我正在尝试使用Keras微调一个模型:

    inception_model = InceptionV3(weights=None, include_top=False, input_shape=(150, 150, 1))    x = inception_model.output    x = GlobalAveragePooling2D()(x)    x = Dense(256, activation='relu', name='fc1')(x)    x = Dropout(0.5)(x)    predictions = Dense(10, activation='softmax', name='predictions')(x)    classifier = Model(inception_model.input, predictions)    ####训练训练训练 ... 保存权重    classifier.load_weights("saved_weights.h5")      classifier.layers.pop()    classifier.layers.pop()    classifier.layers.pop()    classifier.layers.pop()    ###足够的pop操作以恢复到标准的InceptionV3     x = classifier.output    x = GlobalAveragePooling2D()(x)    x = Dense(256, activation='relu', name='fc1')(x)    x = Dropout(0.5)(x)    predictions = Dense(10, activation='softmax', name='predictions')(x)    classifier = Model(classifier.input, predictions)

但我遇到了以下错误:

ValueError: Input 0 is incompatible with layer global_average_pooling2d_3: expected ndim=4, found ndim=2

回答：

你不应该在使用函数式API创建的模型（即keras.models.Model）上使用pop()方法。只有Sequential模型（即keras.models.Sequential）内置了pop()方法（用法：model.pop()）。相反，应该使用层的索引或名称来访问特定的层:

classifier.load_weights("saved_weights.h5")x = classifier.layers[-5].output   # 直接使用层的索引x = GlobalAveragePooling2D()(x)