CV：基于Keras利用CNN主流架构之mini_XCEPTION训练情感分类模型hdf5并保存到指定文件夹下

CV：基于Keras利用CNN主流架构之mini_XCEPTION训练情感分类模型hdf5并保存到指定文件夹下

 

 

目录

图示过程

核心代码

---

 

 

 

图示过程

 

核心代码

def mini_XCEPTION(input_shape, num_classes, l2_regularization=0.01):
    regularization = l2(l2_regularization)

    # base
    img_input = Input(input_shape)
    x = Conv2D(8, (3, 3), strides=(1, 1), kernel_regularizer=regularization,
                                            use_bias=False)(img_input)
    x = BatchNormalization()(x)
    x = Activation('relu')(x)
    x = Conv2D(8, (3, 3), strides=(1, 1), kernel_regularizer=regularization,
                                            use_bias=False)(x)
    x = BatchNormalization()(x)
    x = Activation('relu')(x)

    # module 1
    residual = Conv2D(16, (1, 1), strides=(2, 2),
                      padding='same', use_bias=False)(x)
    residual = BatchNormalization()(residual)

    x = SeparableConv2D(16, (3, 3), padding='same',
                        kernel_regularizer=regularization,
                        use_bias=False)(x)
    x = BatchNormalization()(x)
    x = Activation('relu')(x)
    x = SeparableConv2D(16, (3, 3), padding='same',
                        kernel_regularizer=regularization,
                        use_bias=False)(x)
    x = BatchNormalization()(x)

    x = MaxPooling2D((3, 3), strides=(2, 2), padding='same')(x)
    x = layers.add([x, residual])

    # module 2
    residual = Conv2D(32, (1, 1), strides=(2, 2),
                      padding='same', use_bias=False)(x)
    residual = BatchNormalization()(residual)

    x = SeparableConv2D(32, (3, 3), padding='same',
                        kernel_regularizer=regularization,
                        use_bias=False)(x)
    x = BatchNormalization()(x)
    x = Activation('relu')(x)
    x = SeparableConv2D(32, (3, 3), padding='same',
                        kernel_regularizer=regularization,
                        use_bias=False)(x)
    x = BatchNormalization()(x)

    x = MaxPooling2D((3, 3), strides=(2, 2), padding='same')(x)
    x = layers.add([x, residual])

    # module 3
    residual = Conv2D(64, (1, 1), strides=(2, 2),
                      padding='same', use_bias=False)(x)
    residual = BatchNormalization()(residual)

    x = SeparableConv2D(64, (3, 3), padding='same',
                        kernel_regularizer=regularization,
                        use_bias=False)(x)
    x = BatchNormalization()(x)
    x = Activation('relu')(x)
    x = SeparableConv2D(64, (3, 3), padding='same',
                        kernel_regularizer=regularization,
                        use_bias=False)(x)
    x = BatchNormalization()(x)

    x = MaxPooling2D((3, 3), strides=(2, 2), padding='same')(x)
    x = layers.add([x, residual])

    # module 4
    residual = Conv2D(128, (1, 1), strides=(2, 2),
                      padding='same', use_bias=False)(x)
    residual = BatchNormalization()(residual)

    x = SeparableConv2D(128, (3, 3), padding='same',
                        kernel_regularizer=regularization,
                        use_bias=False)(x)
    x = BatchNormalization()(x)
    x = Activation('relu')(x)
    x = SeparableConv2D(128, (3, 3), padding='same',
                        kernel_regularizer=regularization,
                        use_bias=False)(x)
    x = BatchNormalization()(x)

    x = MaxPooling2D((3, 3), strides=(2, 2), padding='same')(x)
    x = layers.add([x, residual])

    x = Conv2D(num_classes, (3, 3),
            #kernel_regularizer=regularization,
            padding='same')(x)
    x = GlobalAveragePooling2D()(x)
    output = Activation('softmax',name='predictions')(x)

    model = Model(img_input, output)
    return model

 

#CV：利用CNN主流架构之一的XCEPTION训练情感分类模型.hdf5并保存到指定文件夹下边
from keras.callbacks import CSVLogger, ModelCheckpoint, EarlyStopping
from keras.callbacks import ReduceLROnPlateau
from keras.preprocessing.image import ImageDataGenerator

from models.cnn import mini_XCEPTION


# parameters 1、定义参数：每个batch的采样本数、训练轮数、输入shape、部分比例分离用于验证、冗长参数、分类个数、patience、loghdf5保存路径
batch_size = 32    #整数，指定进行梯度下降时每个batch包含的样本数。训练时一个batch的样本会被计算一次梯度下降，使目标函数优化一步。
num_epochs = 10000 #整数，训练终止时的epoch值，训练将在达到该epoch值时停止，当没有设置initial_epoch时，它就是训练的总轮数，否则训练的总轮数为epochs - inital_epoch
input_shape = (64, 64, 1)
validation_split = .2  #0~1之间的浮点数，用来指定训练集的一定比例数据作为验证集。验证集将不参与训练，并在每个epoch结束后测试的模型的指标，如损失函数、精确度等。
verbose = 1  #日志显示，0为不在标准输出流输出日志信息，1为输出进度条记录，2为每个epoch输出一行记录
num_classes = 7
patience = 50  #当monitor不再有改善的时候就会停止训练，这个可以通过patience看出来
base_path = '../trained_models/emotion_models/'

# data generator调用ImageDataGenerator函数实现实时数据增强生成小批量的图像数据。
data_generator = ImageDataGenerator(
                        featurewise_center=False,
                        featurewise_std_normalization=False,
                        rotation_range=10,
                        width_shift_range=0.1,
                        height_shift_range=0.1,
                        zoom_range=.1,
                        horizontal_flip=True)

# model parameters/compilation2、建立XCEPTION模型并compile编译配置参数，最后输出网络摘要
model = mini_XCEPTION(input_shape, num_classes)  #mini_XCEPTION函数(XCEPTION是属于CNN下目前最新的一种模型)实现输入形状、分类个数两个参数建立模型
model.compile(optimizer='adam', loss='categorical_crossentropy',  #model.compile函数(属于keras库)用来配置训练模型参数，可以指定你设想的随机梯度下降中的网络的损失函数、优化方式等参数
              metrics=['accuracy'])
model.summary()  #Prints a string summary of the network.

#3、指定要训练的数据集(emotion→fer2013即喜怒哀乐数据集)
datasets = ['fer2013']
#4、for循环实现callbacks、loading dataset
for dataset_name in datasets:  
    print('Training dataset:', dataset_name)

    # callbacks回调：通过调用CSVLogger、EarlyStopping、ReduceLROnPlateau、ModelCheckpoint等函数得到训练参数存到一个list内
    log_file_path = base_path + dataset_name + '_emotion_training.log'
    csv_logger = CSVLogger(log_file_path, append=False)  #Callback that streams epoch results to a csv file.
    early_stop = EarlyStopping('val_loss', patience=patience) #Stop training when a monitored quantity has stopped improving.
    reduce_lr = ReduceLROnPlateau('val_loss', factor=0.1,  #Reduce learning rate when a metric has stopped improving.
                                  patience=int(patience/4), verbose=1)
    trained_models_path = base_path + dataset_name + '_mini_XCEPTION'
    model_names = trained_models_path + '.{epoch:02d}-{val_acc:.2f}.hdf5'
    model_checkpoint = ModelCheckpoint(model_names, 'val_loss', verbose=1,  #Save the model after every epoch
                                                    save_best_only=True)
    callbacks = [model_checkpoint, csv_logger, early_stop, reduce_lr] #

    # loading dataset加载数据集：通过调用DataManager、
    data_loader = DataManager(dataset_name, image_size=input_shape[:2]) #自定义DataManager函数实现根据数据集name进行加载
    faces, emotions = data_loader.get_data() #自定义get_data函数根据不同数据集name得到各自的ground truth data，
    faces = preprocess_input(faces)  #自定义preprocess_input函数：处理输入的数据，先转为float32类型然后/ 255.0
    num_samples, num_classes = emotions.shape  #shape函数读取矩阵的长度
    train_data, val_data = split_data(faces, emotions, validation_split)  #自定义split_data对数据整理各取所得train_data、 val_data 
    train_faces, train_emotions = train_data
    #training model调用fit_generator函数训练模型
    model.fit_generator(data_generator.flow(train_faces, train_emotions,  #flow函数返回Numpy Array Iterator迭代
                                            batch_size),
                        steps_per_epoch=len(train_faces) / batch_size,
                        epochs=num_epochs, verbose=1, callbacks=callbacks,
                        validation_data=val_data)  #fit_generator函数Fits the model on data generated batch-by-batch by a Python generator