Accessing to Weights and Layers in Tensorflow Hub - tensorflow

Question

When I try to get the model from tensorflow-hub resporitory.
I can see it as a Saved Model format, but I cant get access to model architecture as well as weights store for each layer.
import tensorflow_hub as hub
model = hub.load("https://tfhub.dev/tensorflow/centernet/hourglass_512x512/1")
)
Is there any formal way to work with it?
All the attribute I can get through model.__dict__ is not clear for a specific layer in the original model.
{'_self_setattr_tracking': True,
'_self_unconditional_checkpoint_dependencies': [TrackableReference(name='_model', ref=<tensorflow.python.saved_model.load.Loader._recreate_base_user_object.<locals>._UserObject object at 0x7fe4e4914710>),
TrackableReference(name='signatures', ref=_SignatureMap({'serving_default': <ConcreteFunction signature_wrapper(input_tensor) at 0x7FE4E601F210>})),
TrackableReference(name='_self_saveable_object_factories', ref=DictWrapper({}))],
'_self_unconditional_dependency_names': {'_model': <tensorflow.python.saved_model.load.Loader._recreate_base_user_object.<locals>._UserObject at 0x7fe4e4914710>,
'signatures': _SignatureMap({'serving_default': <ConcreteFunction signature_wrapper(input_tensor) at 0x7FE4E601F210>}),
'_self_saveable_object_factories': {}},
'_self_unconditional_deferred_dependencies': {},
'_self_update_uid': 176794,
'_self_name_based_restores': set(),
'_self_saveable_object_factories': {},
'_model': <tensorflow.python.saved_model.load.Loader._recreate_base_user_object.<locals>._UserObject at 0x7fe4e4914710>,
'signatures': _SignatureMap({'serving_default': <ConcreteFunction signature_wrapper(input_tensor) at 0x7FE4E601F210>}),
'__call__': <tensorflow.python.saved_model.function_deserialization.RestoredFunction at 0x7fe315a28950>,
'graph_debug_info': ,
'tensorflow_version': '2.4.0',
'tensorflow_git_version': 'unknown'}
I have also tried with model.signatures['serving_default'].__dict__, the Tensor represents for each layer is not visible
[<tf.Tensor: shape=(), dtype=resource, numpy=<unprintable>>,
<tf.Tensor: shape=(), dtype=resource, numpy=<unprintable>>,
<tf.Tensor: shape=(), dtype=resource, numpy=<unprintable>>,
<tf.Tensor: shape=(), dtype=resource, numpy=<unprintable>>,
<tf.Tensor: shape=(), dtype=resource, numpy=<unprintable>>,
<tf.Tensor: shape=(), dtype=resource, numpy=<unprintable>>,
<tf.Tensor: shape=(), dtype=resource, numpy=<unprintable>>,
<tf.Tensor: shape=(), dtype=resource, numpy=<unprintable>>,
<tf.Tensor: shape=(), dtype=resource, numpy=<unprintable>>,
<tf.Tensor: shape=(), dtype=resource, numpy=<unprintable>>,
<tf.Tensor: shape=(), dtype=resource, numpy=<unprintable>>,
<tf.Tensor: shape=(), dtype=resource, numpy=<unprintable>>,
<tf.Tensor: shape=(), dtype=resource, numpy=<unprintable>>,
<tf.Tensor: shape=(), dtype=resource, numpy=<unprintable>>],

Answer 1

With the CLI tool saved_model_cli provided by the package tensorflow-serving-api it's possible to inspect a saved model. In the first step I downloaded and cached the model:
from os import environ
import tensorflow_hub as hub
environ['TFHUB_CACHE_DIR'] = '/Users/you/.cache/tfhub_modules'
hub.load("https://tfhub.dev/tensorflow/centernet/hourglass_512x512/1")
Then I inspected the signatures and layers:
saved_model_cli show --dir /Users/you/.cache/tfhub_modules/3085eb2fbe2ad0b69801d50844c97b7a7a5ecade --all
MetaGraphDef with tag-set: 'serve' contains the following SignatureDefs:
signature_def['__saved_model_init_op']:
The given SavedModel SignatureDef contains the following input(s):
The given SavedModel SignatureDef contains the following output(s):
outputs['__saved_model_init_op'] tensor_info:
dtype: DT_INVALID
shape: unknown_rank
name: NoOp
Method name is:
signature_def['serving_default']:
The given SavedModel SignatureDef contains the following input(s):
inputs['input_tensor'] tensor_info:
dtype: DT_UINT8
shape: (1, -1, -1, 3)
name: serving_default_input_tensor:0
The given SavedModel SignatureDef contains the following output(s):
outputs['detection_boxes'] tensor_info:
dtype: DT_FLOAT
shape: (1, 100, 4)
name: StatefulPartitionedCall:0
outputs['detection_classes'] tensor_info:
dtype: DT_FLOAT
shape: (1, 100)
name: StatefulPartitionedCall:1
outputs['detection_scores'] tensor_info:
dtype: DT_FLOAT
shape: (1, 100)
name: StatefulPartitionedCall:2
outputs['num_detections'] tensor_info:
dtype: DT_FLOAT
shape: (1)
name: StatefulPartitionedCall:3
Method name is: tensorflow/serving/predict
After that I used the debugger to understand how a saved model works internally and found the member fields variables and trainable_variables in model.signatures['serving_default'] which stores the data (weights, ...) of the model. Here you see the output of model.signatures['serving_default'].variables:

Answer 2

Short summary for the answer. We can access to the variables of a layer by model.signatures['serving_default'].variables