用keras复现AlexNet

首先，我们用keras复现一下AlexNet的模型结构：

from keras.models import Sequential
from keras.layers import Input,Dense,Conv2D,MaxPooling2D,UpSampling2D,Dropout,Flatten
from keras.layers import Dropout

model = Sequential()
model.add(Conv2D(96,(11,11),strides=(4,4),input_shape=(227,227,3),padding='valid',activation='relu',kernel_initializer='uniform'))
model.add(MaxPooling2D(pool_size=(3,3),strides=(2,2)))
model.add(Conv2D(256,(5,5),strides=(1,1),padding='same',activation='relu',kernel_initializer='uniform'))
model.add(MaxPooling2D(pool_size=(3,3),strides=(2,2)))
model.add(Conv2D(384,(3,3),strides=(1,1),padding='same',activation='relu',kernel_initializer='uniform'))
model.add(Conv2D(384,(3,3),strides=(1,1),padding='same',activation='relu',kernel_initializer='uniform'))
model.add(Conv2D(256,(3,3),strides=(1,1),padding='same',activation='relu',kernel_initializer='uniform'))
model.add(MaxPooling2D(pool_size=(3,3),strides=(2,2)))
model.add(Flatten())
model.add(Dense(4096,activation='relu'))
model.add(Dropout(0.5))
model.add(Dense(4096,activation='relu'))
model.add(Dropout(0.5))
model.add(Dense(1000,activation='softmax'))
model.compile(loss='categorical_crossentropy',optimizer='sgd',metrics=['accuracy'])
model.summary()

实现后打印的模型结构如下：

接下来用cifar10复现Alex论文，代码如下：

import keras
from keras.models import Model
from keras.datasets import cifar10
from keras.utils import to_categorical
from keras.layers import Input,Conv2D,GlobalMaxPooling2D,Dense,BatchNormalization,Activation,MaxPooling2D
from keras.layers import concatenate,Dropout,Flatten

from keras import optimizers
from keras.preprocessing.image import ImageDataGenerator
from keras.callbacks import LearningRateScheduler,TensorBoard,ModelCheckpoint

batch_size=64#一个batch用64张图
iterations=782#一个epoch用782个batch
epochs=300#一共循环300个epoch
num_classes=10
log_filepath='./alexnet'#tensorbroad的文件存储的路径
DATA_FORMAT='channels_last'

#数据预处理并设置学习率的变换规律
def color_preprocessing(x_train,x_test):#??????颜色预处理
    x_train=x_train.astype('float32')
    x_test=x_test.astype('float32')
    mean=[123.307,122.95,113.865]
    std = [62.9932, 62.0887, 66.7048]
    for i in range(3):#i代表通道数
        x_train[:,:,:,i]=(x_train[:,:,:,i]-mean[i])/std[i]
        x_test[:, :, :, i] = (x_test[:, :, :, i] - mean[i]) / std[i]
    return x_train,x_test

def scheduler(epoch):
    if epoch<100:
        return 0.01
    if epoch<200:
        return 0.001
    return 0.0001

#加载数据
(x_train,y_train),(x_test,y_test)=cifar10.load_data()
y_train=to_categorical(y_train,num_classes=num_classes)#num_classes: 总类别数。
y_test=to_categorical(y_test,num_classes)
x_train,x_test=color_preprocessing(x_train,x_test)

#搭建网络
def alexnet(img_input,classes=10):
    x=Conv2D(96,(11,11),strides=(4,4),padding='same',
             activation='relu',kernel_initializer='uniform')(img_input)
    x=MaxPooling2D(pool_size=(3,3),strides=(2,2),padding='same',data_format=DATA_FORMAT)(x)
    x=Conv2D(256,(5,5),strides=(1,1),padding='same',
             activation='relu',kernel_initializer='uniform')(x)
    x = MaxPooling2D(pool_size=(3, 3), strides=(2, 2), padding='same', data_format=DATA_FORMAT)(x)
    x=Conv2D(384,(3,3),strides=(1,1),padding='same',
             activation='relu',kernel_initializer='uniform')(x)
    x=Conv2D(384,(3,3),strides=(1,1),padding='same',
             activation='relu',kernel_initializer='uniform')(x)
    x=Conv2D(256,(3,3),strides=(1,1),padding='same',
             activation='relu',kernel_initializer='uniform')(x)
    x=MaxPooling2D(pool_size=(3,3),strides=(2,2),padding='same',data_format=DATA_FORMAT)(x)
    x=Flatten()(x)
    x=Dense(4096,activation='relu')(x)
    x=Dropout(0.5)(x)
    x=Dense(4096,activation='relu')(x)
    x=Dropout(0.5)(x)
    out=Dense(classes,activation='softmax')(x)
    return out

img_input=Input(shape=(32,32,3))
output=alexnet(img_input)
model=Model(img_input,output)
model.summary()

#momentum: float >= 0. 参数，用于加速 SGD 在相关方向上前进，并抑制震荡。
#nesterov: boolean. 是否使用 Nesterov 动量。
sgd=optimizers.SGD(lr=.1,momentum=0.9,nesterov=True)
model.compile(loss='categorical_crossentropy',optimizer=sgd,metrics=['accuracy'])

# 设置回调函数
#histogram_freq: 对于模型中各个层计算激活值和模型权重直方图的频率（训练轮数中）。 如果设置成 0 ，直方图不会被计算。
tb_cb=TensorBoard(log_dir=log_filepath,histogram_freq=0)#可以可视化测试和训练的标准评估的动态图
#schedule: 一个函数，接受轮索引数作为输入（整数，从 0 开始迭代） 然后返回一个学习速率作为输出（浮点数）。
change_lr=LearningRateScheduler(scheduler)
cbks=[change_lr,tb_cb]#训练时作为参数

#设置数据增强
datagen=ImageDataGenerator(horizontal_flip=True,#随机将一半图像水平翻转
                           width_shift_range=0.125,#图像在水平方向上平移的范围
                           height_shift_range=0.125,
                           fill_mode='constant',cval=0.) #'constant': kkkkkkkk|abcd|kkkkkkkk (cval=k) 平移完用0来填充

#训练模型
model.fit_generator(datagen.flow(x_train,y_train,batch_size=batch_size),
                    steps_per_epoch=iterations,
                    epochs=epochs,
                    callbacks=cbks,
                    validation_data=(x_test,y_test))
model.save('alexnet.h5')