Day022_Captcha_HW.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 『本次練習內容』\n",
    "#### 搭建一個能識別驗證碼的模型(CNN+CTC)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 『本次練習目的』\n",
    "  #### 了解如何結合CNN與CTC Loss\n",
    "  #### 有時間學員也可以延伸閱讀CNN+RNN+CTC模型\n",
    "  #### 沒有GPU的學員可以使用Colab，可以參考問答區的文章"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "## 使用Colab可先run這裡的code\n",
    "#假如要連結自己的雲端硬碟再開啟\n",
    "##from google.colab import drive\n",
    "##drive.mount('/content/gdrive')\n",
    "#檢查GPU\n",
    "##!nvidia-smi\n",
    "##安裝產生驗證碼的套件"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 記得先 pip install captcha"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 136
    },
    "colab_type": "code",
    "id": "AxGi02yvx1Go",
    "outputId": "e7bf7eb2-6699-441b-d724-6b5635ffa2bc"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Collecting captcha\n",
      "  Downloading https://files.pythonhosted.org/packages/90/fe/d4ddf1e6576073b5eaea76e9b2afa022c626212a30c871968480be3ccb7b/captcha-0.3-py3-none-any.whl (101kB)\n",
      "Requirement already satisfied: Pillow in c:\\programdata\\anaconda3\\lib\\site-packages (from captcha) (5.3.0)\n",
      "Installing collected packages: captcha\n",
      "Successfully installed captcha-0.3\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "WARNING: You are using pip version 19.2.1, however version 19.3.1 is available.\n",
      "You should consider upgrading via the 'python -m pip install --upgrade pip' command.\n"
     ]
    }
   ],
   "source": [
    "!pip install captcha"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 產生驗證碼"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 254
    },
    "colab_type": "code",
    "id": "h2VVCBKwk6Zv",
    "outputId": "72cda421-1eca-46b1-d5bd-3b3d9e47eb08"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZ\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "Text(0.5, 1.0, '4X5B')"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "image/png": {
       "height": 203,
       "width": 370
      },
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "from captcha.image import ImageCaptcha\n",
    "import matplotlib.pyplot as plt\n",
    "import numpy as np\n",
    "import random\n",
    "\n",
    "%matplotlib inline\n",
    "%config InlineBackend.figure_format = 'retina'\n",
    "\n",
    "import string\n",
    "##驗證碼包含0-10數字以及26個英文字母\n",
    "characters = string.digits + string.ascii_uppercase\n",
    "print(characters)\n",
    "\n",
    "##設定產生圖片尺寸，以及總類別，n_class之所以要加一是為了留一個位置給Blank\n",
    "width, height, n_len, n_class = 170, 80, 4, len(characters)+1\n",
    "\n",
    "##設定產生驗證碼的generator\n",
    "generator = ImageCaptcha(width=width, height=height)\n",
    "\n",
    "##我們先練習固定長度4個字的驗證碼\n",
    "random_str = ''.join([random.choice(characters) for j in range(4)])\n",
    "img = generator.generate_image(random_str)\n",
    "\n",
    "plt.imshow(img)\n",
    "plt.title(random_str)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 自己搭建一個CNN模型使得output feature Map 高度被壓縮成小於4，8<寬度<20\n",
    "#### 解答的output feature Map 為 (batch_size,1,13,32)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 136
    },
    "colab_type": "code",
    "id": "KX3Jq1sClGGn",
    "outputId": "ea40671a-fc3b-4758-8d29-4692252e840d"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(?, 4, 9, 256)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\ProgramData\\Anaconda3\\lib\\site-packages\\ipykernel_launcher.py:36: UserWarning: Update your `Model` call to the Keras 2 API: `Model(inputs=Tensor(\"in..., outputs=Tensor(\"de...)`\n",
      "C:\\ProgramData\\Anaconda3\\lib\\site-packages\\ipykernel_launcher.py:45: UserWarning: Update your `Model` call to the Keras 2 API: `Model(inputs=[<tf.Tenso..., outputs=[<tf.Tenso...)`\n"
     ]
    }
   ],
   "source": [
    "from keras.models import *\n",
    "from keras.layers import *\n",
    "rnn_size = 128\n",
    "\n",
    "input_tensor = Input((height,width, 3))\n",
    "x = input_tensor\n",
    "\n",
    "###自己設計CNN層######\n",
    "x = Conv2D(64, (3, 3), activation='relu', padding='same', name='block1_conv1')(x)\n",
    "x = Conv2D(64, (3, 3), activation='relu', padding='same', name='block1_conv2')(x)\n",
    "x = MaxPooling2D((2, 2), strides=(2, 2), name='block1_pool')(x)\n",
    "\n",
    "# Block 2\n",
    "x = Conv2D(128, (3, 3), activation='relu', padding='same', name='block2_conv1')(x)\n",
    "x = Conv2D(128, (3, 3), activation='relu', padding='same', name='block2_conv2')(x)\n",
    "x = MaxPooling2D((2, 2), strides=(2, 2), name='block2_pool')(x)\n",
    "\n",
    "# Block 3\n",
    "x = Conv2D(256, (3, 3), activation='relu', padding='same', name='block3_conv1')(x)\n",
    "x = Conv2D(256, (3, 3), activation='relu', padding='same', name='block3_conv2')(x)\n",
    "x = Conv2D(256, (3, 3), activation='relu', padding='same', name='block3_conv3')(x)\n",
    "x = MaxPooling2D((2, 2), strides=(5, 5), name='block3_pool')(x)\n",
    "\n",
    "##記錄輸出CNN尺寸，loss部分需要這個資訊\n",
    "##conv_shape=(Batch_size,輸出高度,輸出寬度,輸出深度)\n",
    "conv_shape = x.get_shape()\n",
    "print(conv_shape)\n",
    "\n",
    "##從(Batch_size,輸出高度,輸出寬度,輸出深度)變成(Batch_size,輸出寬度,輸出深度*輸出高度)，以符合ctc loss需求\n",
    "x = Reshape(target_shape=(int(conv_shape[2]), int(conv_shape[1]*conv_shape[3])))(x)\n",
    "x = Dense(1000, activation='relu')(x)\n",
    "x = Dropout(0.25)(x)\n",
    "x = Dense(n_class, activation='softmax')(x)\n",
    "\n",
    "##包裝用來預測的model\n",
    "base_model = Model(input=input_tensor, output=x)\n",
    "\n",
    "##設定要給CTC Loss的資訊\n",
    "labels = Input(name='the_labels', shape=[n_len], dtype='float32')\n",
    "input_length = Input(name='input_length', shape=[1], dtype='int64')\n",
    "label_length = Input(name='label_length', shape=[1], dtype='int64')\n",
    "loss_out = Lambda(ctc_lambda_func, output_shape=(1,), name='ctc')([x, labels, input_length, label_length])\n",
    "\n",
    "##這裡的model是用來計算loss\n",
    "model = Model(input=[input_tensor, labels, input_length, label_length], output=[loss_out])\n",
    "\n",
    "##之所以要lambda y_true, y_pred: y_pred是因為我們的loss已經包在網路裡，會output:y_true, y_pred，而我們只需要y_pred\n",
    "model.compile(loss={'ctc': lambda y_true, y_pred: y_pred}, optimizer='SGD')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 1000
    },
    "colab_type": "code",
    "id": "m1YRv0TVN_FL",
    "outputId": "c5b995c5-43b0-481b-8cd1-0fd5f85e4f03"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "input_13 (InputLayer)           (None, 80, 170, 3)   0                                            \n",
      "__________________________________________________________________________________________________\n",
      "block1_conv1 (Conv2D)           (None, 80, 170, 64)  1792        input_13[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "block1_conv2 (Conv2D)           (None, 80, 170, 64)  36928       block1_conv1[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "block1_pool (MaxPooling2D)      (None, 40, 85, 64)   0           block1_conv2[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "block2_conv1 (Conv2D)           (None, 40, 85, 128)  73856       block1_pool[0][0]                \n",
      "__________________________________________________________________________________________________\n",
      "block2_conv2 (Conv2D)           (None, 40, 85, 128)  147584      block2_conv1[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "block2_pool (MaxPooling2D)      (None, 20, 42, 128)  0           block2_conv2[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "block3_conv1 (Conv2D)           (None, 20, 42, 256)  295168      block2_pool[0][0]                \n",
      "__________________________________________________________________________________________________\n",
      "block3_conv2 (Conv2D)           (None, 20, 42, 256)  590080      block3_conv1[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "block3_conv3 (Conv2D)           (None, 20, 42, 256)  590080      block3_conv2[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "block3_pool (MaxPooling2D)      (None, 4, 9, 256)    0           block3_conv3[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "reshape_12 (Reshape)            (None, 9, 1024)      0           block3_pool[0][0]                \n",
      "__________________________________________________________________________________________________\n",
      "dense_23 (Dense)                (None, 9, 1000)      1025000     reshape_12[0][0]                 \n",
      "__________________________________________________________________________________________________\n",
      "dropout_12 (Dropout)            (None, 9, 1000)      0           dense_23[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "dense_24 (Dense)                (None, 9, 37)        37037       dropout_12[0][0]                 \n",
      "__________________________________________________________________________________________________\n",
      "the_labels (InputLayer)         (None, 4)            0                                            \n",
      "__________________________________________________________________________________________________\n",
      "input_length (InputLayer)       (None, 1)            0                                            \n",
      "__________________________________________________________________________________________________\n",
      "label_length (InputLayer)       (None, 1)            0                                            \n",
      "__________________________________________________________________________________________________\n",
      "ctc (Lambda)                    (None, 1)            0           dense_24[0][0]                   \n",
      "                                                                 the_labels[0][0]                 \n",
      "                                                                 input_length[0][0]               \n",
      "                                                                 label_length[0][0]               \n",
      "==================================================================================================\n",
      "Total params: 2,797,525\n",
      "Trainable params: 2,797,525\n",
      "Non-trainable params: 0\n",
      "__________________________________________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "from keras import backend as K\n",
    "import keras\n",
    "\n",
    "##CTC Loss需要四個資訊，分別是\n",
    "## Label\n",
    "## 預測\n",
    "## CNN OUTPUT寬度\n",
    "## 預測影像所包含文字長度\n",
    "def ctc_lambda_func(args):\n",
    "    y_pred, labels, input_length, label_length = args\n",
    "    return K.ctc_batch_cost(labels, y_pred, input_length, label_length)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "5Jzb1YnqlJug"
   },
   "outputs": [],
   "source": [
    "##設計generator產生training data\n",
    "## 產生包含要給loss的資訊\n",
    "## X=輸入影像\n",
    "## np.ones(batch_size)*int(conv_shape[2])=CNN輸出feature Map寬度\n",
    "## np.ones(batch_size)*n_len=字串長度(可浮動)\n",
    "\n",
    "def gen(batch_size=128):\n",
    "    X = np.zeros((batch_size,height, width, 3), dtype=np.uint8)\n",
    "    y = np.zeros((batch_size, n_len), dtype=np.uint8)\n",
    "    while True:\n",
    "        generator = ImageCaptcha(width=width, height=height)\n",
    "        for i in range(batch_size):\n",
    "            random_str = ''.join([random.choice(characters) for j in range(4)])\n",
    "            X[i] = np.array(generator.generate_image(random_str))\n",
    "            y[i] = [characters.find(x) for x in random_str]\n",
    "        yield [X, y,np.ones(batch_size)*int(conv_shape[2]), np.ones(batch_size)*n_len], np.ones(batch_size)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 225
    },
    "colab_type": "code",
    "id": "dfcFs6InVQaT",
    "outputId": "3f186d49-2e30-4efd-87c0-0670a58bc4c1"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Label:  [[20 34 11 21]]\n",
      "CNN輸出寬度:  [21.]\n",
      "字串長度(可浮動):  [4.]\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "image/png": {
       "height": 191,
       "width": 370
      },
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "next_ge=gen(batch_size=1)\n",
    "test_ge=next(next_ge)\n",
    "plt.imshow(test_ge[0][0][0])\n",
    "print('Label: ',test_ge[0][1])\n",
    "print('CNN輸出寬度: ',test_ge[0][2])\n",
    "print('字串長度(可浮動): ',test_ge[0][3])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 開始訓練"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 1000
    },
    "colab_type": "code",
    "id": "fqdKnqollUKj",
    "outputId": "3da34151-a9df-4287-ead6-50c414d30e65"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/60\n",
      "  5/300 [..............................] - ETA: 40:27 - loss: 63.7801"
     ]
    },
    {
     "ename": "KeyboardInterrupt",
     "evalue": "",
     "output_type": "error",
     "traceback": [
      "\u001b[1;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[1;31mKeyboardInterrupt\u001b[0m                         Traceback (most recent call last)",
      "\u001b[1;32m<ipython-input-22-c2156d63d20d>\u001b[0m in \u001b[0;36m<module>\u001b[1;34m\u001b[0m\n\u001b[1;32m----> 1\u001b[1;33m \u001b[0mmodel\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0mfit_generator\u001b[0m\u001b[1;33m(\u001b[0m\u001b[0mgen\u001b[0m\u001b[1;33m(\u001b[0m\u001b[1;36m32\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0msteps_per_epoch\u001b[0m\u001b[1;33m=\u001b[0m\u001b[1;36m300\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0mepochs\u001b[0m\u001b[1;33m=\u001b[0m\u001b[1;36m60\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0m",
      "\u001b[1;32mC:\\ProgramData\\Anaconda3\\lib\\site-packages\\keras\\legacy\\interfaces.py\u001b[0m in \u001b[0;36mwrapper\u001b[1;34m(*args, **kwargs)\u001b[0m\n\u001b[0;32m     89\u001b[0m                 warnings.warn('Update your `' + object_name + '` call to the ' +\n\u001b[0;32m     90\u001b[0m                               'Keras 2 API: ' + signature, stacklevel=2)\n\u001b[1;32m---> 91\u001b[1;33m             \u001b[1;32mreturn\u001b[0m \u001b[0mfunc\u001b[0m\u001b[1;33m(\u001b[0m\u001b[1;33m*\u001b[0m\u001b[0margs\u001b[0m\u001b[1;33m,\u001b[0m \u001b[1;33m**\u001b[0m\u001b[0mkwargs\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0m\u001b[0;32m     92\u001b[0m         \u001b[0mwrapper\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0m_original_function\u001b[0m \u001b[1;33m=\u001b[0m \u001b[0mfunc\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m     93\u001b[0m         \u001b[1;32mreturn\u001b[0m \u001b[0mwrapper\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n",
      "\u001b[1;32mC:\\ProgramData\\Anaconda3\\lib\\site-packages\\keras\\engine\\training.py\u001b[0m in \u001b[0;36mfit_generator\u001b[1;34m(self, generator, steps_per_epoch, epochs, verbose, callbacks, validation_data, validation_steps, class_weight, max_queue_size, workers, use_multiprocessing, shuffle, initial_epoch)\u001b[0m\n\u001b[0;32m   1416\u001b[0m             \u001b[0muse_multiprocessing\u001b[0m\u001b[1;33m=\u001b[0m\u001b[0muse_multiprocessing\u001b[0m\u001b[1;33m,\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m   1417\u001b[0m             \u001b[0mshuffle\u001b[0m\u001b[1;33m=\u001b[0m\u001b[0mshuffle\u001b[0m\u001b[1;33m,\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[1;32m-> 1418\u001b[1;33m             initial_epoch=initial_epoch)\n\u001b[0m\u001b[0;32m   1419\u001b[0m \u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m   1420\u001b[0m     \u001b[1;33m@\u001b[0m\u001b[0minterfaces\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0mlegacy_generator_methods_support\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n",
      "\u001b[1;32mC:\\ProgramData\\Anaconda3\\lib\\site-packages\\keras\\engine\\training_generator.py\u001b[0m in \u001b[0;36mfit_generator\u001b[1;34m(model, generator, steps_per_epoch, epochs, verbose, callbacks, validation_data, validation_steps, class_weight, max_queue_size, workers, use_multiprocessing, shuffle, initial_epoch)\u001b[0m\n\u001b[0;32m    215\u001b[0m                 outs = model.train_on_batch(x, y,\n\u001b[0;32m    216\u001b[0m                                             \u001b[0msample_weight\u001b[0m\u001b[1;33m=\u001b[0m\u001b[0msample_weight\u001b[0m\u001b[1;33m,\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[1;32m--> 217\u001b[1;33m                                             class_weight=class_weight)\n\u001b[0m\u001b[0;32m    218\u001b[0m \u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m    219\u001b[0m                 \u001b[0mouts\u001b[0m \u001b[1;33m=\u001b[0m \u001b[0mto_list\u001b[0m\u001b[1;33m(\u001b[0m\u001b[0mouts\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n",
      "\u001b[1;32mC:\\ProgramData\\Anaconda3\\lib\\site-packages\\keras\\engine\\training.py\u001b[0m in \u001b[0;36mtrain_on_batch\u001b[1;34m(self, x, y, sample_weight, class_weight)\u001b[0m\n\u001b[0;32m   1215\u001b[0m             \u001b[0mins\u001b[0m \u001b[1;33m=\u001b[0m \u001b[0mx\u001b[0m \u001b[1;33m+\u001b[0m \u001b[0my\u001b[0m \u001b[1;33m+\u001b[0m \u001b[0msample_weights\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m   1216\u001b[0m         \u001b[0mself\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0m_make_train_function\u001b[0m\u001b[1;33m(\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[1;32m-> 1217\u001b[1;33m         \u001b[0moutputs\u001b[0m \u001b[1;33m=\u001b[0m \u001b[0mself\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0mtrain_function\u001b[0m\u001b[1;33m(\u001b[0m\u001b[0mins\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0m\u001b[0;32m   1218\u001b[0m         \u001b[1;32mreturn\u001b[0m \u001b[0munpack_singleton\u001b[0m\u001b[1;33m(\u001b[0m\u001b[0moutputs\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m   1219\u001b[0m \u001b[1;33m\u001b[0m\u001b[0m\n",
      "\u001b[1;32mC:\\ProgramData\\Anaconda3\\lib\\site-packages\\keras\\backend\\tensorflow_backend.py\u001b[0m in \u001b[0;36m__call__\u001b[1;34m(self, inputs)\u001b[0m\n\u001b[0;32m   2713\u001b[0m                 \u001b[1;32mreturn\u001b[0m \u001b[0mself\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0m_legacy_call\u001b[0m\u001b[1;33m(\u001b[0m\u001b[0minputs\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m   2714\u001b[0m \u001b[1;33m\u001b[0m\u001b[0m\n\u001b[1;32m-> 2715\u001b[1;33m             \u001b[1;32mreturn\u001b[0m \u001b[0mself\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0m_call\u001b[0m\u001b[1;33m(\u001b[0m\u001b[0minputs\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0m\u001b[0;32m   2716\u001b[0m         \u001b[1;32melse\u001b[0m\u001b[1;33m:\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m   2717\u001b[0m             \u001b[1;32mif\u001b[0m \u001b[0mpy_any\u001b[0m\u001b[1;33m(\u001b[0m\u001b[0mis_tensor\u001b[0m\u001b[1;33m(\u001b[0m\u001b[0mx\u001b[0m\u001b[1;33m)\u001b[0m \u001b[1;32mfor\u001b[0m \u001b[0mx\u001b[0m \u001b[1;32min\u001b[0m \u001b[0minputs\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m:\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n",
      "\u001b[1;32mC:\\ProgramData\\Anaconda3\\lib\\site-packages\\keras\\backend\\tensorflow_backend.py\u001b[0m in \u001b[0;36m_call\u001b[1;34m(self, inputs)\u001b[0m\n\u001b[0;32m   2673\u001b[0m             \u001b[0mfetched\u001b[0m \u001b[1;33m=\u001b[0m \u001b[0mself\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0m_callable_fn\u001b[0m\u001b[1;33m(\u001b[0m\u001b[1;33m*\u001b[0m\u001b[0marray_vals\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0mrun_metadata\u001b[0m\u001b[1;33m=\u001b[0m\u001b[0mself\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0mrun_metadata\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m   2674\u001b[0m         \u001b[1;32melse\u001b[0m\u001b[1;33m:\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[1;32m-> 2675\u001b[1;33m             \u001b[0mfetched\u001b[0m \u001b[1;33m=\u001b[0m \u001b[0mself\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0m_callable_fn\u001b[0m\u001b[1;33m(\u001b[0m\u001b[1;33m*\u001b[0m\u001b[0marray_vals\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0m\u001b[0;32m   2676\u001b[0m         \u001b[1;32mreturn\u001b[0m \u001b[0mfetched\u001b[0m\u001b[1;33m[\u001b[0m\u001b[1;33m:\u001b[0m\u001b[0mlen\u001b[0m\u001b[1;33m(\u001b[0m\u001b[0mself\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0moutputs\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m]\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m   2677\u001b[0m \u001b[1;33m\u001b[0m\u001b[0m\n",
      "\u001b[1;32mC:\\ProgramData\\Anaconda3\\lib\\site-packages\\tensorflow\\python\\client\\session.py\u001b[0m in \u001b[0;36m__call__\u001b[1;34m(self, *args, **kwargs)\u001b[0m\n\u001b[0;32m   1437\u001b[0m           ret = tf_session.TF_SessionRunCallable(\n\u001b[0;32m   1438\u001b[0m               \u001b[0mself\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0m_session\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0m_session\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0mself\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0m_handle\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0margs\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0mstatus\u001b[0m\u001b[1;33m,\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[1;32m-> 1439\u001b[1;33m               run_metadata_ptr)\n\u001b[0m\u001b[0;32m   1440\u001b[0m         \u001b[1;32mif\u001b[0m \u001b[0mrun_metadata\u001b[0m\u001b[1;33m:\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m   1441\u001b[0m           \u001b[0mproto_data\u001b[0m \u001b[1;33m=\u001b[0m \u001b[0mtf_session\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0mTF_GetBuffer\u001b[0m\u001b[1;33m(\u001b[0m\u001b[0mrun_metadata_ptr\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n",
      "\u001b[1;31mKeyboardInterrupt\u001b[0m: "
     ]
    }
   ],
   "source": [
    "model.fit_generator(gen(32), steps_per_epoch=300, epochs=60)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 預測"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "characters2 = characters + ' '\n",
    "[X_test, y_test, _, _], _  = next(gen(1))\n",
    "y_pred = base_model.predict(X_test)\n",
    "##用ctc_decode得到解答，自己寫可以參考下方\n",
    "out = K.get_value(K.ctc_decode(y_pred, input_length=np.ones(y_pred.shape[0])*y_pred.shape[1], )[0][0])\n",
    "out = ''.join([characters[x] for x in out[0]])\n",
    "y_true = ''.join([characters[x] for x in y_test[0]])\n",
    "\n",
    "plt.imshow(X_test[0])\n",
    "plt.title('pred:' + str(out) + '\\ntrue: ' + str(y_true))\n",
    "\n",
    "argmax = np.argmax(y_pred, axis=2)[0]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 自己寫 decode CTC"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "metadata": {},
   "outputs": [],
   "source": [
    "## 其中0代表預測為空格，如果預測相同字符之間沒有空格要移除\n",
    "word=''\n",
    "n=''\n",
    "for single_result in y_pred[0].argmax(1):\n",
    "    if single_result != 36:\n",
    "        if n != single_result:\n",
    "            word+=characters[single_result]\n",
    "    n=single_result "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 額外參考：加入RNN的神經網路"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Users/chening/anaconda3/lib/python3.6/site-packages/ipykernel_launcher.py:8: UserWarning: Update your `Conv2D` call to the Keras 2 API: `Conv2D(32, (3, 3), activation=\"relu\")`\n",
      "  \n",
      "/Users/chening/anaconda3/lib/python3.6/site-packages/ipykernel_launcher.py:10: UserWarning: Update your `Conv2D` call to the Keras 2 API: `Conv2D(32, (3, 3), activation=\"relu\")`\n",
      "  # Remove the CWD from sys.path while we load stuff.\n",
      "/Users/chening/anaconda3/lib/python3.6/site-packages/ipykernel_launcher.py:23: UserWarning: Update your `GRU` call to the Keras 2 API: `GRU(128, return_sequences=True, name=\"gru1\", kernel_initializer=\"he_normal\")`\n",
      "/Users/chening/anaconda3/lib/python3.6/site-packages/ipykernel_launcher.py:25: UserWarning: Update your `GRU` call to the Keras 2 API: `GRU(128, return_sequences=True, go_backwards=True, name=\"gru1_b\", kernel_initializer=\"he_normal\")`\n",
      "/Users/chening/anaconda3/lib/python3.6/site-packages/ipykernel_launcher.py:28: UserWarning: Update your `GRU` call to the Keras 2 API: `GRU(128, return_sequences=True, name=\"gru2\", kernel_initializer=\"he_normal\")`\n",
      "/Users/chening/anaconda3/lib/python3.6/site-packages/ipykernel_launcher.py:30: UserWarning: Update your `GRU` call to the Keras 2 API: `GRU(128, return_sequences=True, go_backwards=True, name=\"gru2_b\", kernel_initializer=\"he_normal\")`\n",
      "/Users/chening/anaconda3/lib/python3.6/site-packages/ipykernel_launcher.py:34: UserWarning: Update your `Dense` call to the Keras 2 API: `Dense(37, activation=\"softmax\", kernel_initializer=\"he_normal\")`\n",
      "/Users/chening/anaconda3/lib/python3.6/site-packages/ipykernel_launcher.py:36: UserWarning: Update your `Model` call to the Keras 2 API: `Model(inputs=Tensor(\"in..., outputs=Tensor(\"de...)`\n",
      "/Users/chening/anaconda3/lib/python3.6/site-packages/ipykernel_launcher.py:44: UserWarning: Update your `Model` call to the Keras 2 API: `Model(inputs=[<tf.Tenso..., outputs=[<tf.Tenso...)`\n"
     ]
    }
   ],
   "source": [
    "from keras.models import *\n",
    "from keras.layers import *\n",
    "rnn_size = 128\n",
    "\n",
    "input_tensor = Input((height,width, 3))\n",
    "x = input_tensor\n",
    "for i in range(4):\n",
    "    x = Convolution2D(32, 3, 3, activation='relu')(x)\n",
    "    x = keras.layers.BatchNormalization(axis=-1)(x)\n",
    "    x = Convolution2D(32, 3, 3, activation='relu')(x)\n",
    "    x = keras.layers.BatchNormalization(axis=-1)(x)\n",
    "    if i <3:\n",
    "        x = MaxPooling2D(pool_size=(2, 2))(x)\n",
    "    else:\n",
    "        x = MaxPooling2D(pool_size=(2, 1))(x)\n",
    "\n",
    "\n",
    "conv_shape = x.get_shape()\n",
    "x = Reshape(target_shape=(int(conv_shape[2]), int(conv_shape[1]*conv_shape[3])))(x)\n",
    "\n",
    "x = Dense(32, activation='relu')(x)\n",
    "\n",
    "gru_1 = GRU(rnn_size, return_sequences=True, init='he_normal', name='gru1')(x)\n",
    "gru_1b = GRU(rnn_size, return_sequences=True, go_backwards=True, \n",
    "             init='he_normal', name='gru1_b')(x)\n",
    "gru1_merged = keras.layers.add([gru_1, gru_1b])\n",
    "\n",
    "gru_2 = GRU(rnn_size, return_sequences=True, init='he_normal', name='gru2')(gru1_merged)\n",
    "gru_2b = GRU(rnn_size, return_sequences=True, go_backwards=True, \n",
    "             init='he_normal', name='gru2_b')(gru1_merged)\n",
    "x = keras.layers.Concatenate()([gru_2, gru_2b])\n",
    "\n",
    "x = Dropout(0.25)(x)\n",
    "x = Dense(n_class, init='he_normal', activation='softmax')(x)\n",
    "\n",
    "base_model = Model(input=input_tensor, output=x)\n",
    "\n",
    "labels = Input(name='the_labels', shape=[n_len], dtype='float32')\n",
    "input_length = Input(name='input_length', shape=[1], dtype='int64')\n",
    "label_length = Input(name='label_length', shape=[1], dtype='int64')\n",
    "loss_out = Lambda(ctc_lambda_func, output_shape=(1,), \n",
    "                  name='ctc')([x, labels, input_length, label_length])\n",
    "\n",
    "model = Model(input=[input_tensor, labels, input_length, label_length], output=[loss_out])\n",
    "model.compile(loss={'ctc': lambda y_true, y_pred: y_pred}, optimizer='Adam')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "input_4 (InputLayer)            (None, 80, 170, 3)   0                                            \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_25 (Conv2D)              (None, 78, 168, 32)  896         input_4[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "batch_normalization_25 (BatchNo (None, 78, 168, 32)  128         conv2d_25[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_26 (Conv2D)              (None, 76, 166, 32)  9248        batch_normalization_25[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "batch_normalization_26 (BatchNo (None, 76, 166, 32)  128         conv2d_26[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "max_pooling2d_13 (MaxPooling2D) (None, 38, 83, 32)   0           batch_normalization_26[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_27 (Conv2D)              (None, 36, 81, 32)   9248        max_pooling2d_13[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "batch_normalization_27 (BatchNo (None, 36, 81, 32)   128         conv2d_27[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_28 (Conv2D)              (None, 34, 79, 32)   9248        batch_normalization_27[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "batch_normalization_28 (BatchNo (None, 34, 79, 32)   128         conv2d_28[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "max_pooling2d_14 (MaxPooling2D) (None, 17, 39, 32)   0           batch_normalization_28[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_29 (Conv2D)              (None, 15, 37, 32)   9248        max_pooling2d_14[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "batch_normalization_29 (BatchNo (None, 15, 37, 32)   128         conv2d_29[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_30 (Conv2D)              (None, 13, 35, 32)   9248        batch_normalization_29[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "batch_normalization_30 (BatchNo (None, 13, 35, 32)   128         conv2d_30[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "max_pooling2d_15 (MaxPooling2D) (None, 6, 17, 32)    0           batch_normalization_30[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_31 (Conv2D)              (None, 4, 15, 32)    9248        max_pooling2d_15[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "batch_normalization_31 (BatchNo (None, 4, 15, 32)    128         conv2d_31[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_32 (Conv2D)              (None, 2, 13, 32)    9248        batch_normalization_31[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "batch_normalization_32 (BatchNo (None, 2, 13, 32)    128         conv2d_32[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "max_pooling2d_16 (MaxPooling2D) (None, 1, 13, 32)    0           batch_normalization_32[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "reshape_4 (Reshape)             (None, 13, 32)       0           max_pooling2d_16[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "dense_7 (Dense)                 (None, 13, 32)       1056        reshape_4[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "gru1 (GRU)                      (None, 13, 128)      61824       dense_7[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "gru1_b (GRU)                    (None, 13, 128)      61824       dense_7[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "add_2 (Add)                     (None, 13, 128)      0           gru1[0][0]                       \n",
      "                                                                 gru1_b[0][0]                     \n",
      "__________________________________________________________________________________________________\n",
      "gru2 (GRU)                      (None, 13, 128)      98688       add_2[0][0]                      \n",
      "__________________________________________________________________________________________________\n",
      "gru2_b (GRU)                    (None, 13, 128)      98688       add_2[0][0]                      \n",
      "__________________________________________________________________________________________________\n",
      "concatenate_2 (Concatenate)     (None, 13, 256)      0           gru2[0][0]                       \n",
      "                                                                 gru2_b[0][0]                     \n",
      "__________________________________________________________________________________________________\n",
      "dropout_4 (Dropout)             (None, 13, 256)      0           concatenate_2[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "dense_8 (Dense)                 (None, 13, 37)       9509        dropout_4[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "the_labels (InputLayer)         (None, 4)            0                                            \n",
      "__________________________________________________________________________________________________\n",
      "input_length (InputLayer)       (None, 1)            0                                            \n",
      "__________________________________________________________________________________________________\n",
      "label_length (InputLayer)       (None, 1)            0                                            \n",
      "__________________________________________________________________________________________________\n",
      "ctc (Lambda)                    (None, 1)            0           dense_8[0][0]                    \n",
      "                                                                 the_labels[0][0]                 \n",
      "                                                                 input_length[0][0]               \n",
      "                                                                 label_length[0][0]               \n",
      "==================================================================================================\n",
      "Total params: 398,245\n",
      "Trainable params: 397,733\n",
      "Non-trainable params: 512\n",
      "__________________________________________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "accelerator": "GPU",
  "colab": {
   "collapsed_sections": [],
   "name": "Captcha.ipynb",
   "provenance": []
  },
  "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": "3.7.1"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}