如何在Keras中实现LadderNet (2个U-Net)？(以可用的PyTorch脚本作为参考)

from keras.models import Model
from keras.layers import Input, Conv2D, MaxPooling2D, Dropout, UpSampling2D, BatchNormalization, concatenate

def LadderBlock(input_tensor, filters, kernel_size):
    conv1 = Conv2D(filters, kernel_size, activation='relu', padding='same')(input_tensor)
    conv1 = BatchNormalization()(conv1)
    conv2 = Conv2D(filters, kernel_size, activation='relu', padding='same')(conv1)
    conv2 = BatchNormalization()(conv2)
    return conv2

def LadderNet(input_shape, num_classes):
    inputs = Input(input_shape)

    # Encoder
    conv1 = LadderBlock(inputs, 32, (3, 3))
    pool1 = MaxPooling2D(pool_size=(2, 2))(conv1)
    conv2 = LadderBlock(pool1, 64, (3, 3))
    pool2 = MaxPooling2D(pool_size=(2, 2))(conv2)

    # Decoder
    conv3 = LadderBlock(pool2, 64, (3, 3))
    up1 = UpSampling2D(size=(2, 2))(conv3)
    up1 = concatenate([up1, conv2], axis=-1)
    conv4 = LadderBlock(up1, 32, (3, 3))
    up2 = UpSampling2D(size=(2, 2))(conv4)
    up2 = concatenate([up2, conv1], axis=-1)

    # Output
    outputs = Conv2D(num_classes, (1, 1), activation='softmax')(up2)

    model = Model(inputs=inputs, outputs=outputs)
    return model

input_shape = (256, 256, 3)  # 输入图像的形状
num_classes = 2  # 分割类别数
model = LadderNet(input_shape, num_classes)