简单的卷积神经网络搭建(cifar10案例)

参考视频: https://www.bilibili.com/video/BV16A41157LW?p=34

获取cifar10数据(使用本地数据)

from tensorflow.python.keras.datasets.cifar import load_batch

def get_load_cifar10():
    # 获取数据  直接使用    tf.keras.datasets.cifar10.load_data() 会报错(160M数据,可能是因为网络不好)
    # cifar10数据下载 参考   https://zhuanlan.zhihu.com/p/129078357
    # path 为解压后的路径
    path = 'D:\\Repository\\ai_data\\DeepLearing_TensorFlow2.0-book\\cifar10\\cifar-10-batches-py\\'
    num_train_samples = 50000
    x_train = np.empty((num_train_samples, 3, 32, 32), dtype='uint8')
    y_train = np.empty((num_train_samples,), dtype='uint8')
    for i in range(1, 6):
        fpath = os.path.join(path, 'data_batch_' + str(i))
        (x_train[(i - 1) * 10000:i * 10000, :, :, :],
         y_train[(i - 1) * 10000:i * 10000]) = load_batch(fpath)
    fpath = os.path.join(path, 'test_batch')
    x_test, y_test = load_batch(fpath)
    y_train = np.reshape(y_train, (len(y_train), 1))
    y_test = np.reshape(y_test, (len(y_test), 1))
    if tf.keras.backend.image_data_format() == 'channels_last':
        x_train = x_train.transpose(0, 2, 3, 1)
        x_test = x_test.transpose(0, 2, 3, 1)
    x_test = x_test.astype(x_train.dtype)
    y_test = y_test.astype(y_train.dtype)
    x_train, x_test = x_train / 255.0, x_test / 255.0  # 数据归一化
    return (x_train, y_train), (x_test, y_test)

模型搭建

class MyCifarModel(tf.keras.Model):
    '''
卷积神经网络搭建示例
搭建一层卷积 两层全连接的网络
使用6个 5*5的卷积核 过2*2 的池化核 池化步长2(5*5 conv,filters=6 2*2 pool strides=2)
过128个神经元的全连接层
最后过一个10个神经元的全连接层(因为cifar是一个10分类)
搭建神经网络的口诀是 CBAPD
    C -- 卷积: 核 6*5*5  步长: 1 填充:same(使用全零填充)
    B -- 批标准化:  yes 使用标准化
    A -- 激活函数: relu
    P -- 池化: max(最大池化) 核: 2*2 池化步长: 2 填充 same
    D -- 舍弃: Dropout(0.2)

    Flatten: 拉直层 把卷积送过来的数据拉直
    Dense:  Dense(神经元 128,激活:relu, Dropout: 0.2)
    Dense(神经元:10,激活: softmax)

'''

    def __init__(self):
        super(MyCifarModel, self).__init__()
        self.c1 = tf.keras.layers.Conv2D(filters=6, kernel_size=(5, 5), padding='same')  # 卷积层
        self.b1 = tf.keras.layers.BatchNormalization()  # BN层
        self.a1 = tf.keras.layers.Activation('relu')  # 激活层
        self.p1 = tf.keras.layers.MaxPool2D(pool_size=(2, 2), strides=2, padding='same')  # 池化层
        self.d1 = tf.keras.layers.Dropout(0.2)  # dropout层

        self.flatten = tf.keras.layers.Flatten()
        self.f1 = tf.keras.layers.Dense(128, activation='relu')
        self.d2 = tf.keras.layers.Dropout(0.2)
        self.f2 = tf.keras.layers.Dense(10, activation='softmax')  # softmax 使输出符合概率分布

    def call(self, x):
        # 使输入到输出 过一遍前向传播 返回推理结果
        x = self.c1(x)
        x = self.b1(x)
        x = self.a1(x)
        x = self.p1(x)
        x = self.d1(x)

        x = self.flatten(x)
        x = self.f1(x)
        x = self.d2(x)
        y = self.f2(x)
        return y

读取模型(实现断点续练):

def local_load_model(model_path):
    if os.path.exists(model_path + '/saved_model.pb'):
        tf.print('-------------load the model-----------------')
        local_model = tf.keras.models.load_model(model_path)
    else:
        local_model = MyCifarModel()
        local_model.compile(optimizer='adam',
                            loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=False),
                            metrics=['sparse_categorical_accuracy'])
    return local_model

使用

if __name__ == '__main__':
    (x_train, y_train), (x_test, y_test) = get_load_cifar10()
    model_path = "./data/model/breakpoint"
    my_cifar_model = local_load_model(model_path)
    # 电脑配置低 只训练3次
    history = my_cifar_model.fit(x_train, y_train, batch_size=32, epochs=3, validation_data=(x_test, y_test),
                                 validation_freq=1)
    my_cifar_model.summary()
    # 保存模型
    my_cifar_model.save(model_path, save_format="tf")

完整代码如下

# -*- coding: utf-8 -*-
import tensorflow as tf
import os
import numpy as np
from matplotlib import pyplot as plt
from tensorflow.python.keras.datasets.cifar import load_batch

def get_load_cifar10():
    # 获取数据  直接使用    tf.keras.datasets.cifar10.load_data() 会报错
    # cifar10数据下载 参考   https://zhuanlan.zhihu.com/p/129078357
    # path 为解压后的路径
    path = 'D:\\Repository\\ai_data\\DeepLearing_TensorFlow2.0-book\\cifar10\\cifar-10-batches-py\\'
    num_train_samples = 50000
    x_train = np.empty((num_train_samples, 3, 32, 32), dtype='uint8')
    y_train = np.empty((num_train_samples,), dtype='uint8')
    for i in range(1, 6):
        fpath = os.path.join(path, 'data_batch_' + str(i))
        (x_train[(i - 1) * 10000:i * 10000, :, :, :],
         y_train[(i - 1) * 10000:i * 10000]) = load_batch(fpath)
    fpath = os.path.join(path, 'test_batch')
    x_test, y_test = load_batch(fpath)
    y_train = np.reshape(y_train, (len(y_train), 1))
    y_test = np.reshape(y_test, (len(y_test), 1))
    if tf.keras.backend.image_data_format() == 'channels_last':
        x_train = x_train.transpose(0, 2, 3, 1)
        x_test = x_test.transpose(0, 2, 3, 1)
    x_test = x_test.astype(x_train.dtype)
    y_test = y_test.astype(y_train.dtype)
    x_train, x_test = x_train / 255.0, x_test / 255.0  # 数据归一化
    return (x_train, y_train), (x_test, y_test)


'''
参考视频:  https://www.bilibili.com/video/BV16A41157LW?p=34
'''


class MyCifarModel(tf.keras.Model):
    '''
卷积神经网络搭建示例
搭建一层卷积 两层全连接的网络
使用6个 5*5的卷积核 过2*2 的池化核 池化步长2(5*5 conv,filters=6 2*2 pool strides=2)
过128个神经元的全连接层
最后过一个10个神经元的全连接层(因为cifar是一个10分类)
搭建神经网络的口诀是 CBAPD
    C -- 卷积: 核 6*5*5  步长: 1 填充:same(使用全零填充)
    B -- 批标准化:  yes 使用标准化
    A -- 激活函数: relu
    P -- 池化: max(最大池化) 核: 2*2 池化步长: 2 填充 same
    D -- 舍弃: Dropout(0.2)

    Flatten: 拉直层 把卷积送过来的数据拉直
    Dense:  Dense(神经元 128,激活:relu, Dropout: 0.2)
    Dense(神经元:10,激活: softmax)

'''

    def __init__(self):
        super(MyCifarModel, self).__init__()
        self.c1 = tf.keras.layers.Conv2D(filters=6, kernel_size=(5, 5), padding='same')  # 卷积层
        self.b1 = tf.keras.layers.BatchNormalization()  # BN层
        self.a1 = tf.keras.layers.Activation('relu')  # 激活层
        self.p1 = tf.keras.layers.MaxPool2D(pool_size=(2, 2), strides=2, padding='same')  # 池化层
        self.d1 = tf.keras.layers.Dropout(0.2)  # dropout层

        self.flatten = tf.keras.layers.Flatten()
        self.f1 = tf.keras.layers.Dense(128, activation='relu')
        self.d2 = tf.keras.layers.Dropout(0.2)
        self.f2 = tf.keras.layers.Dense(10, activation='softmax')  # softmax 使输出符合概率分布

    def call(self, x):
        # 使输入到输出 过一遍前向传播 返回推理结果
        x = self.c1(x)
        x = self.b1(x)
        x = self.a1(x)
        x = self.p1(x)
        x = self.d1(x)

        x = self.flatten(x)
        x = self.f1(x)
        x = self.d2(x)
        y = self.f2(x)
        return y


def local_load_model(model_path):
    if os.path.exists(model_path + '/saved_model.pb'):
        tf.print('-------------load the model-----------------')
        local_model = tf.keras.models.load_model(model_path)
    else:
        local_model = MyCifarModel()
        local_model.compile(optimizer='adam',
                            loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=False),
                            metrics=['sparse_categorical_accuracy'])
    return local_model


def show_train_line(history):
    # 显示训练集和验证集的acc和loss曲线
    acc = history.history['sparse_categorical_accuracy']
    val_acc = history.history['val_sparse_categorical_accuracy']
    loss = history.history['loss']
    val_loss = history.history['val_loss']

    plt.subplot(1, 2, 1)
    plt.plot(acc, label='Training Accuracy')
    plt.plot(val_acc, label='Validation Accuracy')
    plt.title('Training and Validation Accuracy')
    plt.legend()

    plt.subplot(1, 2, 2)
    plt.plot(loss, label='Training Loss')
    plt.plot(val_loss, label='Validation Loss')
    plt.title('Training and Validation Loss')
    plt.legend()
    plt.show()


if __name__ == '__main__':
    (x_train, y_train), (x_test, y_test) = get_load_cifar10()
    model_path = "./data/model/breakpoint"
    my_cifar_model = local_load_model(model_path)
    history = my_cifar_model.fit(x_train, y_train, batch_size=32, epochs=5, validation_data=(x_test, y_test),
                                 validation_freq=1)
    show_train_line(history)
    my_cifar_model.summary()
    # 保存模型
    my_cifar_model.save(model_path, save_format="tf")