Assemble two model into one, but encount ValueError: Graph disconnected

General Discussion
, ,
#1

i have two models:
first one is mobile_net, which is a mobilenet assign with input shape of 192,192,3, and output shape of 6,6,1280.it contains pretrain weights.
second model is model_0 assign with input shape of 6,6,1280, which has two layers of GlobalAveragePooling2D and Dense.( model_0 is well trianed)

now i connect the mobile_net and model_0 together, hoping the data will flow through.
I finally assembled the two model successfully, the assembled model is named model_1
But when i extract some layer outputs of the model_1, wired things happened~~
i can get the -4 layer output successfully
i can get the -3 layer output successfully
i can not get the -2 layer output
i can not get the -1 layer output

the demo code is here
inputs=tf.keras.layers.Input([6,6,1280])
x=tf.keras.layers.GlobalAveragePooling2D(
name=‘global_average_pooling2d_0’)(inputs)
y=tf.keras.layers.Dense(5,name=‘dense_0’)(x)
model_0 = tf.keras.Model(inputs,y,name=‘model_0’)

mobile_net = tf.keras.applications.MobileNetV2(
input_shape=(192,192,3),
include_top=False,
weights=‘imagenet’)
assert mobile_net.output_shape==(None, 6, 6, 1280)
y1=model_0.get_layer(‘global_average_pooling2d_0’)(mobile_net.output)
y2=model_0.get_layer(‘dense_0’)(y1)
model_1 = tf.keras.Model(mobile_net.input,y2,name=‘assembled_model’)

tmp=tf.keras.Model(model_1.input,model_1.layers[-3].output) # successfully
tmp=tf.keras.Model(model_1.input,model_1.layers[-2].output) # failed
tmp=tf.keras.Model(model_1.input,model_1.layers[-1].output) # failed

the failed outputs are

ValueError: Graph disconnected: cannot obtain value for tensor KerasTensor(type_spec=TensorSpec(shape=(None, 6, 6, 1280), dtype=tf.float32, name=‘input_12’), name=‘input_12’, description=“created by layer ‘input_12’”) at layer “global_average_pooling2d_0”. The following previous layers were accessed without issue: []

ValueError: Graph disconnected: cannot obtain value for tensor KerasTensor(type_spec=TensorSpec(shape=(None, 6, 6, 1280), dtype=tf.float32, name=‘input_12’), name=‘input_12’, description=“created by layer ‘input_12’”) at layer “global_average_pooling2d_0”. The following previous layers were accessed without issue: []

#3

Hi @Eric_ZHANG

Welcome to the TensorFlow Forum!

This error occurred because both the functional models have different input shapes and unable to connect the graph correctly with the above method. You can ensemble these both models using below code:

input1 = model_0.input
input2 = model_1.input

m1 = model_0(input1)
m2 = model_1(input2)

output = tf.keras.layers.Concatenate()([m1,m2])
output = tf.keras.layers.Dense(10, activation='softmax')(output) 
final_model = tf.keras.Model(inputs=[input1,input2], outputs=output)

tf.keras.utils.plot_model(final_model, show_shapes=True, dpi=50)

Output:
img

Please refer to the attached replicated gist for your reference. Thank you.