验证码识别-TFLearn 版-单字符-简化-训练.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 验证码识别-TFLearn 版-单字符"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "同是从带多个字符的图片中学习特征，同时对应多个输出分类(4个字符)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 准备数据"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "%matplotlib inline\n",
    "import os\n",
    "import cv2\n",
    "import numpy as np\n",
    "from matplotlib import pyplot as plt\n",
    "import matplotlib.image as mpimg\n",
    "\n",
    "image_dir = \"/Volumes/Cloud/DataSet/src/captcha\"\n",
    "model_dir = \"/Volumes/Cloud/Models/captcha/resnet\"\n",
    "\n",
    "LABEL_LENGTH = 4        # 验证码字符数\n",
    "#LABELS = \"0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZ\"\n",
    "LABELS = \"0123456789\"   # 验证码字符组成\n",
    "\n",
    "IMAGE_H = 28            # 缩放后单个字符图片大小\n",
    "IMAGE_W = 28            # 缩放后单个字符图片大小\n",
    "IMAGE_C = 3             # 图片通道数"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# 读图\n",
    "def read_image(filename):\n",
    "    image = cv2.imread(filename)  # 读 PNG 只会读到3个通道\n",
    "    #image = mpimg.imread(filename) \n",
    "    h,w = image.shape[:2]\n",
    "    image = image[0:h,12:w-6]\n",
    "    image = cv2.resize(image,(IMAGE_W*LABEL_LENGTH,IMAGE_H),cv2.INTER_LINEAR) # 缩放大小\n",
    "    return image\n",
    "\n",
    "filename = os.path.join(image_dir,os.listdir(image_dir)[9])\n",
    "\n",
    "image = read_image(filename)\n",
    "\n",
    "# 看等比例切分是否合适\n",
    "h = image.shape[0]\n",
    "sw = IMAGE_W\n",
    "for i in range(LABEL_LENGTH-1):\n",
    "    x = sw*(i+1)\n",
    "    image[0:h,x:x+1]=0\n",
    "plt.imshow(image);"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 图片前期处理"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 576x216 with 5 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "def plot_image(image):\n",
    "    plt.axis('off');\n",
    "    plt.imshow(image)\n",
    "\n",
    "# 切分图片\n",
    "def split_image(image):\n",
    "    images = []\n",
    "    h = image.shape[0]\n",
    "    sw = IMAGE_W\n",
    "    for i in range(LABEL_LENGTH):\n",
    "        x = sw*i\n",
    "        images.append(image[0:h,x:x+sw])\n",
    "\n",
    "    return images\n",
    "\n",
    "image = read_image(os.path.join(image_dir,os.listdir(image_dir)[0]))\n",
    "simgs = split_image(image)\n",
    "\n",
    "plt.figure(figsize=(8,3))\n",
    "plt.subplot(1, 5, 1)\n",
    "plot_image(image)\n",
    "\n",
    "for i in range(len(simgs)):\n",
    "    plt.subplot(1, 5, i+2)\n",
    "    plot_image(simgs[i])\n",
    "\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 576x216 with 8 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# 对比度增强\n",
    "def enhance_contrast(image,contrast=1.5): \n",
    "    rows,cols,channels=image.shape\n",
    "    dst=image.copy()\n",
    "\n",
    "    a=contrast\n",
    "    b=100\n",
    "    for i in range(rows):\n",
    "        for j in range(cols):\n",
    "            for c in range(3):\n",
    "                color=image[i,j][c]*a+b\n",
    "                if color>255:\n",
    "                    dst[i,j][c]=255\n",
    "                elif color<0:\n",
    "                    dst[i,j][c]=0\n",
    "    return dst\n",
    "\n",
    "# 返回去燥用的掩码\n",
    "def get_noise_mask_hsv(image):\n",
    "    # 转换为 HSV 后只进行去燥掩码计算\n",
    "    #H是色彩，范围0° ~ 360°，红（0°）、绿（120°）、蓝（240°）；\n",
    "    #S是深浅， S = 0时，只有灰度，越往圆心的位置越偏白；\n",
    "    #V是明暗\n",
    "\n",
    "    hsv = cv2.cvtColor(image, cv2.COLOR_BGR2HSV).reshape(-1,3)\n",
    "\n",
    "    def is_keep_color(x,v,t=60):\n",
    "        return v-t<=x and x<=v+t\n",
    "\n",
    "    h = []\n",
    "    s = []\n",
    "    v = []\n",
    "    for x in hsv:\n",
    "        if not(x[0]==0 and x[1]==0 and x[2]==255):\n",
    "            h.append(x[0])\n",
    "            s.append(x[1])\n",
    "            v.append(x[2])\n",
    "\n",
    "    h = np.argmax(np.bincount(np.array(h,dtype=np.uint8)))  \n",
    "    s = np.argmax(np.bincount(np.array(s,dtype=np.uint8)))  \n",
    "    v = np.argmax(np.bincount(np.array(v,dtype=np.uint8)))  \n",
    "\n",
    "    mask = []\n",
    "    for x in hsv:\n",
    "        #if is_keep_color(x[0], h, 20) and is_keep_color(x[1], s) and is_keep_color(x[2], v):\n",
    "        if is_keep_color(x[0], h, 20):\n",
    "            mask.append(255)\n",
    "        else:\n",
    "            mask.append(0)\n",
    "\n",
    "    mask = np.array(mask).reshape((image.shape[:2])) # 掩码\n",
    "    return mask\n",
    "\n",
    "\n",
    "# 用HSV聚类进行去燥\n",
    "def remove_noise_hsv(image):\n",
    "    h,w,_ = image.shape\n",
    "    sw = int(w/LABEL_LENGTH)\n",
    "    for i in range(LABEL_LENGTH):\n",
    "        x = sw*i\n",
    "        mask = image.copy()\n",
    "        mask[0:h,0:x]    = 255\n",
    "        mask[0:h,x+sw:w] = 255\n",
    "\n",
    "        mask = get_noise_mask_hsv(mask)\n",
    "\n",
    "        mask[0:h,0:x]    = 255\n",
    "        mask[0:h,x+sw:w] = 255\n",
    "\n",
    "        image[mask==0]=255  # 进行掩码计算\n",
    "\n",
    "    return image\n",
    "\n",
    "# 验证码去燥\n",
    "def remove_noise_gray(image):\n",
    "    gray = cv2.cvtColor(image,cv2.COLOR_BGR2GRAY)\n",
    "    gray = cv2.GaussianBlur(gray, (5, 5), 0)\n",
    "\n",
    "    # 自适应阈值\n",
    "    gray = cv2.adaptiveThreshold(gray,255,cv2.ADAPTIVE_THRESH_MEAN_C, cv2.THRESH_BINARY,11,2) \n",
    "\n",
    "    image = image.copy()\n",
    "    image[gray!=0]=255\n",
    "\n",
    "    return image\n",
    "\n",
    "from scipy import ndimage\n",
    "\n",
    "# 验证码去燥\n",
    "def remove_noise(image):\n",
    "    image = enhance_contrast(image)\n",
    "    #image = remove_noise_hsv(image)\n",
    "    image = remove_noise_gray(image)\n",
    "    gray = cv2.cvtColor(image,cv2.COLOR_BGR2GRAY)\n",
    "    gray[gray!=255]=0\n",
    "\n",
    "    gray = ndimage.binary_erosion(gray,iterations=5)\n",
    "    #gray = ndimage.binary_dilation(gray,iterations=1)\n",
    "    #gray = ndimage.binary_erosion(gray,iterations=2)\n",
    "\n",
    "    image[gray!=0]=255\n",
    "\n",
    "    gray = cv2.cvtColor(image,cv2.COLOR_BGR2GRAY)\n",
    "    gray[gray!=255]=0\n",
    "\n",
    "    gray = ndimage.binary_erosion(gray,iterations=2)\n",
    "    gray = ndimage.binary_dilation(gray,iterations=2)\n",
    "\n",
    "    # 清理一下边界\n",
    "    gray[0:4,:] = 255\n",
    "    gray[gray.shape[0] - 4:, :] = 255\n",
    "    gray[:,0:4] = 255\n",
    "    gray[:,gray.shape[1] - 4:] = 255\n",
    "\n",
    "    return gray\n",
    "\n",
    "# 验证码去燥\n",
    "def remove_noise(image):\n",
    "    #image = enhance_contrast(image)\n",
    "    #image = remove_noise_gray(image)\n",
    "    #image = cv2.cvtColor(image,cv2.COLOR_BGR2GRAY)\n",
    "    #image[image!=255]=0\n",
    "    #global IMAGE_C\n",
    "    #IMAGE_C = 1\n",
    "\n",
    "    return image\n",
    "\n",
    "filename=os.path.join(image_dir,os.listdir(image_dir)[0])\n",
    "image = read_image(os.path.join(image_dir,os.listdir(image_dir)[9]))\n",
    "simgs = split_image(image)\n",
    "\n",
    "plt.figure(figsize=(8,3))\n",
    "\n",
    "for i in range(len(simgs)):\n",
    "    plt.subplot(2, LABEL_LENGTH, i+1)\n",
    "    plt.axis('off');\n",
    "    plt.imshow(simgs[i])\n",
    "    plt.subplot(2, LABEL_LENGTH, i+1+LABEL_LENGTH)\n",
    "    plot_image(remove_noise(simgs[i]))\n",
    "\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 构建训练集和测试集"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 999/999 [00:00<00:00, 1803.38it/s]\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "((3996, 28, 28, 3), (3996, 10), (3996,))"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import os\n",
    "import cv2\n",
    "import numpy as np\n",
    "from sklearn import model_selection\n",
    "import tqdm\n",
    "\n",
    "# 读取图片、二值化、灰度图，Label 编码为 OneHot 格式\n",
    "def load_data(path):\n",
    "    # OneHot\n",
    "    def char_to_vec(c):\n",
    "        y = np.zeros((len(LABELS),))\n",
    "        y[LABELS.index(c)] = 1.0\n",
    "        return y\n",
    "\n",
    "    labels = []\n",
    "    images = []\n",
    "    sfiles = []\n",
    "    fnames = os.listdir(path)\n",
    "    with tqdm.tqdm(total=len(fnames)) as pbar:\n",
    "        for i,name in enumerate(fnames):\n",
    "            if name.endswith(\".jpg\") or name.endswith(\".jpeg\") or name.endswith(\".png\"):\n",
    "                image = read_image(os.path.join(path,name))\n",
    "                # Convert from [0, 255] -> [0.0, 1.0].\n",
    "                image = image.astype(np.float32)\n",
    "                image = image/255.0\n",
    "\n",
    "                simgs = split_image(image)\n",
    "                label = name[:LABEL_LENGTH].upper()\n",
    "\n",
    "                for k in range(LABEL_LENGTH):\n",
    "                    labels.append(char_to_vec(label[k]))\n",
    "                    images.append(remove_noise(simgs[k]))\n",
    "                    sfiles.append(name)\n",
    "                pbar.update(1)\n",
    "\n",
    "    images = np.array(images)\n",
    "    if IMAGE_C == 1:\n",
    "        images = images.reshape(images.shape + (1,))\n",
    "    labels = np.array(labels)\n",
    "    labels = labels.reshape((labels.shape[0],-1))\n",
    "    sfiles = np.array(sfiles)\n",
    "\n",
    "    return images, labels, sfiles\n",
    "\n",
    "# \n",
    "class DataSet(object):\n",
    "    def __init__(self, images, labels):\n",
    "        assert images.shape[0] == labels.shape[0], (\n",
    "            \"images.shape: %s labels.shape: %s\" % (images.shape,\n",
    "                                                   labels.shape))\n",
    "        self._num_examples = images.shape[0]\n",
    "        self._images = images\n",
    "        self._labels = labels\n",
    "        self._epochs_completed = 0\n",
    "        self._index_in_epoch = 0\n",
    "\n",
    "    @property\n",
    "    def images(self):\n",
    "        return self._images\n",
    "\n",
    "    @property\n",
    "    def labels(self):\n",
    "        return self._labels\n",
    "\n",
    "    @property\n",
    "    def num_examples(self):\n",
    "        return self._num_examples\n",
    "\n",
    "    @property\n",
    "    def epochs_completed(self):\n",
    "        return self._epochs_completed\n",
    "\n",
    "    def next_batch(self, batch_size):\n",
    "        \"\"\"Return the next `batch_size` examples from this data set.\"\"\"\n",
    "        start = self._index_in_epoch\n",
    "        self._index_in_epoch += batch_size\n",
    "        if self._index_in_epoch > self._num_examples:\n",
    "            # Finished epoch\n",
    "            self._epochs_completed += 1\n",
    "            # Shuffle the data\n",
    "            perm = np.arange(self._num_examples)\n",
    "            np.random.shuffle(perm)\n",
    "            self._images = self._images[perm]\n",
    "            self._labels = self._labels[perm]\n",
    "            # Start next epoch\n",
    "            start = 0\n",
    "            self._index_in_epoch = batch_size\n",
    "            assert batch_size <= self._num_examples\n",
    "        end = self._index_in_epoch\n",
    "        return self._images[start:end], self._labels[start:end]\n",
    "class DataSets(object):\n",
    "    pass\n",
    "\n",
    "images,labels,sfiles = load_data(image_dir)\n",
    "images.shape, labels.shape, sfiles.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(array([1., 0., 0., 0., 0., 0., 0., 0., 0., 0.]), '0')"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Onehot 编码转换回字符串\n",
    "def onehot2number(label):\n",
    "    return LABELS[np.argmax(label)]\n",
    "\n",
    "labels[0],onehot2number(labels[0])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 数据分布"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 576x216 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "a = []\n",
    "for i,label in enumerate(labels):\n",
    "    label = onehot2number(label)\n",
    "    a.append([i,label])\n",
    "\n",
    "import pandas as pd\n",
    "df = pd.DataFrame(a,columns=['i','label'])\n",
    "import seaborn as sns\n",
    "\n",
    "plt.figure(figsize=(8,3))\n",
    "sns.countplot(df['label']);\n",
    "plt.xlabel('数字分布');"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 进行数据平衡"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "max 443\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 576x216 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "groupby_counts = df.groupby('label').size()\n",
    "samples = groupby_counts.max().max()\n",
    "print('max', samples)\n",
    "df_groupby = df.groupby('label')\n",
    "df = df_groupby.apply(lambda x: x.sample(samples, replace=True)).reset_index(drop=True)\n",
    "\n",
    "plt.figure(figsize=(8,3))\n",
    "sns.countplot(df['label']);\n",
    "plt.xlabel('数字分布');"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "new_images=[]\n",
    "new_labels=[]\n",
    "new_sfiles=[]\n",
    "for i in df['i']:\n",
    "    new_images.append(images[i])\n",
    "    new_labels.append(labels[i])\n",
    "    new_sfiles.append(sfiles[i])\n",
    "images = np.array(new_images)\n",
    "labels = np.array(new_labels)\n",
    "sfiles = np.array(new_sfiles)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "((3544, 28, 28, 3), (3544, 10))"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "def make_data_sets(images,labels):\n",
    "    trainX, testX, trainY, testY = model_selection.train_test_split(images, labels, test_size=0.20, random_state=42)\n",
    "\n",
    "    data_sets = DataSets()\n",
    "    data_sets.train = DataSet(trainX,trainY)\n",
    "    data_sets.test  = DataSet(testX,testY)\n",
    "    \n",
    "    return data_sets\n",
    "\n",
    "data_sets = make_data_sets(images,labels)\n",
    "data_sets.train.images.shape, data_sets.train.labels.shape"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 做完的数据集可视化一下"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 576x216 with 6 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "%matplotlib inline\n",
    "from matplotlib import pyplot as plt\n",
    "import numpy as np\n",
    "\n",
    "# 显示图像\n",
    "def display(data, labels, i):\n",
    "    image = data[i]\n",
    "\n",
    "    # Convert from [0.0, 1.0] -> [0, 255] \n",
    "    image = (image*255).astype(int)\n",
    "  \n",
    "    if IMAGE_C == 1:\n",
    "        image = image.reshape(IMAGE_H,IMAGE_W)\n",
    "\n",
    "    plot_image(image)\n",
    "    plt.title(onehot2number(labels[i]))\n",
    "    plt.axis('off')\n",
    "    \n",
    "batch_images, batch_labels = data_sets.train.next_batch(6)\n",
    "\n",
    "plt.figure(figsize=(8,3))\n",
    "for i in range(len(batch_images)):\n",
    "    plt.subplot(1, 6, i+1)\n",
    "    display(batch_images, batch_labels, i)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 建模训练\n",
    "## 残差网络模型"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training Step: 947  | total loss: \u001b[1m\u001b[32m0.02454\u001b[0m\u001b[0m | time: 4.251s\n",
      "| Momentum | epoch: 009 | loss: 0.02454 - acc: 0.9987 -- iter: 1900/3544\n",
      "Successfully left training! Final model accuracy: 0.9986549019813538\n"
     ]
    },
    {
     "ename": "KeyboardInterrupt",
     "evalue": "",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m                         Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-14-77f329100172>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[1;32m     82\u001b[0m                   \u001b[0mshow_metric\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;32mTrue\u001b[0m\u001b[0;34m,\u001b[0m       \u001b[0;31m# 是否显示学习过程中的准确率\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     83\u001b[0m                   \u001b[0mcallbacks\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mfit_cb\u001b[0m\u001b[0;34m,\u001b[0m       \u001b[0;31m# 用于提前结束训练\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 84\u001b[0;31m                   run_id='vcode_resnet')\n\u001b[0m\u001b[1;32m     85\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     86\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m/usr/local/lib/python3.6/site-packages/tflearn/models/dnn.py\u001b[0m in \u001b[0;36mfit\u001b[0;34m(self, X_inputs, Y_targets, n_epoch, validation_set, show_metric, batch_size, shuffle, snapshot_epoch, snapshot_step, excl_trainops, validation_batch_size, run_id, callbacks)\u001b[0m\n\u001b[1;32m    214\u001b[0m                          \u001b[0mexcl_trainops\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mexcl_trainops\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    215\u001b[0m                          \u001b[0mrun_id\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mrun_id\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 216\u001b[0;31m                          callbacks=callbacks)\n\u001b[0m\u001b[1;32m    217\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    218\u001b[0m     \u001b[0;32mdef\u001b[0m \u001b[0mfit_batch\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mX_inputs\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mY_targets\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m/usr/local/lib/python3.6/site-packages/tflearn/helpers/trainer.py\u001b[0m in \u001b[0;36mfit\u001b[0;34m(self, feed_dicts, n_epoch, val_feed_dicts, show_metric, snapshot_step, snapshot_epoch, shuffle_all, dprep_dict, daug_dict, excl_trainops, run_id, callbacks)\u001b[0m\n\u001b[1;32m    337\u001b[0m                                                        \u001b[0;34m(\u001b[0m\u001b[0mbool\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mbest_checkpoint_path\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;34m|\u001b[0m \u001b[0msnapshot_epoch\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    338\u001b[0m                                                        \u001b[0msnapshot_step\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 339\u001b[0;31m                                                        show_metric)\n\u001b[0m\u001b[1;32m    340\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    341\u001b[0m                             \u001b[0;31m# Update training state\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m/usr/local/lib/python3.6/site-packages/tflearn/helpers/trainer.py\u001b[0m in \u001b[0;36m_train\u001b[0;34m(self, training_step, snapshot_epoch, snapshot_step, show_metric)\u001b[0m\n\u001b[1;32m    816\u001b[0m         \u001b[0mtflearn\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mis_training\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;32mTrue\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0msession\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0msession\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    817\u001b[0m         _, train_summ_str = self.session.run([self.train, self.summ_op],\n\u001b[0;32m--> 818\u001b[0;31m                                              feed_batch)\n\u001b[0m\u001b[1;32m    819\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    820\u001b[0m         \u001b[0;31m# Retrieve loss value from summary string\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m/usr/local/lib/python3.6/site-packages/tensorflow/python/client/session.py\u001b[0m in \u001b[0;36mrun\u001b[0;34m(self, fetches, feed_dict, options, run_metadata)\u001b[0m\n\u001b[1;32m    927\u001b[0m     \u001b[0;32mtry\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    928\u001b[0m       result = self._run(None, fetches, feed_dict, options_ptr,\n\u001b[0;32m--> 929\u001b[0;31m                          run_metadata_ptr)\n\u001b[0m\u001b[1;32m    930\u001b[0m       \u001b[0;32mif\u001b[0m \u001b[0mrun_metadata\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    931\u001b[0m         \u001b[0mproto_data\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mtf_session\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mTF_GetBuffer\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mrun_metadata_ptr\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m/usr/local/lib/python3.6/site-packages/tensorflow/python/client/session.py\u001b[0m in \u001b[0;36m_run\u001b[0;34m(self, handle, fetches, feed_dict, options, run_metadata)\u001b[0m\n\u001b[1;32m   1150\u001b[0m     \u001b[0;32mif\u001b[0m \u001b[0mfinal_fetches\u001b[0m \u001b[0;32mor\u001b[0m \u001b[0mfinal_targets\u001b[0m \u001b[0;32mor\u001b[0m \u001b[0;34m(\u001b[0m\u001b[0mhandle\u001b[0m \u001b[0;32mand\u001b[0m \u001b[0mfeed_dict_tensor\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1151\u001b[0m       results = self._do_run(handle, final_targets, final_fetches,\n\u001b[0;32m-> 1152\u001b[0;31m                              feed_dict_tensor, options, run_metadata)\n\u001b[0m\u001b[1;32m   1153\u001b[0m     \u001b[0;32melse\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1154\u001b[0m       \u001b[0mresults\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;34m[\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m/usr/local/lib/python3.6/site-packages/tensorflow/python/client/session.py\u001b[0m in \u001b[0;36m_do_run\u001b[0;34m(self, handle, target_list, fetch_list, feed_dict, options, run_metadata)\u001b[0m\n\u001b[1;32m   1326\u001b[0m     \u001b[0;32mif\u001b[0m \u001b[0mhandle\u001b[0m \u001b[0;32mis\u001b[0m \u001b[0;32mNone\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1327\u001b[0m       return self._do_call(_run_fn, feeds, fetches, targets, options,\n\u001b[0;32m-> 1328\u001b[0;31m                            run_metadata)\n\u001b[0m\u001b[1;32m   1329\u001b[0m     \u001b[0;32melse\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1330\u001b[0m       \u001b[0;32mreturn\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_do_call\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0m_prun_fn\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mhandle\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mfeeds\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mfetches\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m/usr/local/lib/python3.6/site-packages/tensorflow/python/client/session.py\u001b[0m in \u001b[0;36m_do_call\u001b[0;34m(self, fn, *args)\u001b[0m\n\u001b[1;32m   1332\u001b[0m   \u001b[0;32mdef\u001b[0m \u001b[0m_do_call\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mfn\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m*\u001b[0m\u001b[0margs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1333\u001b[0m     \u001b[0;32mtry\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m-> 1334\u001b[0;31m       \u001b[0;32mreturn\u001b[0m \u001b[0mfn\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m*\u001b[0m\u001b[0margs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m   1335\u001b[0m     \u001b[0;32mexcept\u001b[0m \u001b[0merrors\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mOpError\u001b[0m \u001b[0;32mas\u001b[0m \u001b[0me\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1336\u001b[0m       \u001b[0mmessage\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mcompat\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mas_text\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0me\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mmessage\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m/usr/local/lib/python3.6/site-packages/tensorflow/python/client/session.py\u001b[0m in \u001b[0;36m_run_fn\u001b[0;34m(feed_dict, fetch_list, target_list, options, run_metadata)\u001b[0m\n\u001b[1;32m   1317\u001b[0m       \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_extend_graph\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1318\u001b[0m       return self._call_tf_sessionrun(\n\u001b[0;32m-> 1319\u001b[0;31m           options, feed_dict, fetch_list, target_list, run_metadata)\n\u001b[0m\u001b[1;32m   1320\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1321\u001b[0m     \u001b[0;32mdef\u001b[0m \u001b[0m_prun_fn\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mhandle\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mfeed_dict\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mfetch_list\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m/usr/local/lib/python3.6/site-packages/tensorflow/python/client/session.py\u001b[0m in \u001b[0;36m_call_tf_sessionrun\u001b[0;34m(self, options, feed_dict, fetch_list, target_list, run_metadata)\u001b[0m\n\u001b[1;32m   1405\u001b[0m     return tf_session.TF_SessionRun_wrapper(\n\u001b[1;32m   1406\u001b[0m         \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_session\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0moptions\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mfeed_dict\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mfetch_list\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mtarget_list\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m-> 1407\u001b[0;31m         run_metadata)\n\u001b[0m\u001b[1;32m   1408\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1409\u001b[0m   \u001b[0;32mdef\u001b[0m \u001b[0m_call_tf_sessionprun\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mhandle\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mfeed_dict\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mfetch_list\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m: "
     ]
    }
   ],
   "source": [
    "# 构造网络模型\n",
    "from __future__ import absolute_import, division, print_function\n",
    "import tflearn\n",
    "import tensorflow as tf\n",
    "import time\n",
    "\n",
    "tf.reset_default_graph()\n",
    "tflearn.config.init_training_mode()\n",
    "\n",
    "# 提前结束训练，用于保存最好的模型\n",
    "class EarlyStoppingCallback(tflearn.callbacks.Callback):\n",
    "    def __init__(self, val_acc_thresh):\n",
    "        self.val_acc_thresh = val_acc_thresh\n",
    "\n",
    "    #def on_epoch_end(self, training_state):\n",
    "    def on_batch_end(self, training_state, snapshot):\n",
    "        if training_state.val_acc is None: return\n",
    "        if training_state.val_acc >= self.val_acc_thresh and training_state.acc_value >= self.val_acc_thresh:\n",
    "            raise StopIteration\n",
    "\n",
    "    def on_train_end(self, training_state):\n",
    "        self.val_acc   = training_state.val_acc\n",
    "        self.acc_value = training_state.acc_value\n",
    "        print(\"Successfully left training! Final model accuracy:\", training_state.acc_value)\n",
    "\n",
    "\n",
    "def build_model(learning_rate):\n",
    "    img_prep = tflearn.ImagePreprocessing()\n",
    "    #img_prep.add_featurewise_zero_center(per_channel=True)  # 输入图像要减图像均值\n",
    "\n",
    "    img_aug = tflearn.ImageAugmentation()\n",
    "    img_aug.add_random_rotation(max_angle=10.0)   # 随机旋转角度\n",
    "    #img_aug.add_random_blur(sigma_max=5.0)\n",
    "\n",
    "    # Building Residual Network\n",
    "    net = tflearn.input_data(shape=[None, IMAGE_H, IMAGE_W, IMAGE_C], \n",
    "                             data_preprocessing=img_prep,\n",
    "                             data_augmentation=img_aug,\n",
    "                             name='input')\n",
    "\n",
    "    net = tflearn.conv_2d(net, 16, 3,\n",
    "                          regularizer='L2',\n",
    "                          weights_init='variance_scaling',\n",
    "                          weight_decay=0.0001,\n",
    "                          name=\"conv1\")  # 卷积处理, 16个卷积，卷积核大小为3，L2 正则化减少过拟合\n",
    "\n",
    "    net = tflearn.residual_block(net, 1, 16, name=\"res1\")  # 1 个残差层，输出16特征\n",
    "    net = tflearn.residual_block(net, 1, 32, downsample=True, name=\"res2\")  # 1 个残差层，输出32特征，降维1/2\n",
    "    net = tflearn.residual_block(net, 1, 64, downsample=True, name=\"res3\")  # 1 个残差层，输出64特征，降维1/2\n",
    "\n",
    "    # Regression\n",
    "    net = tflearn.fully_connected(net, len(LABELS), activation='softmax')\n",
    "    mom = tflearn.Momentum(learning_rate, lr_decay=0.1, decay_step=32000, staircase=True)\n",
    "    net = tflearn.regression(net, optimizer=mom, loss='categorical_crossentropy')\n",
    "\n",
    "    # Training\n",
    "    model = tflearn.DNN(net, max_checkpoints=1, tensorboard_verbose=3)\n",
    "\n",
    "    return model\n",
    "\n",
    "model = build_model(learning_rate=0.01)\n",
    "\n",
    "# Load a model\n",
    "if tf.gfile.Exists(model_dir +'/model.tfl'+'.index'):\n",
    "    model.load(model_dir +'/model.tfl')\n",
    "\n",
    "# Training\n",
    "fit = True\n",
    "if fit:\n",
    "    start_time = time.time()\n",
    "    fit_cb = EarlyStoppingCallback(val_acc_thresh=0.998)\n",
    "    try:\n",
    "        model.fit(data_sets.train.images, \n",
    "                  data_sets.train.labels, \n",
    "                  validation_set=(\n",
    "                      data_sets.test.images, \n",
    "                      data_sets.test.labels\n",
    "                  ),\n",
    "                  n_epoch=100,            # 完整数据集投喂次数，太多或太少会导致过拟合或欠拟合\n",
    "                  batch_size=100,         # 每次训练获取的样本数\n",
    "                  shuffle=True,           # 是否对数据进行洗牌\n",
    "                  show_metric=True,       # 是否显示学习过程中的准确率\n",
    "                  callbacks=fit_cb,       # 用于提前结束训练\n",
    "                  run_id='vcode_resnet')\n",
    "\n",
    "\n",
    "    except StopIteration as e:\n",
    "        print(\"early stop\")\n",
    "\n",
    "    # Manually save model\n",
    "    if not tf.gfile.Exists(model_dir):\n",
    "        tf.gfile.MakeDirs(model_dir)\n",
    "\n",
    "    model.save(model_dir +'/model.tfl')\n",
    "    print('save trained model to ', model_dir +'/model.tfl')   \n",
    "    \n",
    "    duration = time.time() - start_time\n",
    "    print('Training Duration %.3f sec' % (duration))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 测试"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "def display_preds(data, y_trues, y_preds, i):\n",
    "    image = data[i]\n",
    "\n",
    "    # Convert from [0.0, 1.0] -> [0, 255] \n",
    "    image = (image*255).astype(int)\n",
    "\n",
    "    plot_image(image)\n",
    "    y_true = onehot2number(y_trues[i])\n",
    "    y_pred = onehot2number(y_preds[i])\n",
    "    plt.title(y_true+(' == ' if y_true==y_pred else ' != ')+'pred('+y_pred+')')\n",
    "    plt.axis('off')\n",
    "\n",
    "x_data, y_trues = data_sets.test.next_batch(6)\n",
    "\n",
    "y_preds = model.predict(x_data)\n",
    "\n",
    "plt.figure(figsize=(8,3))\n",
    "for i in range(len(x_data)):\n",
    "    plt.subplot(2, 3, i+1)\n",
    "    display_preds(x_data, y_trues, y_preds, i)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 剔除错误数据\n",
    "查找原始数据可能就标记错的数据，然后继续训练会进一步提高精度，最终接近100%"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "scrolled": false
   },
   "outputs": [],
   "source": [
    "# Predict\n",
    "x_data, y_trues = images,labels\n",
    "y_preds = model.predict(x_data)\n",
    "\n",
    "y_error = []\n",
    "for i in range(len(y_preds)):\n",
    "    y_true = y_trues[i].argmax()\n",
    "    y_pred = y_preds[i].argmax()\n",
    "\n",
    "    if y_pred != y_true:\n",
    "        y_error.append(i)\n",
    "        \n",
    "acc = 1-len(y_error)/len(y_preds)\n",
    "print(\"accuracy: \", acc)\n",
    "\n",
    "for i in y_error:\n",
    "    y_true = y_trues[i].argmax()\n",
    "    y_pred = y_preds[i].argmax()\n",
    "    print('incorrect', sfiles[i], y_true, ' != ', y_pred)"
   ]
  }
 ],
 "metadata": {
  "anaconda-cloud": {},
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.5"
  },
  "toc": {
   "base_numbering": 1,
   "nav_menu": {},
   "number_sections": false,
   "sideBar": true,
   "skip_h1_title": false,
   "title_cell": "Table of Contents",
   "title_sidebar": "Contents",
   "toc_cell": false,
   "toc_position": {
    "height": "calc(100% - 180px)",
    "left": "10px",
    "top": "150px",
    "width": "375px"
   },
   "toc_section_display": true,
   "toc_window_display": true
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}