TensorFlow报错：Cannot iterate over a Tensor with unknown first dimension求助

在Kaggle平台运行自定义TensorFlow模型代码，原本基于预训练模型做迁移学习，现在改为自行创建模型训练，出现报错：

Cannot iterate over a Tensor with unknown first dimension.

已指定输入形状，但模型仍无法识别维度，目标是正常打印model.summary()。

原代码 #

def load_layers():
    input_tensor = Input(shape=(IMG_SIZE, IMG_SIZE, ColorChannels))

    #model.add(layers.Input(shape=input_tensor))
    model = models.Sequential()
    model.add(layers.Conv2D(filters=32, kernel_size=(9,9), input_shape=input_tensor))
    model.add(layers.MaxPooling2D(strides=(2, 2)))
    model.add(layers.Conv2D(filters=64, kernel_size=(6,6), activation='relu'))
    model.add(layers.Dropout(0.4))
    model.add(layers.MaxPooling2D(strides=(2, 2)))
    model.add(layers.Conv2D(filters=128, kernel_size=(5,5), activation='relu'))
    model.add(layers.Dropout(0.4))
    model.add(layers.MaxPooling2D(strides=(2, 2)))
    model.add(layers.Conv2D(filters=128, kernel_size=(3,3), activation='relu'))
    model.add(layers.MaxPooling2D(strides=(2, 2)))
    model.add(layers.Flatten())
    model.add(layers.Dense(1024, activation='relu'))
    model.add(layers.Dropout(0.5))
    model.add(layers.Dense(512, activation='relu'))
    model.add(layers.Dropout(0.5))
    model.add(layers.Dense(10, activation='sigmoid'))
    
    baseModel = model(input_tensor=input_tensor)
    headModel = baseModel.output   
    headModel = Dense(10, activation="sigmoid")(headModel)
    model = Model(inputs=baseModel.input, outputs=headModel)
    
    for layer in baseModel.layers:
        layer.trainable = False

    print("Compiling model...")
    model.compile(loss="binary_crossentropy",
                    optimizer='adam',
                    metrics=["accuracy"])

    return model

if TPU_INIT:
    with tpu_strategy.scope():
        model = load_layers()
else:
    model = load_layers()

model.summary()

错误原因 #

• 混淆了Sequential模型与函数式API的用法：把Sequential模型当成层调用（model(input_tensor=input_tensor)），导致baseModel变成了张量而非模型对象，后续遍历baseModel.layers时自然报错（张量没有layers属性，且维度未知）。
• Conv2D的input_shape参数错误：传入了Input张量，而非要求的形状元组。

修正后的代码 #

def load_layers():
    # 用函数式API构建模型，避免混用Sequential
    input_tensor = Input(shape=(IMG_SIZE, IMG_SIZE, ColorChannels))
    
    # 逐层定义张量传递
    x = layers.Conv2D(filters=32, kernel_size=(9,9), activation='relu')(input_tensor)
    x = layers.MaxPooling2D(strides=(2, 2))(x)
    x = layers.Conv2D(filters=64, kernel_size=(6,6), activation='relu')(x)
    x = layers.Dropout(0.4)(x)
    x = layers.MaxPooling2D(strides=(2, 2))(x)
    x = layers.Conv2D(filters=128, kernel_size=(5,5), activation='relu')(x)
    x = layers.Dropout(0.4)(x)
    x = layers.MaxPooling2D(strides=(2, 2))(x)
    x = layers.Conv2D(filters=128, kernel_size=(3,3), activation='relu')(x)
    x = layers.MaxPooling2D(strides=(2, 2))(x)
    x = layers.Flatten()(x)
    x = layers.Dense(1024, activation='relu')(x)
    x = layers.Dropout(0.5)(x)
    x = layers.Dense(512, activation='relu')(x)
    x = layers.Dropout(0.5)(x)
    base_output = layers.Dense(10, activation='sigmoid')(x)
    
    # 明确创建基础模型对象
    baseModel = Model(inputs=input_tensor, outputs=base_output)
    
    # 冻结基础模型层（按需选择）
    for layer in baseModel.layers:
        layer.trainable = False
    
    # 若不需要额外输出层，可直接返回baseModel；如需保留原迁移学习结构，保留以下代码
    headModel = baseModel.output
    headModel = layers.Dense(10, activation="sigmoid")(headModel)
    model = Model(inputs=baseModel.input, outputs=headModel)
    
    print("Compiling model...")
    model.compile(loss="binary_crossentropy",
                  optimizer='adam',
                  metrics=["accuracy"])

    return model

if TPU_INIT:
    with tpu_strategy.scope():
        model = load_layers()
else:
    model = load_layers()

model.summary()

关键修正点 #

• 统一使用函数式API，避免Sequential与函数式API混用，明确区分模型对象和张量。
• 修正Conv2D的输入参数：不再将Input张量传给input_shape，而是通过函数式API的张量传递定义输入。
• baseModel现在是真正的Model对象，可正常遍历layers属性，维度信息也能被正确识别。

内容的提问来源于stack exchange，提问作者slave