使用定制数据生成器连体网络输入问题

import numpy as np
import tensorflow as tf
from tensorflow.keras.models import Model
from tensorflow.keras.layers import Input, Conv2D, Flatten, Dense

# 定义定制数据生成器
class CustomDataGenerator:
    def __init__(self, num_samples, img_shape):
        self.num_samples = num_samples
        self.img_shape = img_shape

    def generate(self):
        for _ in range(self.num_samples):
            # 生成随机图像数据
            img = np.random.rand(*self.img_shape)
            yield img, img  # 假设生成相同图像作为输入和标签

# 定义连体网络结构
def create_siamese_network(input_shape):
    input_a = Input(shape=input_shape)
    input_b = Input(shape=input_shape)

    # 共享权重的卷积层
    conv_layer = Conv2D(32, (3, 3), activation='relu')
    processed_a = conv_layer(input_a)
    processed_b = conv_layer(input_b)

    # 展平并连接
    flatten_layer = Flatten()
    flattened_a = flatten_layer(processed_a)
    flattened_b = flatten_layer(processed_b)

    # 全连接层
    dense_layer = Dense(128, activation='relu')
    dense_a = dense_layer(flattened_a)
    dense_b = dense_layer(flattened_b)

    # 计算相似度
    distance = tf.keras.layers.Lambda(lambda tensors: tf.abs(tensors[0] - tensors[1]))([dense_a, dense_b])
    output = Dense(1, activation='sigmoid')(distance)

    model = Model(inputs=[input_a, input_b], outputs=output)
    return model

# 使用示例
num_samples = 1000
img_shape = (64, 64, 3)
batch_size = 32

data_generator = CustomDataGenerator(num_samples, img_shape)
siamese_network = create_siamese_network(img_shape)

siamese_network.compile(optimizer='adam', loss='binary_crossentropy')
siamese_network.fit(data_generator.generate(), steps_per_epoch=num_samples // batch_size, epochs=10)