以下网络架构的设计目的是为了寻找两张图像之间的相似性。
最初，我采用了VGGNet16并移除了分类头：
vgg_model = VGG16(weights="imagenet", include_top=False,input_tensor=Input(shape=(img_width, img_height, channels)))之后，我设置了参数layer.trainable = False，使网络作为特征提取器工作。
我将两张不同的图像输入到网络中：
encoded_left = vgg_model(input_left)encoded_right = vgg_model(input_right)这将产生两个特征向量。然后，为了进行分类（判断它们是否相似），我使用了一个由2个卷积层、后接池化层和4个全连接层组成的度量网络。
merge(encoded_left, encoded_right) -> conv-pool -> conv-pool -> reshape -> dense * 4 -> output因此，模型看起来像这样：
model = Model(inputs=[left_image, right_image], outputs=output)在只训练度量网络后，为了微调卷积层，我设置了最后一个卷积块进行训练。因此，在第二个训练阶段，除了度量网络外，最后一个卷积块也被训练。
现在我想将这个微调后的网络用于其他目的。这里是网络摘要：
__________________________________________________________________________________________________Layer (type) Output Shape Param # Connected to==================================================================================================input_1 (InputLayer) (None, 224, 224, 3) 0__________________________________________________________________________________________________input_2 (InputLayer) (None, 224, 224, 3) 0__________________________________________________________________________________________________vgg16 (Model) (None, 7, 7, 512) 14714688 input_1[0][0] input_2[0][0]__________________________________________________________________________________________________Merged_feature_map (Concatenate (None, 7, 7, 1024) 0 vgg16[1][0] vgg16[2][0]__________________________________________________________________________________________________mnet_conv1 (Conv2D) (None, 7, 7, 1024) 4195328 Merged_feature_map[0][0]__________________________________________________________________________________________________batch_normalization_1 (BatchNor (None, 7, 7, 1024) 4096 mnet_conv1[0][0]__________________________________________________________________________________________________activation_1 (Activation) (None, 7, 7, 1024) 0 batch_normalization_1[0][0]__________________________________________________________________________________________________mnet_pool1 (MaxPooling2D) (None, 3, 3, 1024) 0 activation_1[0][0]__________________________________________________________________________________________________mnet_conv2 (Conv2D) (None, 3, 3, 2048) 8390656 mnet_pool1[0][0]__________________________________________________________________________________________________batch_normalization_2 (BatchNor (None, 3, 3, 2048) 8192 mnet_conv2[0][0]__________________________________________________________________________________________________activation_2 (Activation) (None, 3, 3, 2048) 0 batch_normalization_2[0][0]__________________________________________________________________________________________________mnet_pool2 (MaxPooling2D) (None, 1, 1, 2048) 0 activation_2[0][0]__________________________________________________________________________________________________reshape_1 (Reshape) (None, 1, 2048) 0 mnet_pool2[0][0]__________________________________________________________________________________________________fc1 (Dense) (None, 1, 256) 524544 reshape_1[0][0]__________________________________________________________________________________________________batch_normalization_3 (BatchNor (None, 1, 256) 1024 fc1[0][0]__________________________________________________________________________________________________activation_3 (Activation) (None, 1, 256) 0 batch_normalization_3[0][0]__________________________________________________________________________________________________fc2 (Dense) (None, 1, 128) 32896 activation_3[0][0]__________________________________________________________________________________________________batch_normalization_4 (BatchNor (None, 1, 128) 512 fc2[0][0]__________________________________________________________________________________________________activation_4 (Activation) (None, 1, 128) 0 batch_normalization_4[0][0]__________________________________________________________________________________________________fc3 (Dense) (None, 1, 64) 8256 activation_4[0][0]__________________________________________________________________________________________________batch_normalization_5 (BatchNor (None, 1, 64) 256 fc3[0][0]__________________________________________________________________________________________________activation_5 (Activation) (None, 1, 64) 0 batch_normalization_5[0][0]__________________________________________________________________________________________________fc4 (Dense) (None, 1, 1) 65 activation_5[0][0]__________________________________________________________________________________________________batch_normalization_6 (BatchNor (None, 1, 1) 4 fc4[0][0]__________________________________________________________________________________________________activation_6 (Activation) (None, 1, 1) 0 batch_normalization_6[0][0]__________________________________________________________________________________________________reshape_2 (Reshape) (None, 1) 0 activation_6[0][0]==================================================================================================Total params: 27,880,517Trainable params: 13,158,787Non-trainable params: 14,721,730由于VGGNet的最后一个卷积块已经在自定义数据集上进行了训练，我想在层上切断网络：
__________________________________________________________________________________________________vgg16 (Model) (None, 7, 7, 512) 14714688 input_1[0][0] input_2[0][0]__________________________________________________________________________________________________并将其用作强大的特征提取器。为此任务，我加载了微调后的模型：
model = load_model('model.h5')然后尝试创建新的模型如下：
new_model = Model(Input(shape=(img_width, img_height, channels)), model.layers[2].output)这导致了以下错误：
`AttributeError: Layer vgg16 has multiple inbound nodes, hence the notion of "layer output" is ill-defined. Use `get_output_at(node_index)` instead.`请指导我哪里做错了。
回答：
