{
  "cells": [
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "x4HI2mpwlrcn"
      },
      "outputs": [],
      "source": [
        "##### Copyright 2019 The TensorFlow Authors."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "cellView": "form",
        "id": "679Lmwt3l1Bk",
        "vscode": {
          "languageId": "python"
        }
      },
      "outputs": [],
      "source": [
        "#@title Licensed under the Apache License, Version 2.0 (the \"License\");\n",
        "# you may not use this file except in compliance with the License.\n",
        "# You may obtain a copy of the License at\n",
        "#\n",
        "# https://www.apache.org/licenses/LICENSE-2.0\n",
        "#\n",
        "# Unless required by applicable law or agreed to in writing, software\n",
        "# distributed under the License is distributed on an \"AS IS\" BASIS,\n",
        "# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n",
        "# See the License for the specific language governing permissions and\n",
        "# limitations under the License."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "DSPCom-KmApV"
      },
      "source": [
        "# 卷积神经网络（Convolutional Neural Network, CNN）"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "klAltGp8ycek"
      },
      "source": [
        "<table class=\"tfo-notebook-buttons\" align=\"left\">\n",
        "  <td>     <a target=\"_blank\" href=\"https://tensorflow.google.cn/tutorials/images/cnn\"><img src=\"https://tensorflow.google.cn/images/tf_logo_32px.png\">在 TensorFlow.org 上查看</a> </td>\n",
        "  <td>     <a target=\"_blank\" href=\"https://colab.research.google.com/github/tensorflow/docs-l10n/blob/master/site/zh-cn/tutorials/images/cnn.ipynb\"><img src=\"https://tensorflow.google.cn/images/colab_logo_32px.png\">在 Google Colab 中运行</a>   </td>\n",
        "  <td>     <a target=\"_blank\" href=\"https://github.com/tensorflow/docs-l10n/blob/master/site/zh-cn/tutorials/images/cnn.ipynb\"><img src=\"https://tensorflow.google.cn/images/GitHub-Mark-32px.png\">在 GitHub 上查看源代码</a>   </td>\n",
        "  <td>     <a href=\"https://storage.googleapis.com/tensorflow_docs/docs-l10n/site/zh-cn/tutorials/images/cnn.ipynb\"><img src=\"https://tensorflow.google.cn/images/download_logo_32px.png\">下载笔记本</a> </td>\n",
        "</table>"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "qLGkt5qiyz4E"
      },
      "source": [
        "本教程展示了如何训练一个简单的[卷积神经网络](https://developers.google.com/machine-learning/glossary/#convolutional_neural_network) (CNN) 来对 [CIFAR 图像](https://www.cs.toronto.edu/~kriz/cifar.html)进行分类。由于本教程使用的是 [Keras Sequential API](https://tensorflow.google.cn/guide/keras/overview)，创建和训练模型只需要几行代码。\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "m7KBpffWzlxH"
      },
      "source": [
        "### 导入 TensorFlow"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "iAve6DCL4JH4",
        "vscode": {
          "languageId": "python"
        }
      },
      "outputs": [],
      "source": [
        "import tensorflow as tf\n",
        "\n",
        "from tensorflow.keras import datasets, layers, models\n",
        "import matplotlib.pyplot as plt"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "jRFxccghyMVo"
      },
      "source": [
        "### 下载并准备 CIFAR10 数据集\n",
        "\n",
        "CIFAR10 数据集包含 10 类，共 60000 张彩色图片，每类图片有 6000 张。此数据集中 50000 个样例被作为训练集，剩余 10000 个样例作为测试集。类之间相互独立，不存在重叠的部分。"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "JWoEqyMuXFF4",
        "vscode": {
          "languageId": "python"
        }
      },
      "outputs": [],
      "source": [
        "(train_images, train_labels), (test_images, test_labels) = datasets.cifar10.load_data()\n",
        "\n",
        "# Normalize pixel values to be between 0 and 1\n",
        "train_images, test_images = train_images / 255.0, test_images / 255.0"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "7wArwCTJJlUa"
      },
      "source": [
        "### 验证数据\n",
        "\n",
        "为了验证数据集看起来是否正确，我们绘制训练集中的前 25 张图像并在每张图像下方显示类名称：\n"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "K3PAELE2eSU9",
        "vscode": {
          "languageId": "python"
        }
      },
      "outputs": [],
      "source": [
        "class_names = ['airplane', 'automobile', 'bird', 'cat', 'deer',\n",
        "               'dog', 'frog', 'horse', 'ship', 'truck']\n",
        "\n",
        "plt.figure(figsize=(10,10))\n",
        "for i in range(25):\n",
        "    plt.subplot(5,5,i+1)\n",
        "    plt.xticks([])\n",
        "    plt.yticks([])\n",
        "    plt.grid(False)\n",
        "    plt.imshow(train_images[i])\n",
        "    # The CIFAR labels happen to be arrays, \n",
        "    # which is why you need the extra index\n",
        "    plt.xlabel(class_names[train_labels[i][0]])\n",
        "plt.show()"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Oewp-wYg31t9"
      },
      "source": [
        "### 构造卷积神经网络模型"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "3hQvqXpNyN3x"
      },
      "source": [
        "下方展示的 6 行代码声明了了一个常见卷积神经网络，由几个 [Conv2D](https://tensorflow.google.cn/api_docs/python/tf/keras/layers/Conv2D) 和 [MaxPooling2D](https://tensorflow.google.cn/api_docs/python/tf/keras/layers/MaxPool2D) 层组成。\n",
        "\n",
        "CNN 将形状为 (image_height, image_width, color_channels) 的张量作为输入，忽略批次大小。如果您不熟悉这些维度，color_channels 是指 (R,G,B)。在此示例中，您将配置 CNN 以处理形状为 (32, 32, 3) 的输入，即 CIFAR 图像的格式。您可以通过将参数 `input_shape` 传递给第一层来实现此目的。\n"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "L9YmGQBQPrdn",
        "vscode": {
          "languageId": "python"
        }
      },
      "outputs": [],
      "source": [
        "model = models.Sequential()\n",
        "model.add(layers.Conv2D(32, (3, 3), activation='relu', input_shape=(32, 32, 3)))\n",
        "model.add(layers.MaxPooling2D((2, 2)))\n",
        "model.add(layers.Conv2D(64, (3, 3), activation='relu'))\n",
        "model.add(layers.MaxPooling2D((2, 2)))\n",
        "model.add(layers.Conv2D(64, (3, 3), activation='relu'))"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "lvDVFkg-2DPm"
      },
      "source": [
        "到目前为止，模型的架构如下："
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "8-C4XBg4UTJy",
        "vscode": {
          "languageId": "python"
        }
      },
      "outputs": [],
      "source": [
        "model.summary()"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "_j-AXYeZ2GO5"
      },
      "source": [
        "在上面的结构中，您可以看到每个 Conv2D 和 MaxPooling2D 层的输出都是一个三维的张量 (Tensor)，其形状描述了 (height, width, channels)。越深的层中，宽度和高度都会收缩。每个 Conv2D 层输出的通道数量 (channels) 取决于声明层时的第一个参数（如：上面代码中的 32 或 64）。这样，由于宽度和高度的收缩，您便可以（从运算的角度）增加每个 Conv2D 层输出的通道数量 (channels)。"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "_v8sVOtG37bT"
      },
      "source": [
        "### 增加 Dense 层\n",
        "\n",
        "为了完成模型，您需要将卷积基（形状为 (4, 4, 64)）的最后一个输出张量馈送到一个或多个 Dense 层以执行分类。Dense 层将向量作为输入（即 1 维），而当前输出为 3 维张量。首先，将 3 维输出展平（或展开）为 1 维，然后在顶部添加一个或多个 Dense 层。CIFAR 有 10 个输出类，因此使用具有 10 个输出的最终 Dense 层。"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "mRs95d6LUVEi",
        "vscode": {
          "languageId": "python"
        }
      },
      "outputs": [],
      "source": [
        "model.add(layers.Flatten())\n",
        "model.add(layers.Dense(64, activation='relu'))\n",
        "model.add(layers.Dense(10))"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ipGiQMcR4Gtq"
      },
      "source": [
        "下面是模型的完整架构："
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "8Yu_m-TZUWGX",
        "vscode": {
          "languageId": "python"
        }
      },
      "outputs": [],
      "source": [
        "model.summary()"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "xNKXi-Gy3RO-"
      },
      "source": [
        "网络摘要显示 (4, 4, 64) 输出在经过两个 Dense 层之前被展平为形状为 (1024) 的向量。"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "P3odqfHP4M67"
      },
      "source": [
        "### 编译并训练模型"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "MdDzI75PUXrG",
        "vscode": {
          "languageId": "python"
        }
      },
      "outputs": [],
      "source": [
        "model.compile(optimizer='adam',\n",
        "              loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),\n",
        "              metrics=['accuracy'])\n",
        "\n",
        "history = model.fit(train_images, train_labels, epochs=10, \n",
        "                    validation_data=(test_images, test_labels))"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "jKgyC5K_4O0d"
      },
      "source": [
        "### 评估模型"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "gtyDF0MKUcM7",
        "vscode": {
          "languageId": "python"
        }
      },
      "outputs": [],
      "source": [
        "plt.plot(history.history['accuracy'], label='accuracy')\n",
        "plt.plot(history.history['val_accuracy'], label = 'val_accuracy')\n",
        "plt.xlabel('Epoch')\n",
        "plt.ylabel('Accuracy')\n",
        "plt.ylim([0.5, 1])\n",
        "plt.legend(loc='lower right')\n",
        "\n",
        "test_loss, test_acc = model.evaluate(test_images,  test_labels, verbose=2)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "0LvwaKhtUdOo",
        "vscode": {
          "languageId": "python"
        }
      },
      "outputs": [],
      "source": [
        "print(test_acc)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "8cfJ8AR03gT5"
      },
      "source": [
        "您的简单 CNN 的测试准确率已达到 70% 以上。对于只有几行的代码来说，效果不错！对于另一种 CNN 风格，请参阅[适合专家的 TensorFlow 2 快速入门](https://tensorflow.google.cn/tutorials/quickstart/advanced)示例，此示例使用了 Keras 子类化 API 和 `tf.GradientTape`。"
      ]
    }
  ],
  "metadata": {
    "accelerator": "GPU",
    "colab": {
      "collapsed_sections": [],
      "name": "cnn.ipynb",
      "toc_visible": true
    },
    "kernelspec": {
      "display_name": "Python 3",
      "name": "python3"
    }
  },
  "nbformat": 4,
  "nbformat_minor": 0
}