如何在Keras中使用flow_from_directory进行多类语义分割？

from keras.preprocessing.image import ImageDataGenerator
from keras.models import Model
from keras.layers import Conv2D, MaxPooling2D, Dropout, UpSampling2D, concatenate

data_gen_args = dict(rescale=1./255, validation_split=0.2)  # 数据增强参数
image_datagen = ImageDataGenerator(**data_gen_args)
mask_datagen = ImageDataGenerator(**data_gen_args)

image_generator = image_datagen.flow_from_directory(
    'path_to_images_folder',
    target_size=(height, width),
    class_mode=None,
    subset='training',
    seed=seed)
    
mask_generator = mask_datagen.flow_from_directory(
    'path_to_masks_folder',
    target_size=(height, width),
    class_mode=None,
    subset='training',
    seed=seed)

train_generator = zip(image_generator, mask_generator)

inputs = Input(shape=(height, width, channels))
conv1 = Conv2D(32, (3, 3), activation='relu', padding='same')(inputs)
conv1 = Conv2D(32, (3, 3), activation='relu', padding='same')(conv1)
pool1 = MaxPooling2D(pool_size=(2, 2))(conv1)

# 定义更多的卷积和池化层

conv5 = Conv2D(64, (3, 3), activation='relu', padding='same')(conv5)
conv5 = Conv2D(64, (3, 3), activation='relu', padding='same')(conv5)
up6 = concatenate([UpSampling2D(size=(2, 2))(conv5), conv4], axis=3)

# 定义更多的上采样和连接层

conv9 = Conv2D(num_classes, (1, 1), activation='softmax')(conv9)

model = Model(inputs=inputs, outputs=conv9)
model.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['accuracy'])
model.fit_generator(train_generator, steps_per_epoch=num_samples // batch_size, epochs=num_epochs)