如何在输入层和“模型作为层”之间放置层？

Question

This works:

img = Input(shape=(224,224,3))

efnet = EfficientNetB0(

    weights = 'noisy-student',
    include_top = False,
    pooling = None,
    classes = None

)
for layer in efnet.layers:
    layer.trainable = False

x = efnet(img)
# ... any number of layers ...
x = Dense(1)(x)

model = Model(inputs=img, outputs=x)

This does not ("Graph disconnected"):

img = Input(shape=(224,224,3))
img = Dropout(0.2)(img)
# ^ "Preprocessing" could be anything, Dropout is a simple example

efnet = EfficientNetB0(

    weights = 'noisy-student',
    include_top = False,
    pooling = None,
    classes = None

)
for layer in efnet.layers:
    layer.trainable = False

x = efnet(img)
# ... any number of layers ... 
x = Dense(1)(x)

model = Model(inputs=img, outputs=x)

The only difference between the two is the "preprocessing". Why doesn't this work, and how can I put intermediate layers between an input and a "model as layer" as shown above?

(Specifying an input_shape and/or input_tensor in the efnet declaration has no effect. In fact, specifying an input_tensor mysteriously causes the efnet weights to fail to load, because efnet apparently then has 131 layers (???) instead of the expected 130.)

Answer 1

I found a fix for the problem, though I don't know why the fix makes any difference.

This works:

img = Input(shape=(224,224,3))
x = Dropout(0.2)(img)
x = efnet(x) 
x = Dense(1)(x)
model = Model(inputs=img, outputs=x)

This doesn't:

img = Input(shape=(224,224,3))
img = Dropout(0.2)(img)
x = efnet(img) 
x = Dense(1)(x)
model = Model(inputs=img, outputs=x)