CODEWITHSUNDEEP
tensorflow2.0tensorflow-hub
nattafahh
nattafahh

Reputation: 1

How to get all layers from Keras Model with TensorFlow hub

I use efficientnetv2 from hub.KerasLayer, I would like to see all layers when using model.summary(), but it shows only "keras_layer (KerasLayer)"

Layer (type) Output Shape Param #
keras_layer (KerasLayer) (None, 1280) 5919312
dropout (Dropout) (None, 1280) 0
dense (Dense) (None, 2) 2562

Upvotes: 0

Views: 2306

Answers (2)

Andrey Popov
Andrey Popov

Reputation: 721

TensorFlow's SavedModel is essentially a computation graph. While you could in principle inspect its structure, there is no information about high-level architectural blocks.

If you'd like to access individual layers, a better option might be using the model from the authors' original implementation. It can be constructed as follows:

from effnetv2_model import get_model
model = get_model('efficientnetv2-s', weights='imagenet21k-ft1k', with_endpoints=True)

Pretrained weights are available for the same configurations as in TensorFlow Hub. model.summary() displays individual (Fused)MBConv blocks.

Also note that a future version of TensorFlow will include EfficientNet v2 (already available in nightly builds).

Upvotes: 0

Experience_In_AI
Experience_In_AI

Reputation: 419

This way you can "jump inside" the model;

import tensorflow_hub as hub

malli = hub.KerasLayer("https://tfhub.dev/google/imagenet/efficientnet_v2_imagenet21k_b0/feature_vector/2")
print("Thickness of the model:", len(malli.weights))
for i in range(len(malli.weights)):
    print("In layer ",malli.weights[i].name," the content is: ", malli.weights[i])

...of course note the output is quite a loooong but modify the output print according to your need.

Upvotes: 1

Related Questions