图示过程

核心代码

from keras.callbacks import CSVLogger, ModelCheckpoint, EarlyStopping

from keras.callbacks import ReduceLROnPlateau

from models.cnn import mini_XCEPTION

# parameters1、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、do_random_crop

batch_size = 32

num_epochs = 1000

validation_split = .2

do_random_crop = False  #random crop only works for classification since the current implementation does no transform bounding boxes

patience = 100

num_classes = 2

dataset_name = 'imdb'

input_shape = (64, 64, 1)

#if判断，然后指定图像、log、loghdf5各自保存路径

if input_shape[2] == 1:

   grayscale = True

images_path = '../datasets/imdb_crop/'

log_file_path = '../trained_models/gender_models/gender_training.log'

trained_models_path = '../trained_models/gender_models/gender_mini_XCEPTION'

# data loader

data_loader = DataManager(dataset_name) #自定义DataManager函数实现根据数据集name进行加载

ground_truth_data = 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，

train_keys, val_keys = split_imdb_data(ground_truth_data, validation_split)

print('Number of training samples:', len(train_keys))

print('Number of validation samples:', len(val_keys))

#调用ImageDataGenerator函数实现实时数据增强生成小批量的图像数据。

image_generator = ImageGenerator(ground_truth_data, batch_size,

                                input_shape[:2],

                                train_keys, val_keys, None,

                                path_prefix=images_path,

                                vertical_flip_probability=0,

                                grayscale=grayscale,

                                do_random_crop=do_random_crop)

# 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)

model.compile(optimizer='adam',

             loss='categorical_crossentropy',

             metrics=['accuracy'])

model.summary()

#3、指定要训练的数据集(gender→imdb即男女数据集)

# model callbacks

# callbacks4、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

early_stop = EarlyStopping('val_loss', patience=patience)

reduce_lr = ReduceLROnPlateau('val_loss', factor=0.1,

                             patience=int(patience/2), verbose=1)

csv_logger = CSVLogger(log_file_path, append=False)

model_names = trained_models_path + '.{epoch:02d}-{val_acc:.2f}.hdf5'

model_checkpoint = ModelCheckpoint(model_names,

                                  monitor='val_loss',

                                  verbose=1,

                                  save_best_only=True,

                                  save_weights_only=False)

callbacks = [model_checkpoint, csv_logger, early_stop, reduce_lr]

# training model5、调用fit_generator函数训练模型

model.fit_generator(image_generator.flow(mode='train'),

                   steps_per_epoch=int(len(train_keys) / batch_size),

                   epochs=num_epochs, verbose=1,

                   callbacks=callbacks,

                   validation_data=image_generator.flow('val'),

                   validation_steps=int(len(val_keys) / batch_size))