tensorflowtensorflow-litekaggle

How can I convert a TensorFlow 1 "BigGAN" model on Kaggle to TensorFlow Lite format?

How can I convert a TensorFlow 1 "BigGAN" model on Kaggle to TensorFlow Lite format?

https://www.tensorflow.org/hub/tutorials/biggan_generation_with_tf_hub?hl=ja

I am trying to implement the following source code to define a TensorFlow.Keras model and convert it to the TensorFlow Lite format, but when I try to define a model in the Function API using the Layer created with hub.KerasLayer, the following exception occurs:

And I also don't know how to specify the arguments for model.build().

This model requires multiple inputs, so I was unable to use the solution to build a model using the Sequential API shown in the URL below. How can I save a model with input shape (1, None, None, 3) with None fixed to 256?

Are there any good solutions?

Non-working source code

import tensorflow.compat.v1 as tf
tf.disable_v2_behavior()
from tensorflow.compat.v1 import keras
from tensorflow.compat.v1.keras import layers
import tensorflow_hub as hub

input_truncation = keras.Input(shape=(), name='truncation')
input_y = keras.Input(shape=(1000, ), name='y')
input_z = keras.Input(shape=(128, ), name='z')
hub_layer = hub.KerasLayer(
            "https://www.kaggle.com/models/deepmind/biggan/TensorFlow1/128/2",
            trainable=False,
            signature="default", 
            signature_outputs_as_dict=True,
            input_shape = [[], [1000], [128]],
            output_shape = [128, 128, 3],
        )

### -->> TypeError occur HERE
output = hub_layer([input_truncation, input_y, input_z])
### <<-- TypeError occur HERE

model = tf.keras.models.Model(inputs=[input_truncation, input_y, input_z], outputs=[output])

### -->> How to build this model 
model.build([1], [1, 1000], [1, 128])
### <<-- How to build this model

model.summary()
model.save("biggan-128")

converter = tf.lite.TFLiteConverter.from_keras_model(model)
tflite_model = converter.convert()
open("converted_model.tflite", "wb").write(tflite_model)

the exception that occurred

 TypeError
in user code:

    File "/home/shino/anaconda3/envs/movenet/lib/python3.10/site-packages/tensorflow_hub/keras_layer.py", line 242, in call  *
        result = f()

    TypeError: pruned(truncation, y, z) takes 0 positional arguments, got 1.
  File "/tmp/__autograph_generated_fileedgegq8b.py", line 74, in tf__call
    ag__.if_stmt(ag__.not_(ag__.ld(self)._has_training_argument), if_body_3, else_body_3, get_state_3, set_state_3, ('result', 'training'), 1)
  File "/tmp/__autograph_generated_fileedgegq8b.py", line 37, in if_body_3
    result = ag__.converted_call(ag__.ld(f), (), None, fscope)
TypeError: pruned(truncation, y, z) takes 0 positional arguments, got 1.

During handling of the above exception, another exception occurred:

  File "/tmp/__autograph_generated_fileedgegq8b.py", line 37, in if_body_3
    result = ag__.converted_call(ag__.ld(f), (), None, fscope)
  File "/tmp/__autograph_generated_fileedgegq8b.py", line 74, in tf__call
    ag__.if_stmt(ag__.not_(ag__.ld(self)._has_training_argument), if_body_3, else_body_3, get_state_3, set_state_3, ('result', 'training'), 1)
  File "/home/shino/sandbox/python/biggan/biggan_export.py", line 21, in <module>
    output = hub_layer([input_truncation, input_y, input_z])
TypeError: in user code:

    File "/home/shino/anaconda3/envs/movenet/lib/python3.10/site-packages/tensorflow_hub/keras_layer.py", line 242, in call  *
        result = f()

    TypeError: pruned(truncation, y, z) takes 0 positional arguments, got 1.

requirements.txt

absl-py==2.2.1
astunparse==1.6.3
cachetools==5.5.2
certifi==2025.1.31
charset-normalizer==3.4.1
coloredlogs==15.0.1
flatbuffers==1.12
gast==0.4.0
google-auth==2.38.0
google-auth-oauthlib==0.4.6
google-pasta==0.2.0
grpcio==1.71.0
h5py==3.13.0
humanfriendly==10.0
idna==3.10
keras==2.9.0
Keras-Preprocessing==1.1.2
libclang==18.1.1
Markdown==3.7
MarkupSafe==3.0.2
mpmath==1.3.0
numpy==1.26.4
oauthlib==3.2.2
onnx==1.14.1
onnx-graphsurgeon==0.5.7
onnx2tf==1.26.9
onnxruntime==1.21.0
opt_einsum==3.4.0
packaging==24.2
protobuf==3.20.3
psutil==7.0.0
pyasn1==0.6.1
pyasn1_modules==0.4.2
requests==2.32.3
requests-oauthlib==2.0.0
rsa==4.9
six==1.17.0
sng4onnx==1.0.4
sympy==1.13.3
tensorboard==2.9.0
tensorboard-data-server==0.6.1
tensorboard-plugin-wit==1.8.1
tensorflow==2.9.0
tensorflow-estimator==2.9.0
tensorflow-hub==0.16.1
tensorflow-io-gcs-filesystem==0.37.1
tensorflow-neuron==1.0
termcolor==3.0.0
tf-keras==2.14.1
tf2onnx==1.13.0
typing_extensions==4.13.0
urllib3==2.3.0
Werkzeug==3.1.3
wrapt==1.17.2
Solution

  • I solved it myself.

    By modifying the source code as follows, we were able to successfully convert BigGAN into TensorFlow Lite format.

    Thank you.

    import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import layers
import tensorflow_hub as hub

model_url = "https://www.kaggle.com/models/deepmind/biggan/TensorFlow1/128/2"
input_truncation = keras.Input(batch_shape=(), name='truncation')
input_y = keras.Input(shape=(1000, ), name='y')
input_z = keras.Input(shape=(120, ), name='z')
hub_layer = hub.KerasLayer(
            model_url,
            trainable=False,
            signature="default", 
            signature_outputs_as_dict=True,
        )

inputs = dict(y=input_y, z=input_z, truncation=input_truncation)
output = hub_layer(inputs)
model = tf.keras.models.Model(inputs=[input_truncation, input_y, input_z], outputs=[output])
model.build(([1], [1, 1000], [1, 120]))
model.summary()
model.save("biggan-128")

converter = tf.lite.TFLiteConverter.from_keras_model(model)
tflite_model = converter.convert()
open("biggan-128.tflite", "wb").write(tflite_model)