "(train_data, train_label), (test_data, test_label) = mnist.load_data()\n",
    "train_data = train_data.reshape(train_data.shape[0], 1, 28, 28)\n",
    "train_data = train_data.astype('float32')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2023-01-04T02:42:45.455009Z",
     "start_time": "2023-01-04T02:42:45.382002Z"
    }
   },
   "outputs": [],
   "source": [
    "# 创建图像生成器，指定对图像操作的内容，rotation_range指定了旋转的最大角度，\n",
    "# 但是每张图片的旋转角度是在这个最大角度中的随机数\n",
    "datagen = ImageDataGenerator(rotation_range=90)\n",
    "# 图像生成器要训练的数据\n",
    "datagen.fit(train_data)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2023-01-04T02:42:46.574964Z",
     "start_time": "2023-01-04T02:42:46.298047Z"
    }
   },
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAfoAAAGgCAYAAABCAKXYAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjUuMywgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy/NK7nSAAAACXBIWXMAAA9hAAAPYQGoP6dpAABGvElEQVR4nO3de1RU5f4/8A8YDBQwiApIMkplWZl1IkDSzIpE7eal2+lmp4s/dahMyxadrFXf+vL9djVLq/UtoU55iU5qadkFFLNQg9JCjW6WFM54KWbwBgr790fLffbng8wwzG3vPe/XWrPWfrMHeIAPPOx59vM8UYqiKAQAAACmFB3uBgAAAEDwoKMHAAAwMXT0AAAAJoaOHgAAwMTQ0QMAAJgYOnoAAAATQ0cPAABgYujoAQAATAwdPQAAgImhowcAADCxoHX08+bNowEDBlBcXBzl5eXRxo0bg/WpAAIKtQtGhdqFY4kKxlr3S5YsoVtuuYVefvllysvLozlz5lB5eTnV19dTamqqx/dtb2+nxsZGSkxMpKioqEA3DYJAURRqbm6mjIwMio429otEqN3Igtr9C2rXeHyqXSUIcnNzFbvdrua2tjYlIyNDKSkp8fq+DQ0NChHhYcBHQ0NDMMoppFC7kflA7aJ2jfroSu0G/F/Y1tZWqq2tpYKCAvVt0dHRVFBQQNXV1R2e39LSQm63W30o2EzPsBITE8PdBL+gdiMXahe1a1Rdqd2Ad/R79uyhtrY2SktLY29PS0sjh8PR4fklJSVktVrVh81mC3STIESM/pIfajdyoXZRu0bVldoN+6BUcXExuVwu9dHQ0BDuJgF0CWoXjAq1G1mOC/QH7N27N/Xo0YOcTid7u9PppPT09A7Pt1gsZLFYAt0MAJ+hdsGoULvgScCv6GNjYyk7O5sqKirUt7W3t1NFRQXl5+cH+tMBBAxqF4wKtQsedf8ez84tXrxYsVgsSllZmbJ161Zl8uTJSnJysuJwOLy+r8vlCvtdjHh07+FyuYJRTiGF2o3MB2oXtWvUR1dqN+Av3RMRXXfddbR79256+OGHyeFw0DnnnEOrVq3qcKNIpBsyZAjLLpeL5Z07d7Lc2toa9DZFOtQuGFUk1a4cdvD0NR4+fJjlI0eOsLx7926W5c1tiglmJARlwRx/uN1uslqt4W5GSJito3e5XJSUlBTuZoRNJNWu2aB2jVW76Oj/oyu1G/a77gEAACB40NEDAACYWFDG6OE/zj33XPX4gQceYOfk3bDLly9n+bXXXmP522+/ZbmtrS0QTQTwmVxbu729PUwtgUjQq1cvlpctW8byvn37WI6JiVGPY2Nj2bl169ax/P3337MsX8r/4IMPWNb7S/nHgit6AAAAE0NHDwAAYGLo6AEAAEwMY/QBJpebvOmmm9TjkSNHsnN9+vRh+Y477mA5NzeX5euuu47lHTt2sIxxUggk7RSmzMxMdu7EE09kubm5mWVt3RNRh7XUP/74Y5a3bNnS7XaC+cTFxbFst9tZHjx4MMtyaqB2HF3+XczOzmZZjuGvWbOG5aFDh7I8e/bsTlqtX7iiBwAAMDF09AAAACaGl+79pJ0+R0T06KOPsjx27Fj12Nu+wT169GBZ7hE9fvx4ll966SWWDx065LmxAD4488wz1eNbb72VnZO1KKc/ydXIvK3cVVVVxfI999zD8ubNmz2+P5ibHDpKTEz0+Hzt39r9+/ezc95q8ZJLLmG5oKCA5dNPP53lq6++2uPH0wNc0QMAAJgYOnoAAAATQ0cPAABgYhij95Mco5fjN9qxIrm0olzSVn6snj17suxtjB4gkLKystRjOW4pdws77jj+p0Quzyyn16WmprLcv39/ls866yyW5e8KppKamxyD/+WXX1huampiOSUlheXPP/9cPZ47dy47J6fmyWnMl156Kcvy3qqJEyeyvHDhQpZvuOEG0htc0QMAAJgYOnoAAAATQ0cPAABgYhij95Ecr5G5d+/eLGvHKisqKti5N954g+VXXnmF5eTkZJb//PNPluVyu3IcC8Af2rnr27ZtY+fk/SZ79+5luaamhmU5pjpjxgyW5ZK6ct50QkICy263u5NWgxnI+pJ/O+W6DnKMXlt/snblduCjR49m+YcffmD52muvZVkuXf73v/+d5YMHD7J8++23U7jhih4AAMDE0NEDAACYGDp6AAAAE8MYvRdyDP60005j+e6772Y5Pj6e5aVLl6rHzz77LDsnxy0//fRTlq+88kqWq6urWf711187aTWAd7K2tVt7EvG57/fddx8753Q6WZZj6HLfBTnPPj8/n2U591jef9LS0kIQudavX8/ya6+9xvJ///d/s6wdw9+4cSM7V1dXx/IHH3zA8meffcay/Dsr92GQ95fcdtttLMvanTZtGoUarugBAABMDB09AACAiaGjBwAAMDGM0XuhXe+biGjWrFksy7Xt5RxK7VimnFss59xnZGSwLPf4PuGEE1iW+yq7XC4C6Cq5l4KsJ0/ryUdH82sEOSYvf28KCwtZvuaaa1j+6aefWJbjqBijBy25fv3555/P8oUXXqge5+XlsXOrVq1iWa5PIvMzzzzDsvy9kGvjy1q/4447WNbW8r333kuhgCt6AAAAE0NHDwAAYGI+d/Rr166lK664gjIyMigqKoqWLVvGziuKQg8//DD17duX4uPjqaCgoMOSggDhgNoFo0Ltgj98HqPfv38/nX322XTbbbfRhAkTOpx/8sknae7cufT6669TVlYWzZ49mwoLC2nr1q0UFxcXkEYHk2yjHIO/+OKLWfY2Vrly5cpOn3vyySezLOdjynn2Mh84cICg68xeu76SY4kzZ85k+ccff1SP5b0nspbl/SYxMTEsy98juVb9kiVLWF69enVnzY5IqF1Orvkg576fccYZ6rEcv7darSw3NzezfOTIEY+f+7nnnmP5ww8/ZHnkyJEsT548meWbb75ZPZb9RXFxscfP3V0+d/RjxoyhMWPGHPOcoig0Z84ceuihh+iqq64ior82bklLS6Nly5bR9ddf3+F9Wlpa2M0J2KwCggW1C0aF2gV/BHSMfvv27eRwOKigoEB9m9Vqpby8vA6ruh1VUlJCVqtVfchdqwBCAbULRoXaBW8C2tE7HA4iIkpLS2NvT0tLU89JxcXF5HK51Id22U2AUEHtglGhdsGbsM+jt1gsZLFYwt0MlRwz+f3331n2Np+3qqqK5e+//77T5/bv359lOY9errEs/zs/fPiwx7ZAcOmtdr2R+zTcddddLJ977rkesz/k3OTHH3+c5fLycpYbGxsD9rmhI6PVriT/TldWVrKs3YNk4MCB7Nw//vEPlh955BG/2vLdd9+xLPsMuSbE7Nmz1eObbrqJnZN/819++WW/2nZUQK/oj25EITe8cDqdHTapANAT1C4YFWoXvAloR5+VlUXp6elUUVGhvs3tdtOGDRs67FYFoCeoXTAq1C544/NL9/v27WPTbrZv306bNm2ilJQUstlsNH36dHr88cdp4MCB6jSPjIwMGjduXCDbDeAz1C4YFWoX/OFzR19TU0MXXXSRmmfMmEFERJMmTaKysjKaNWsW7d+/nyZPnkxNTU00fPhwWrVqlWHmcsr95M877zyW5dexfft2lt98881Oz8u5x3LsSI6Zaf9DJ/I83t8Vcm1z7f0GkTAn3+y1602fPn1YlvW4d+9elnfu3KkeH3/88ezcSSed5PFz7d69m+X58+ezvHDhQpbly87Ama12Tz31VJblz9/bvh1yHn1bWxvL2r0TLrjgAnZO1q7cU0T+HvhKzstft24dyw888IB6PHz4cHbut99+Y1n+3nX377TPHf3IkSM7fJO1oqKi6LHHHqPHHnusWw0CCBbULhgVahf8gbXuAQAATAwdPQAAgImFfR693sh58u+++y7LN9xwA8ty/u/atWtZTkhIUI+Tk5PZuaFDh7K8ZcsWluUayr6OHckx//vvv5/l119/XT3W3uhD1HHMC4zvq6++YvnWW29luV+/fixr71c57jj+p0LeX3LjjTeyLGtdjj3u2rXLa3vBPB599FGW5Tj51KlT/fr48l4p7XomI0aMYOeuuOIKlufNm8eyv2P03mzatEk9lusByD0lAnXvFK7oAQAATAwdPQAAgInhpXuhvb2d5T/++IPladOmsSzvhM3Ly2P58ssvV48vvfRSdk6+/Llv3z6W5dQLOdWivr6eZbkDlVxSV7t1IxFffvGZZ55h5+SWuBB6cvqbnColl0D2tiSyfBlQ1o9cytOTxMREluW0VLvdzrJs+wknnMCyrH0wF7lVrBwqlLXuKzmlTbskrvZvMBFRSkoKy3JbWtkW2ScEki+/c/7AFT0AAICJoaMHAAAwMXT0AAAAJoYxeh/98MMPLI8ePZrl0tJSlnv37q0eR0VFefzYctzyoYceYlluYSjz1q1bWZZjS3L6lHZnK7ltKISfvKdCLscsxyU///xzljvbi/woTyuteSPH1FeuXMnyrFmzWJbT7eQ0IjCXtLQ0ln/55ReWs7KyWD7nnHNYltOUvZF/67Rj3/I+K7k9+GWXXcaydvrbsT62EeGKHgAAwMTQ0QMAAJgYOnoAAAATwxi9j+Rc9gkTJrAs5xf7o0ePHizLZSNl1m5jSUT0559/sizvAdAuE3naaaexc998841vjYWAKywsZPnuu+9m2Wq1sizHyeXaCJs3b2bZn2WO5fi+rEU5N/nMM89kWc67xzx6c5E/X/n3JTc3l2W51PiGDRtYlkuTe6OdCy+Xfpb3ugwZMoRluYWuvPfJiHBFDwAAYGLo6AEAAEwMHT0AAICJYYzeCzn/d9KkSSyPHTuWZbk1rJYc15Tz6t98802W5bil3NZWzlVNSkpiWW6BqN0yl4jo999/V4/leD6ER2xsrHos5+9mZmay/NNPP7Es5823trayLO/5CORWxHJfBfmx5ZitXNMhVGt+Q2jI9eJlLcq/kyUlJSyvWbOG5e+//96nz79z5071+NNPP2XntHt8EHXcU2Tw4MEeP7e8/8QIcEUPAABgYujoAQAATAwdPQAAgIlhjF6Q4+Zy3PuWW25huU+fPizLNZ21ex/LMdcnnniC5VdeeYVlOU9arg8t12yura1lWc6rl+Omb731lnq8f/9+An3Ztm0byzt27GD5xBNPZFnOZZf3ZARzze5vv/2WZTkmK9e2R72Zm9yHQe7LIe8/kvcTyf3r5d9VWV+efP311yyvX7+e5ZEjR7Is2+7PnhB6gSt6AAAAE0NHDwAAYGLo6AEAAEwMY/SCnO8rx+TluLjT6WT5yy+/ZHnUqFHqsRynKi8vZ1mOW8msnfdORLRu3TryRM7/lHNXsSe4/mjHHuX+8q+++irL999/P8sjRoxg+cCBAyzLucp1dXXdbqf8PYmLi/OY5f0FDQ0N3f7coH9yXY6PPvqI5auvvpplea+TXAt/yZIl3W7LgAEDWJZ/w+XviazdQK43ES64ogcAADAxdPQAAAAm5lNHX1JSQjk5OZSYmEipqak0btw4qq+vZ885dOgQ2e126tWrFyUkJNDEiRM7vLwNEGqoXTAq1C74y6cx+qqqKrLb7ZSTk0NHjhyhBx98kEaNGkVbt25V9zq/9957aeXKlVReXk5Wq5WKiopowoQJHcYb9ULOmy8oKGBZjoPK/ejlnMxNmzaxfOWVV6rHci6onDsaaHLedCSPyRuxduV83gULFrAs92G46qqrWJbjoNdeey3LL730EstyP3vt+vPy/pLCwkKWb7/9dpblGLxcEwC6zoi1K9eDl3sZLF68mOU777yT5fHjx7Ms60neb+KJ/Dt43HG82+vbty/LckzfDHzq6FetWsVyWVkZpaamUm1tLY0YMYJcLhe99tprtHDhQrr44ouJiKi0tJROP/10Wr9+fYdNWYiIWlpaqKWlRc1ut7s7XweAR6hdMCrULvjLrzF6l8tFRP9Z/a22tpYOHz7MrooHDRpENpuNqqurj/kxSkpKyGq1qg+5QxdAMKB2wahQu+Crbnf07e3tNH36dBo2bJi6rZ/D4aDY2NgOLymmpaV12ELzqOLiYnK5XOoD024g2FC7YFSoXeiObs+jt9vtVFdX53UutzcWi8XjHu7BJsfozzrrLI/Pl3t6y7nql156Kcval8fkmstY7zs8jFq7jY2NLM+dO5dluT/9pEmTWJZzk+UY/pAhQ1g+euVIRGSz2dg5uQeE3I++srKSZYzRB4ZRa/ebb75h+f3332d57NixLMt6mjZtGstyrftnnnmm08/922+/sSz/qZF7iPTs2ZNl7X4lRER//PFHp59Lr7p1RV9UVEQrVqyg1atXU79+/dS3p6enU2trKzU1NbHnO51OSk9P96uhAIGA2gWjQu1Cd/nU0SuKQkVFRbR06VKqrKzs8J9QdnY2xcTEUEVFhfq2+vp62rFjB+Xn5wemxQDdgNoFo0Ltgr98eunebrfTwoULafny5ZSYmKiO/1itVoqPjyer1Uq33347zZgxg1JSUigpKYnuuusuys/PP+adnwChgtoFo0Ltgr986uiPzruV+/eWlpbSrbfeSkREzz33HEVHR9PEiROppaWFCgsLaf78+QFpbDDIOZY//vgjy3KN7jPPPJPl66+/nmU5dqRd8/mzzz5j5+QayxA8ZqxdOdb4xhtvsFxbW8vy1KlTWZZXe3IPcO39JXL9iMOHD7Ms7z9ZuHAhy3K/eug6M9buihUrWJbz5m+77TaW5Vz3e+65h2V5r9WHH36oHqemprJzcoaBts6JOt4LY4Z7qXzq6BVF8fqcuLg4mjdvHs2bN6/bjQIINNQuGBVqF/yFte4BAABMDB09AACAiWE/ekGOayYkJHh8vtyXWzv3mIiPm8pxzK68JAfQGVk/cixx48aNLMuNUOSY70UXXcTynj171OMffviBnZNzjb3t+QDgidwrITY2luUbb7yRZTlm/9BDD7GsXSVQjt/L+6z27dvHsrxvS47hGxGu6AEAAEwMHT0AAICJ4aV7QW5hKKda9O/f3+P7b968meX33ntPPZYrVwGEkhxWWr58ucesJYeozPjyJujHzTffzHJcXBzLEydOZFkuySy3UfZk69atLK9Zs4Zl+dK/EYdccUUPAABgYujoAQAATAwdPQAAgIlF/Bi9HH+R0zZOOeUUj+8vpzS9/fbbLG/ZssWP1gHow8GDB8PdBIhg11xzDctym9tLLrmEZe0WvHJM/fPPP2e5tLSUZbkNrRHH5CVc0QMAAJgYOnoAAAATQ0cPAABgYhE/Ri/HX2pqalhesmQJy3KLQzkmv2jRogC2DgAApKuuuorlF154geX09HT1WK6FIrcL/+STT1jWbi1uFriiBwAAMDF09AAAACaGjh4AAMDEohSdTRJ0u91ktVrD3QyV3I7TjOM3geJyuTqsOR1J9Fa70HWoXWPXrtyjZMCAAeqxw+Fg51pbWz1mo+lK7eKKHgAAwMTQ0QMAAJiY7qbX6WwkQXft0bNI/15F+tdvZJH+szP61y/br91GWZ4z+tcqdeXr0V1H39zcHO4mMNhDvuuam5sNPc7nL73VLnQdatfYtdvW1sbyzz//HKaWhF5Xald3N+O1t7dTY2MjKYpCNpuNGhoaIvomGV+53W7KzMwM6fdNURRqbm6mjIwMio6O3NEg1K5/ULvhg9r1j95rV3dX9NHR0dSvXz9yu91ERJSUlISC64ZQf98i+WroKNRuYKB2Qw+1Gxh6rd3I/RcWAAAgAqCjBwAAMDHddvQWi4UeeeQRslgs4W6KoeD7Fn74GXQPvm/hh59B9+j9+6a7m/EAAAAgcHR7RQ8AAAD+Q0cPAABgYujoAQAATAwdPQAAgImhowcAADAx3Xb08+bNowEDBlBcXBzl5eXRxo0bw90k3SgpKaGcnBxKTEyk1NRUGjduHNXX17PnHDp0iOx2O/Xq1YsSEhJo4sSJ5HQ6w9TiyILa7RxqV99Qu50zdO0qOrR48WIlNjZWWbBggbJlyxblzjvvVJKTkxWn0xnupulCYWGhUlpaqtTV1SmbNm1Sxo4dq9hsNmXfvn3qc6ZMmaJkZmYqFRUVSk1NjTJ06FDl/PPPD2OrIwNq1zPUrn6hdj0zcu3qsqPPzc1V7Ha7mtva2pSMjAylpKQkjK3Sr127dilEpFRVVSmKoihNTU1KTEyMUl5erj5n27ZtChEp1dXV4WpmREDt+ga1qx+oXd8YqXZ199J9a2sr1dbWUkFBgfq26OhoKigooOrq6jC2TL9cLhcREaWkpBARUW1tLR0+fJh9DwcNGkQ2mw3fwyBC7foOtasPqF3fGal2ddfR79mzh9ra2igtLY29PS0tjRwOR5hapV/t7e00ffp0GjZsGA0ePJiIiBwOB8XGxlJycjJ7Lr6HwYXa9Q1qVz9Qu74xWu3qbpta8I3dbqe6ujpat25duJsC4BPULhiV0WpXd1f0vXv3ph49enS4U9HpdFJ6enqYWqVPRUVFtGLFClq9ejX169dPfXt6ejq1trZSU1MTez6+h8GF2u061K6+oHa7zoi1q7uOPjY2lrKzs6miokJ9W3t7O1VUVFB+fn4YW6YfiqJQUVERLV26lCorKykrK4udz87OppiYGPY9rK+vpx07duB7GESoXe9Qu/qE2vXO0LUbrLv8XnzxRaV///6KxWJRcnNzlQ0bNnT5fRcvXqxYLBalrKxM2bp1qzJ58mQlOTlZcTgcwWquoUydOlWxWq3KmjVrlJ07d6qPAwcOqM+ZMmWKYrPZlMrKSqWmpkbJz89X8vPzw9hq40DtBg9qN7hQu8Fj5NoNyja1S5YsoVtuuYVefvllysvLozlz5lB5eTnV19dTamqqx/dtb2+nxsZGWrhwIb3wwgvkdDppyJAh9OSTT9J5550X6KYaktVqPebb58+fTzfeeCMR/bVwwz//+U965513qKWlhS655BJ69tlnO9xsEwiKolBzczNlZGRQdLTuXiTyCWo3uFC7wYPaDS5D124w/nvwZz5mQ0ODQkR4GPDR0NAQjHIKKdRuZD5Qu6hdoz66UrsB/xfW1/mYLS0t5Ha71YcS+BcYIEQSExPD3QS/oHYjF2oXtWtUXandgHf0vs7HLCkpIavVqj5sNlugmwQhEhUVFe4m+AW1G7lQu6hdo+pK7YZ9UKq4uJhcLpf6aGhoCHeTALoEtQtGhdqNLAFfMMfX+ZgWi4UsFkugmwHgM9QuGBVqFzwJ+BU95mOCUaF2wahQu+BR9+/x7Jw/8zFdLlfY72LEo3sPl8sVjHIKKdRuZD5Qu6hdoz66UrtBWzDnhRdeUGw2mxIbG6vk5uYq69ev79L7oeCM+zDDH0tFQe1G4gO1i9o16qMrtRuUBXP84Xa7O12YAPTN5XJRUlJSuJsRNqhd40LtonaNqiu1G/a77gEAACB40NEDAACYGDp6AAAAE0NHDwAAYGLo6AEAAEwMHT0AAICJoaMHAAAwsYCvdQ8A4SHn0h44cIDltrY2lnW2hAYABAmu6AEAAEwMHT0AAICJ4aV7AIOKj49n+aOPPmJ5xYoVLJeXl7P8/fffB6dhAKAruKIHAAAwMXT0AAAAJoaOHgAAwMQwRg9gUL169WJ53bp1LI8ePZrlhIQEll955RWWd+zYwXJ7e7u/TQQAHcAVPQAAgImhowcAADAxdPQAAAAmhjF6L+RcZTlumZiYyPKePXuC3iYAIqLffvuN5RdffNHj86+99lqW+/Xrx3JZWRnLn332mXrc2trajRYCgB7gih4AAMDE0NEDAACYGDp6AAAAE8MYvRePPvooy+np6Sz//PPPLP/6668sf/XVV+rx5s2bA9w6gP+QY/avvfYay/v27WP5xhtvZLlPnz6d5k8//ZSd27t3L8vY8hZAv3BFDwAAYGLo6AEAAEwMHT0AAICJYYxeGDFiBMv333+/T+9fV1fHsnYsc//+/ezcG2+8wfKSJUt8+lwAWm1tbSx/9913LD/33HMsHzlyhOXrrruO5fnz56vH8+bNY+deffVVluW9KQBGFRUVxbK8/8RqtbLc0tLC8nnnncey/D3r3bu3eizXp4iJiWF55cqVXWixd7iiBwAAMDGfO/q1a9fSFVdcQRkZGRQVFUXLli1j5xVFoYcffpj69u1L8fHxVFBQQD/88EOg2gvQbahdMCrULvjD545+//79dPbZZ3d4Ke+oJ598kubOnUsvv/wybdiwgU444QQqLCykQ4cO+d1YAH+gdsGoULvgjyjFjwmwUVFRtHTpUho3bhwR/fVfZUZGBs2cOZPuu+8+IiJyuVyUlpZGZWVldP3113v9mG63u8MYSCi9++67LI8fPz5kn1veD7B48WKW5TxpvXG5XJSUlBTuZnSJGWvXGzn2eOKJJ7I8ZswYlu12u3os93TYsGEDy++//z7Ly5cvZ/nAgQO+NTbEULv6rl1/HH/88Swfdxy/NS0lJYVlWQc5OTksa8fYiYiGDh3K8gknnMDy8OHDWW5ublaPLRYLO+dyuVh+9tlnWX7++edJ6krtBnSMfvv27eRwOKigoEB9m9Vqpby8PKqurj7m+7S0tJDb7WYPgFBD7YJRoXbBm4B29A6Hg4iI0tLS2NvT0tLUc1JJSQlZrVb1kZmZGcgmAXQJaheMCrUL3oT9rvvi4mJyuVzqo6GhIdxNAugS1C4YFWo3sgR0Hv3RdeCdTif17dtXfbvT6aRzzjnnmO9jsVg6jFOEkhxfkWPycv95eSdrUVERy3L98FtuuUU9jo72/H/VpEmTWG5qamJZzl2GwNFj7Z5yyiksJyQksLxp0yafPp68HUfe8/H666+zvGfPHvV41qxZ7Jyccy/3tpf+/e9/s4z97QNHj7UbavJ+k7POOks9zsrKYufGjh3L8sGDB1m+4IILWF6zZg3L8vdQ/p3u0aMHy0899RTL2nsG5P4S8n4BOb6/YMEC9VhRlA77V3QmoFf0WVlZlJ6eThUVFerb3G43bdiwgfLz8wP5qQACCrULRoXaBW98vqLft28f/fjjj2revn07bdq0iVJSUshms9H06dPp8ccfp4EDB1JWVhbNnj2bMjIy1DtEAcIFtQtGhdoFf/jc0dfU1NBFF12k5hkzZhDRXy87l5WV0axZs2j//v00efJkampqouHDh9OqVasoLi4ucK0G6AbULhgVahf84dc8+mAI9XzOuXPnsjxt2jSW5fhLcXExy9oxEyLq8FLZ008/rR7n5uayc3JesyT3un/wwQdZfu+991iWY02hZqS5yMEQ6NqdM2cOyyeffDLLctxbzl3/888//fr82vnCf//739m5Rx99lGU5z147vk9EtHTpUpbl/QDffvsty6Ged4/a1dc8ernmu8xXXnkly6NHj2Z54MCB6vH555/Pzsnakn9nZW3LqYeVlZUsp6amsiznwst7IbQfr2fPnuyc/Bk0NjZ6bPvRzxfSefQAAACgL+joAQAATAwdPQAAgIlF3H70cr5vRkYGy3IOZK9evVjW3vlK1HEPcDmWpD3vbUxeOumkk1iW8+zlnP6vvvrKp48P+iPn6GpddtllLJ922mksy9qVeyXs37/fp7bs3btXPX7rrbfYuaNzt4+aOXOmx/PXXHMNy3L9cTlmv27dOpYPHz7chRaDUcXHx7OsnQdPRPTEE0+wPGDAAJZlbdfX16vH2rXliTqujSLvZZF/wz/++ONOWv0XOY4uefq92717t8ccKLiiBwAAMDF09AAAACYWcS/dy5dR5Ms00q5du1iWU5zk7lDy4w0ZMsTXJnZKbiOanJzMspwC9euvvwbsc0NotLS0qMdffPEFOydf/pYvX8qpoX/88QfLcjqmHHaStDNv5ceS29LKZUXlsILcuvPaa69lWS5F/corr7AspxLu2LGjs2aDAcTGxrI8ZcoUluWysU6nk+VvvvmG5SeffJJl7Uv377zzDjt36aWXsnzHHXew/Oabb3psqxGXb8YVPQAAgImhowcAADAxdPQAAAAmZvoxejmN54wzzmB52LBhHt9fLm8opz/Ijz9x4kSWtdOl5Ji5HOeU4/9yTF7KyclhedSoUSz/61//YvnQoUMePx6En3Yamdwe8+6772ZZTreT4+T/93//x3JmZibLcsqcdjqdN3KMXE71PO44/qelvLyc5ZEjR7J84YUXsnznnXeyLKdAacdRjxw54r3BEFZynFtOFb7nnntYlvX09ttvsyzH8D1t17ps2TKW5ZK3chvbU089lWUz1Beu6AEAAEwMHT0AAICJoaMHAAAwMdOP0ctt/eRSmnK8Ri55K99/+PDhLK9cuZLl8ePHd9oWOYf/3XffZXn16tUsyy1xZZb3B8itG7///nuW5bKi3uZRQ3jJNRzkXPKffvqJZbmGw7hx41i+6aabWP7uu+9YrqioYNlTfcjfIzmGLreplVt7fvrppywf3V/9KPl7JLdo1i6ROnv2bHYu1Fvcgnfyb9W5557LsqzdqVOnsvzhhx92+3PLWpXj/3LZaXkvlVye19elpPUAV/QAAAAmho4eAADAxNDRAwAAmJjpx+iljRs3suxpW1CijmNLZ555JsvPPvssy3369On0Y0VH8/+r5Ji8JM/LtezlVo6XX365x8/3+++/syy33AVj2bx5M8v3338/y7/88gvLcj3xuXPnsvz//t//Y1l7T4ccr5dbf8palPOm5TznTZs2saxd45+I6ODBgyzL+w0KCgrU46VLl3babtAHuT68nKsut3qVY/byb5ncatZisXR6Xs6Tv+6661jW7ulwrI8ta9OIcEUPAABgYujoAQAATAwdPQAAgIlF3Bi93MNd7lUs5+RKf/vb31jWjhV68/zzz3f5uUQd5znL9cLlOKjcA1zOVZX7MGvHveSYaENDg09thdCTY4sOh4PlJUuWsJyWlsaynFc/c+ZMlrXzj+W9LdnZ2SynpKSwLOtJ1rJUV1fH8v/8z/94fL52X4e+fft6fC6En1zb4KWXXmJ5wYIFLN93330sjxgxgmU5112uR6/d1yE9PZ2dy8vLY1muV/HJJ594/NhGhCt6AAAAE0NHDwAAYGLo6AEAAEwsSpEDfWHmdrvJarWG7PPJPd/lmtpybXtv5Nikdq76kCFDfGydZ3K/8VtvvZVluSe4pF0LX+73/Nhjj7Es14s+FpfLRUlJSV6fZ1ahrl1vevTowfIpp5zCcllZGcuDBg1i+aOPPlKPv/76a3Zu6NChLMvfE3k/yT//+U+W5dr43vTs2ZNl7b00cjxXzoPuCtRueGtX7uNxySWXsCzXM5H3m8h59tu3b1eP5b4LX331Fcvz589nWd5PoPcx+q7ULq7oAQAATMynjr6kpIRycnIoMTGRUlNTady4cVRfX8+ec+jQIbLb7dSrVy9KSEigiRMnktPpDGijAXyF2gWjQu2Cv3zq6Kuqqshut9P69evpk08+ocOHD9OoUaPYtn333nsvvf/++1ReXk5VVVXU2NhIEyZMCHjDAXyB2gWjQu2Cv/wao9+9ezelpqZSVVUVjRgxglwuF/Xp04cWLlxIV199NRH9NX/29NNPp+rq6g7jescS6rEiuf+8dn4uEVF+fj7L8mvIyclhedGiRSwvXLhQPV6xYkW320lEFBUVxfLtt9/Oslx3PzExscsfe+vWrSzfc889LMv9w4/FSOOcZqhdX8lxTLnOwuOPP86y9ndh79697Jz8vfnmm29YlveLyPqSa597I+830c7bl/OguwO1G97alfeTxMXFsSz3UpBr28ufndvtVo/l/vPylQ55T4fOblvzKuhj9C6Xi4j+80tXW1tLhw8fZovIDBo0iGw2G1VXVx/zY7S0tJDb7WYPgGBD7YJRoXbBV93u6Nvb22n69Ok0bNgwGjx4MBH9tTJXbGxsh9Xn0tLSOqzadVRJSQlZrVb1oV3RCCAYULtgVKhd6I5ud/R2u53q6upo8eLFfjWguLiYXC6X+sDSqxBsqF0wKtQudEe31rovKiqiFStW0Nq1a6lfv37q29PT06m1tZWamprYf5dOp7PDesNHWSyWDuMtoSTHHuUY+xdffMGyHO/KyMhg+d1332VZrsnsDzl29Oqrr7L82WefsSzHXOWe3tpxzzPOOIOdC8S4px6ZqXZ9Jcci5bj58uXLWdauZ9+7d292Ts6rl/Pk5X7z/pJzmc1an56YuXbb2tpY1t5oSES0fv36UDbHdHy6olcUhYqKimjp0qVUWVlJWVlZ7Hx2djbFxMRQRUWF+rb6+nrasWNHh5vaAEIJtQtGhdoFf/l0RW+322nhwoW0fPlySkxMVMd/rFYrxcfHk9Vqpdtvv51mzJhBKSkplJSURHfddRfl5+d36c5PgGBB7YJRoXbBXz519Ee3Fhw5ciR7e2lpqTqd5rnnnqPo6GiaOHEitbS0UGFhYYclBgFCDbULRoXaBX9F/Fr3/oqPj2dZrnUfSnLuqXzZbvTo0Sxrx+WbmprYuZtvvtnnz2+kucjBYLTalbTjvkRETz31lHp81VVXsXOff/45y3LdBTn+r3eoXWPXbiTDWvcAAAARDh09AACAiaGjBwAAMLFuzaOH/wjnmLx06NAhllevXs1yXV0dy9o9nX/77bfgNQwMobGxkeUnnnhCPZbzsXNzc1keNmwYy99//z3Let/TG8DMcEUPAABgYujoAQAATAwv3UeQ3bt3e8wQ2eQSudu3b1eP5drqcttYuUQuXqoH0A9c0QMAAJgYOnoAAAATQ0cPAABgYhijB4Bj0m4VKrewXbVqFcsHDhwISZsAwHe4ogcAADAxdPQAAAAmho4eAADAxDBGDwBeORyOcDcBALoJV/QAAAAmho4eAADAxNDRAwAAmBg6egAAABNDRw8AAGBi6OgBAABMTHcdvaIo4W4CdFOk/+wi/es3skj/2UX6129kXfnZ6a6jb25uDncToJsi/WcX6V+/kUX6zy7Sv34j68rPLkrR2b9y7e3t1NjYSIqikM1mo4aGBkpKSgp3swzD7XZTZmZmSL9viqJQc3MzZWRkUHS07v53DBnUrn9Qu+GD2vWP3mtXdyvjRUdHU79+/cjtdhMRUVJSEgquG0L9fbNarSH7XHqF2g0M1G7ooXYDQ6+1G7n/wgIAAEQAdPQAAAAmptuO3mKx0COPPEIWiyXcTTEUfN/CDz+D7sH3LfzwM+gevX/fdHczHgAAAASObq/oAQAAwH/o6AEAAEwMHT0AAICJoaMHAAAwMd129PPmzaMBAwZQXFwc5eXl0caNG8PdJN0oKSmhnJwcSkxMpNTUVBo3bhzV19ez5xw6dIjsdjv16tWLEhISaOLEieR0OsPU4siC2u0calffULudM3TtKjq0ePFiJTY2VlmwYIGyZcsW5c4771SSk5MVp9MZ7qbpQmFhoVJaWqrU1dUpmzZtUsaOHavYbDZl37596nOmTJmiZGZmKhUVFUpNTY0ydOhQ5fzzzw9jqyMDatcz1K5+oXY9M3Lt6rKjz83NVex2u5rb2tqUjIwMpaSkJIyt0q9du3YpRKRUVVUpiqIoTU1NSkxMjFJeXq4+Z9u2bQoRKdXV1eFqZkRA7foGtasfqF3fGKl2dffSfWtrK9XW1lJBQYH6tujoaCooKKDq6uowtky/XC4XERGlpKQQEVFtbS0dPnyYfQ8HDRpENpsN38MgQu36DrWrD6hd3xmpdnXX0e/Zs4fa2tooLS2NvT0tLY0cDkeYWqVf7e3tNH36dBo2bBgNHjyYiIgcDgfFxsZScnIyey6+h8GF2vUNalc/ULu+MVrt6m73OvCN3W6nuro6WrduXbibAuAT1C4YldFqV3dX9L1796YePXp0uFPR6XRSenp6mFqlT0VFRbRixQpavXo19evXT317eno6tba2UlNTE3s+vofBhdrtOtSuvqB2u86Itau7jj42Npays7OpoqJCfVt7eztVVFRQfn5+GFumH4qiUFFRES1dupQqKyspKyuLnc/OzqaYmBj2Payvr6cdO3bgexhEqF3vULv6hNr1ztC1G9ZbATuxePFixWKxKGVlZcrWrVuVyZMnK8nJyYrD4Qh303Rh6tSpitVqVdasWaPs3LlTfRw4cEB9zpQpUxSbzaZUVlYqNTU1Sn5+vpKfnx/GVkcG1K5nqF39Qu16ZuTaDVpH/+KLLyr9+/dXLBaLkpubq2zYsMGn93/hhRcUm82mxMbGKrm5ucr69euD1FLjIaJjPkpLS9XnHDx4UJk2bZrSs2dP5fjjj1fGjx+v7Ny5M3yNNhDUbvCgdoMLtRs8Rq7doGxTu2TJErrlllvo5Zdfpry8PJozZw6Vl5dTfX09paamenzf9vZ2amxspMTERIqKigp00yAIFEWh5uZmysjIoOho3Y0G+QS1G1lQu39B7RqPT7UbjP8e/Fl4oaGhodP/nPDQ96OhoSEY5RRSqN3IfKB2UbtGfXSldgP+L6yvCy+0tLSQ2+1WH0rgX2AIqejoaPY4/vjj1YfZJSYmhrsJfon02o1kqF3UrlF1pXYD3tH7uvBCSUkJWa1W9WGz2QLdpJCKiorq9GF2Rv8aI712IxlqF7VrVF2p3bAvmFNcXEwzZsxQs9vtpszMzDC2yD9XXnkly/379+/0ue+99x7LP//8c1DaBMFhttqFyIHajSwB7+h9XXjBYrGQxWIJdDMAfIbaBaNC7YInAX/pHgsvgFGhdsGoULvgSVBeup8xYwZNmjSJzjvvPMrNzaU5c+bQ/v376R//+EcwPh1AwKB2wahQu9CZoHT01113He3evZsefvhhcjgcdM4559CqVas63ChiRgcPHmT5gQceUI+PHDnCzj300EMs//Of/2T5lVdeCXDrwJtIrl2pV69e6rG8s/eXX34JcWvAG9QudCYoC+b4w+12k9VqDXczum306NEsl5aWqseyo4+Pj2fZ6B29y+WipKSkcDcjbIxeu1IkdfSoXXPVbiTpSu0aeykoAAAA8AgdPQAAgImFfR692ezZs6fTfNppp7Fzxx3Hv/0vv/wyy8OHD2d56tSpLO/bt6/b7QTwZtq0aerxwIED2TntkBQR0Zo1a1jW2YggQETDFT0AAICJoaMHAAAwMXT0AAAAJoYx+gD76quvWL777rvV4+LiYnYuISGB5by8PJblGP1NN93E8pIlS1j+888/fWssgMaZZ57J8uDBg9XjnJwcdm7+/PkhaRNAoMmlf+X9JK2traFsTkjgih4AAMDE0NEDAACYGF66D7D29naW165dqx5XVVWxc9qX9YmIsrOzWZZb3Mo1q7/55huWN27cyLJciQ9AKy4ujuUhQ4awfPnll6vHckjq8OHDLGM6HYSTHAa12Wzq8bBhw9g5uX/7ueeey7LL5WJ51apVLMvfBfl8PcIVPQAAgImhowcAADAxdPQAAAAmhjF6P0VH8/+V5Bh9W1tbp+/70ksvsbxz506W58yZw7IcQ5Xb3M6YMYPl7777rtPPDXDo0CGWr7jiCpa105Dq6+vZudra2uA1DMBH119/PcvanUDlUuP9+vVjWf6N3rt3L8sXX3wxywMGDGB50aJFLG/evJnld955h+VwjOnjih4AAMDE0NEDAACYGDp6AAAAE8MYvY9uvfVWluX4T0NDA8vffvutetzY2MjOtbS0sLxy5UqW5fj/008/zfJZZ53F8kUXXcSyHPM3wnxPCB15z0d6ejrLBw8eVI/ffPPNkLQJoDvksrY///yzeiyXdpb3UfXo0YPl+Ph4luX24omJiSzL9U3kmH9hYSHL2nuzQrW9M67oAQAATAwdPQAAgImhowcAADAxjNF70adPH5blevRyzL6uro7ljz76SD1et24dOyfHZ+Ta9HK+plxfPDMzk+WioiKWv/zyS5ZramoIIpdc49vtdrPcu3dvlrX1YsatO8E85LbJn332mXq8YMECdi45OZll+XdSrm1/8skns3zOOeewnJKSwnJGRgbLEydOZPnrr79Wj+V6FPJ3MlBwRQ8AAGBi6OgBAABMDB09AACAiWGM3ovdu3ezLPcivummm1jOy8tj+dRTT1WP5VriP/zwA8vaectEREOHDmVZ7k8v51zK+wnkHs0Q2WS9XHbZZSwnJSWxrK3PDRs2BK9hAH6S69FPmjRJPT7hhBPYuaVLl7I8e/Zslnfs2MGyvD9FjsFfcsklLD/++OMs22w2lrX3wsh1WIIFV/QAAAAmho4eAADAxHzu6NeuXUtXXHEFZWRkUFRUFC1btoydVxSFHn74Yerbty/Fx8dTQUFBh5eoAcIBtQtGhdoFf/g8QLB//346++yz6bbbbqMJEyZ0OP/kk0/S3Llz6fXXX6esrCyaPXs2FRYW0tatWykuLi4gjQ6n6upqllesWMHyhRdeyLJ27+OePXuyc3Kternmslzr3ts6yOvXr2f5jz/+8Pj8SBPptSvX8Jb69u3L8rvvvqsey/W9paampm63C7yL9NqV5P1Kd999N8sjRoxQj7Xz1omowz9JP/30E8ve/s7KPUs++eQTlvPz81m+4YYbWNb+nQ/V32ifO/oxY8bQmDFjjnlOURSaM2cOPfTQQ3TVVVcREdEbb7xBaWlptGzZMrr++us7vE9LSwvb3CVYCwYAoHbBqFC74I+AjtFv376dHA4HFRQUqG+zWq2Ul5fX4Ur4qJKSErJarepDrvYGEAqoXTAq1C54E9CO3uFwEBFRWloae3taWpp6TiouLiaXy6U+5DavAKGA2gWjQu2CN2GfR2+xWDrsJaxn3333Hctz5sxhWc7B1M6dHzx4MDsn51D+8ssvLMfExHh8/v79+1l+4403WP71118JgsdotSvXaZBzj/fu3cvy+eefrx7ffPPN7Jy8H0TuVy9rWe7jAOGl99qV95PI+5nkvh5yTYgvvvhCPZ4xYwY7t3XrVpb93QN+165dLG/ZsoVl+c/Wnj171GP5N17uZxIoAb2iT09PJyIip9PJ3u50OtVzAHqE2gWjQu2CNwHt6LOysig9PZ0qKirUt7ndbtqwYUOHOxEB9AS1C0aF2gVvfH7pft++ffTjjz+qefv27bRp0yZKSUkhm81G06dPp8cff5wGDhyoTvPIyMigcePGBbLdAD5D7YJRoXbBHz539DU1NXTRRRep+ej4x6RJk6isrIxmzZpF+/fvp8mTJ1NTUxMNHz6cVq1aZcq5nEQd9xOWczK1688PHDiQnZNjOXIN5s8//5xl7Zx8IqJt27axLOd3ulyuzpodkSK9docPH85ybm4uy3KdhzvvvFM9rqurY+dmzpzJsnyJ+L777mN53759vjUWmEirXbnPx5QpU1geO3Ysy3IPknnz5qnHmzZtCmzjhPb2dpbPOOMMluXvhraPiI2NZeeCNUbvc0c/cuRIjzcvREVF0WOPPUaPPfaYXw0DCDTULhgVahf8gbXuAQAATAwdPQAAgImFfR692ci57QsWLFCP5Rj9BRdcwHJ5eTnLGzduZHnNmjUBaCFEioSEBJanTp3KspzDK8dztfX24YcfsnPyJq9Ro0axLOfdL1q0iGWsjQ9ashYvv/xylkeOHMlyZWUly9oxeSKiL7/8MnCN82LQoEEsn3LKKSzLWtfO45f9RbDgih4AAMDE0NEDAACYmOlfurdarSzLl0rk1rDaHZ26Q06P0E6/k1OU5GIWd9xxB8ty6US5zGhbW1u32wnmJ6e0yXrq1auXx+drX/4sLS1l57799luW5UunV199Nctr165lGS/dRza5Lr98qV5u7XrgwAGW7733Xpa3b9/Osr/L2vrCZrOxnJqaynJVVRXL4ah9XNEDAACYGDp6AAAAE0NHDwAAYGKmHKPv3bu3eiynsI0YMYLlZcuWsVxdXc1ya2trwNq1fPlyluUyj+eddx7L2iUvj/X+GOcET+S2xrt372ZZ3uMha117T4l83/r6epajoqJYPvvss1mWv3dyC2UskRtZ5BK2s2bNYllu911WVubxfCjH5OX2vrL2k5KSWJbTov/888+gtMsTXNEDAACYGDp6AAAAE0NHDwAAYGKmHKPXuvLKK1mWY0NjxoxhWc5lX7dunV+fXzuv/tNPP2XnLr30UpbluOaFF17I8iWXXMLyv//9b7/aBuYmt89sbm5mWc5N/u2331iWc9+1fv75Z5b/93//l+UHH3yQZTlGL+83wRi9uSUnJ7Ms/w4fPHiQ5Tlz5rD8ySefsHzkyJGAtc1Xp512GsvTp09nWd7PIn+P3G53UNrlCa7oAQAATAwdPQAAgImhowcAADAxU47Ra9ez//jjj9m50aNHsyzXXJ48eTLLcmypoqKCZTm25IlcV1+OU8pxzNNPP91jWwA8iY7m/8fL9enlvg9yLrLcttbTx/7xxx89vu/gwYNZTk9PZ7mxsbHTzwXGk5iYyLK8Hyk3N5dluQ3yO++8w7LcQySU5HbPhYWFLMu17uX24qHaitYTXNEDAACYGDp6AAAAE0NHDwAAYGKmHKPXjptr99QmIqqpqWH5sssuY/mmm25iWY41yfeXY0ee5nfKtcW/+OILluUY6rnnnsvyGWecwbIc88f+9KAla1GOq8u9EuSe3p7GzeUcfVl7v//+u8fP3dLS0unHBuPr378/y+PHj2fZ2/0j4fxbFhMTw/J1113Hst1uZ3nXrl0sy3X5XS5X4BrXTbiiBwAAMDF09AAAACaGjh4AAMDETDlGryX3LZ45cybLcqzxiiuuYPmqq65iWc7/lescy3sC5MfXSk1NZXnv3r0sy32O5br9b7zxBsubN2/u9HNB5JHjoHLMPjY2lmV5/4ncz94Tuce2HIOX61XEx8ezLGs9lPuLQ+CdeOKJLF9wwQUs9+vXj+UzzzyTZXk/0i+//MKyt70RtOs4yLofMGAAy3/7299YnjBhAst5eXksy4+3aNEilj/77DOWPfUBoYIregAAABNDRw8AAGBiPnX0JSUllJOTQ4mJiZSamkrjxo2j+vp69pxDhw6R3W6nXr16UUJCAk2cOJGcTmdAGw3gK9QuGBVqF/zl0xh9VVUV2e12ysnJoSNHjtCDDz5Io0aNoq1bt9IJJ5xARET33nsvrVy5ksrLy8lqtVJRURFNmDCBPv/886B8Ad60tray3NDQwPK//vUvluX4y8UXX8zy0KFDWX7++edZfvTRR1nWruEsxyXlOJUcg5fjlnL9cHm/AMboO2fE2vWXHBvcs2cPy/L+kp07d7L8xx9/dPlzbdmyhWW5vrf8WPL+AIzJd84otav9eyX/7h5//PEe31feGyXJr0PePyLn4Q8cOFA9lnuIyLXpzzrrLJYzMzNZlrVdWlrKsrxXSg9j8pJPHf2qVatYLisro9TUVKqtraURI0aQy+Wi1157jRYuXKh2kKWlpXT66afT+vXrO3SSRH/dtKO9ccftdnfn6wDwCLULRoXaBX/5NUZ/dMWflJQUIiKqra2lw4cPU0FBgfqcQYMGkc1mo+rq6mN+jJKSErJarepD/jcFEAyoXTAq1C74qtsdfXt7O02fPp2GDRumbkHpcDgoNja2w3aqaWlp5HA4jvlxiouLyeVyqQ/50jpAoKF2wahQu9Ad3Z5Hb7fbqa6ujtatW+dXAywWC1ksFr8+hi/k/vFy32M5jpmTk8Oy/GWSe8bfddddLJ922mnqsfxeyefKsSO53rNcE6C2tpbAd0atXX999913LP/8888sy3FVXwwfPpzlU045hWX5fZK/Z9A1eq5d7X0W8n4P2V45Ji/vV5o6darHLOfV19XVsTxy5Ej1WK4nIfcnkfdC/frrryw/99xzLK9YsYJluf6JHnXrir6oqIhWrFhBq1evZj+g9PR0am1t7bBZhtPp7HDjGEA4oHbBqFC70F0+dfSKolBRUREtXbqUKisrKSsri53Pzs6mmJgYqqioUN9WX19PO3bsoPz8/MC0GKAbULtgVKhd8JdPL93b7XZauHAhLV++nBITE9XxH6vVSvHx8WS1Wun222+nGTNmUEpKCiUlJdFdd91F+fn5x7zzEyBUULtgVKhd8FeU4sMEVjmWcVRpaSndeuutRPTXwg0zZ86kRYsWUUtLCxUWFtL8+fO7/BKS2+0mq9Xa1SYFXXZ2Nstz585l2dt/zNrxGzkPXo7BJyUlefxYco7+U089xfKBAwc8vn+wuVwur19DuERi7frq6Jzso+RceK2jd3wf9V//9V8syz2833rrLZaffvpplsN9MxhqN7C1K+e5X3311Sw/88wzLGdkZLAsx/jlviA9evRgWY67a6cLyvH7TZs2sSxvWNSufUJE9NNPP7HsbZ39UOtK7fp0Rd+V/wni4uJo3rx5NG/ePF8+NEBQoXbBqFC74C+sdQ8AAGBi6OgBAABMzPT70ftLzlUfNmwYyyUlJSw/8MADLGvHveS4lRwDlXt6y3nP27ZtYzncY/JgbCeddBLLxcXFLGvH1eU458MPP8zyqFGjWJa1+t5777H822+/+dZYMBS5Z8jixYtZ/vTTT1keN24cy/K+BO3a9UQd10OR8/C1c+G1sxGIqMOGQHJNBznvXo9r1/sKV/QAAAAmho4eAADAxHyaXhcKRp+iNGbMGJZvuOEG9VhO45ArWclpGx999BHLcn9p+VJ/uOl5ilIoGK125dSrmTNnsnzGGWeoxzExMeycnFYqp8fJ5Z2//vprln3ZAjcUULv6rl05NVlueyuHQbU785ldV2oXV/QAAAAmho4eAADAxNDRAwAAmBim1wWYXD5RO41EThmRWS6JK2+fkOcB/CHvGfnggw9YPvHEE9XjzMxMdu7tt99medGiRSxXVVWxLKdbAfji0KFDHjN4hit6AAAAE0NHDwAAYGLo6AEAAEwMY/RBJrdPBNCr1atXs7x582b12GazsXNyfF9mANAPXNEDAACYGDp6AAAAE0NHDwAAYGIYoweAY9KuR6+3tekBoOtwRQ8AAGBi6OgBAABMTHcdvc52zQUfRPrPLtK/fiOL9J9dpH/9RtaVn53uOvrm5uZwNwG6KdJ/dpH+9RtZpP/sIv3rN7Ku/OyiFJ39K9fe3k6NjY2kKArZbDZqaGigpKSkcDfLMNxuN2VmZob0+6YoCjU3N1NGRgZFR+vuf8eQQe36B7UbPqhd/+i9dnV31310dDT169eP3G43ERElJSWh4Loh1N83q9Uass+lV6jdwEDthh5qNzD0WruR+y8sAABABEBHDwAAYGK67egtFgs98sgjZLFYwt0UQ8H3LfzwM+gefN/CDz+D7tH79013N+MBAABA4Oj2ih4AAAD8h44eAADAxNDRAwAAmBg6egAAABNDRw8AAGBiuu3o582bRwMGDKC4uDjKy8ujjRs3hrtJulFSUkI5OTmUmJhIqampNG7cOKqvr2fPOXToENntdurVqxclJCTQxIkTyel0hqnFkQW12znUrr6hdjtn6NpVdGjx4sVKbGyssmDBAmXLli3KnXfeqSQnJytOpzPcTdOFwsJCpbS0VKmrq1M2bdqkjB07VrHZbMq+ffvU50yZMkXJzMxUKioqlJqaGmXo0KHK+eefH8ZWRwbUrmeoXf1C7Xpm5NrVZUefm5ur2O12Nbe1tSkZGRlKSUlJGFulX7t27VKISKmqqlIURVGampqUmJgYpby8XH3Otm3bFCJSqqurw9XMiIDa9Q1qVz9Qu74xUu3q7qX71tZWqq2tpYKCAvVt0dHRVFBQQNXV1WFsmX65XC4iIkpJSSEiotraWjp8+DD7Hg4aNIhsNhu+h0GE2vUdalcfULu+M1Lt6q6j37NnD7W1tVFaWhp7e1paGjkcjjC1Sr/a29tp+vTpNGzYMBo8eDARETkcDoqNjaXk5GT2XHwPgwu16xvUrn6gdn1jtNrV3Ta14Bu73U51dXW0bt26cDcFwCeoXTAqo9Wu7q7oe/fuTT169Ohwp6LT6aT09PQwtUqfioqKaMWKFbR69Wrq16+f+vb09HRqbW2lpqYm9nx8D4MLtdt1qF19Qe12nRFrV3cdfWxsLGVnZ1NFRYX6tvb2dqqoqKD8/Pwwtkw/FEWhoqIiWrp0KVVWVlJWVhY7n52dTTExMex7WF9fTzt27MD3MIhQu96hdvUJteudoWs3rLcCdmLx4sWKxWJRysrKlK1btyqTJ09WkpOTFYfDEe6m6cLUqVMVq9WqrFmzRtm5c6f6OHDggPqcKVOmKDabTamsrFRqamqU/Px8JT8/P4ytjgyoXc9Qu/qF2vXMyLWry45eURTlhRdeUGw2mxIbG6vk5uYq69evD3eTdIOIjvkoLS1Vn3Pw4EFl2rRpSs+ePZXjjz9eGT9+vLJz587wNTqCoHY7h9rVN9Ru54xcu9iPHgAAwMR0N0YPAAAAgYOOHgAAwMTQ0QMAAJgYOnoAAAATQ0cPAABgYujoAQAATAwdPQAAgImhowcAADAxdPQAAAAmho4eAADAxNDRAwAAmNj/B3y4Z8utuVQDAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<Figure size 640x480 with 9 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# 这是个图像生成迭代器，是可以无限生成各种新图片，我们指定每轮迭代只生成9张图片\n",
    "for batch_data, batch_label in datagen.flow(train_data, train_label, batch_size=9):\n",
    "    for i in range(0, 9):\n",
    "        # 创建一个 3*3的九宫格，以显示图片\n",
    "        plt.subplot(330 + 1 + i)\n",
    "        plt.imshow(batch_data[i].reshape(28, 28), cmap=plt.get_cmap('gray'))\n",
    "    plt.show()\n",
    "    break"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.10"
  },
  "toc": {
   "base_numbering": 1,
   "nav_menu": {},
   "number_sections": true,
   "sideBar": true,
   "skip_h1_title": false,
   "title_cell": "Table of Contents",
   "title_sidebar": "Contents",
   "toc_cell": false,
   "toc_position": {},
   "toc_section_display": true,
   "toc_window_display": false
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}