python - How to pass Input to a tensorflow backbone model without getting AttributeError: 'tuple' object has no

I am trying to use ResNet3D from tensorflow-models library but I am getting this weird error when trying to run the block

!pip install tf-models-official==2.17.0

Tensorflow version is 2.18 on the Kaggle notebook.

After installing tf-models-official

from tensorflow.keras.callbacks import EarlyStopping, ReduceLROnPlateau
from tensorflow.keras.models import Model
from tensorflow.keras.layers import Dense, GlobalAveragePooling3D, Input
from tensorflow.keras.optimizers import AdamW
import tensorflow_models as tfm

def create_model():
    base_model = tfm.vision.backbones.ResNet3D(model_id = 50,
        temporal_strides= [3,3,3,3],
        temporal_kernel_sizes = [(5,5,5),(5,5,5,5),(5,5,5,5,5,5),(5,5,5)],
        input_specs=tf.keras.layers.InputSpec(shape=(None, None, IMG_SIZE, IMG_SIZE, 3))
    )
    
    # Unfreeze the base model layers
    base_model.trainable = True
    
    # Create the model
    inputs = Input(shape=[None, None, IMG_SIZE, IMG_SIZE, 3])
    x = base_model(inputs) # B,1,7,7,2048
    x = GlobalAveragePooling3D(data_format="channels_last", keepdims=False)(x)
    x = Dense(1024, activation='relu')(x)
    x = tf.keras.layers.Dropout(0.3)(x)  # Add dropout to prevent overfitting
    outputs = Dense(NUM_CLASSES, activation='softmax')(x)
    
    model = Model(inputs, outputs)
    
    # Compile the model with class weights
    optimizer = AdamW(learning_rate=1e-4, weight_decay=1e-5)
    modelpile(
        optimizer=optimizer,
        loss='sparse_categorical_crossentropy',
        metrics=['accuracy', tf.keras.metrics.AUC()]
    )
    
    return model

# Create and display model
model = create_model()
model.summary()

When I run this, I get the error below:

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-56-363271b4dda8> in <cell line: 39>()
     37 
     38 # Create and display model
---> 39 model = create_model()
     40 model.summary()

<ipython-input-56-363271b4dda8> in create_model()
     18     # Create the model
     19     inputs = Input(shape=(None, None, IMG_SIZE, IMG_SIZE, 3))
---> 20     x = base_model(inputs) # B,1,7,7,2048

/usr/local/lib/python3.10/dist-packages/tf_keras/src/engine/training.py in __call__(self, *args, **kwargs)
    586             layout_map_lib._map_subclass_model_variable(self, self._layout_map)
    587 
--> 588         return super().__call__(*args, **kwargs)

/usr/local/lib/python3.10/dist-packages/tf_keras/src/engine/base_layer.py in __call__(self, *args, **kwargs)
   1101             training=training_mode,
   1102         ):
-> 1103             input_spec.assert_input_compatibility(
   1104                 self.input_spec, inputs, self.name
   1105             )

/usr/local/lib/python3.10/dist-packages/tf_keras/src/engine/input_spec.py in assert_input_compatibility(input_spec, inputs, layer_name)
    300                             "incompatible with the layer: "
    301                             f"expected shape={spec.shape}, "
--> 302                             f"found shape={display_shape(x.shape)}"
    303                         )
    304 

/usr/local/lib/python3.10/dist-packages/tf_keras/src/engine/input_spec.py in display_shape(shape)
    305 
    306 def display_shape(shape):
--> 307     return str(tuple(shape.as_list()))
    308 
    309 

AttributeError: 'tuple' object has no attribute 'as_list'

I have tried passing the input to the shape argument as a list, but still getting the same error.

The error is occurring with this

!pip install tf-models-official==2.17.0

import tensorflow as tf

inputs = tf.keras.Input(shape=[None, None, IMG_SIZE, IMG_SIZE, 3])
print(inputs.shape.as_list())

Error:

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-39-6e88680ff7df> in <cell line: 2>()
      1 inputs = tf.keras.Input(shape=[None, None, IMG_SIZE, IMG_SIZE, 3])
----> 2 print(inputs.shape.as_list())

AttributeError: 'tuple' object has no attribute 'as_list'

I am trying to use ResNet3D from tensorflow-models library but I am getting this weird error when trying to run the block

!pip install tf-models-official==2.17.0

Tensorflow version is 2.18 on the Kaggle notebook.

After installing tf-models-official

from tensorflow.keras.callbacks import EarlyStopping, ReduceLROnPlateau
from tensorflow.keras.models import Model
from tensorflow.keras.layers import Dense, GlobalAveragePooling3D, Input
from tensorflow.keras.optimizers import AdamW
import tensorflow_models as tfm

def create_model():
    base_model = tfm.vision.backbones.ResNet3D(model_id = 50,
        temporal_strides= [3,3,3,3],
        temporal_kernel_sizes = [(5,5,5),(5,5,5,5),(5,5,5,5,5,5),(5,5,5)],
        input_specs=tf.keras.layers.InputSpec(shape=(None, None, IMG_SIZE, IMG_SIZE, 3))
    )
    
    # Unfreeze the base model layers
    base_model.trainable = True
    
    # Create the model
    inputs = Input(shape=[None, None, IMG_SIZE, IMG_SIZE, 3])
    x = base_model(inputs) # B,1,7,7,2048
    x = GlobalAveragePooling3D(data_format="channels_last", keepdims=False)(x)
    x = Dense(1024, activation='relu')(x)
    x = tf.keras.layers.Dropout(0.3)(x)  # Add dropout to prevent overfitting
    outputs = Dense(NUM_CLASSES, activation='softmax')(x)
    
    model = Model(inputs, outputs)
    
    # Compile the model with class weights
    optimizer = AdamW(learning_rate=1e-4, weight_decay=1e-5)
    model.compile(
        optimizer=optimizer,
        loss='sparse_categorical_crossentropy',
        metrics=['accuracy', tf.keras.metrics.AUC()]
    )
    
    return model

# Create and display model
model = create_model()
model.summary()

When I run this, I get the error below:

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-56-363271b4dda8> in <cell line: 39>()
     37 
     38 # Create and display model
---> 39 model = create_model()
     40 model.summary()

<ipython-input-56-363271b4dda8> in create_model()
     18     # Create the model
     19     inputs = Input(shape=(None, None, IMG_SIZE, IMG_SIZE, 3))
---> 20     x = base_model(inputs) # B,1,7,7,2048

/usr/local/lib/python3.10/dist-packages/tf_keras/src/engine/training.py in __call__(self, *args, **kwargs)
    586             layout_map_lib._map_subclass_model_variable(self, self._layout_map)
    587 
--> 588         return super().__call__(*args, **kwargs)

/usr/local/lib/python3.10/dist-packages/tf_keras/src/engine/base_layer.py in __call__(self, *args, **kwargs)
   1101             training=training_mode,
   1102         ):
-> 1103             input_spec.assert_input_compatibility(
   1104                 self.input_spec, inputs, self.name
   1105             )

/usr/local/lib/python3.10/dist-packages/tf_keras/src/engine/input_spec.py in assert_input_compatibility(input_spec, inputs, layer_name)
    300                             "incompatible with the layer: "
    301                             f"expected shape={spec.shape}, "
--> 302                             f"found shape={display_shape(x.shape)}"
    303                         )
    304 

/usr/local/lib/python3.10/dist-packages/tf_keras/src/engine/input_spec.py in display_shape(shape)
    305 
    306 def display_shape(shape):
--> 307     return str(tuple(shape.as_list()))
    308 
    309 

AttributeError: 'tuple' object has no attribute 'as_list'

I have tried passing the input to the shape argument as a list, but still getting the same error.

The error is occurring with this

!pip install tf-models-official==2.17.0

import tensorflow as tf

inputs = tf.keras.Input(shape=[None, None, IMG_SIZE, IMG_SIZE, 3])
print(inputs.shape.as_list())

Error:

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-39-6e88680ff7df> in <cell line: 2>()
      1 inputs = tf.keras.Input(shape=[None, None, IMG_SIZE, IMG_SIZE, 3])
----> 2 print(inputs.shape.as_list())

AttributeError: 'tuple' object has no attribute 'as_list'

• python
• tensorflow
• machine-learning
• deep-learning
• tf.keras

Share Improve this question Follow edited Feb 7 at 15:37 Daraan 4,23277 gold badges2222 silver badges4747 bronze badges asked Feb 4 at 11:22 SiladittyaSiladittya 1,20533 gold badges1717 silver badges4444 bronze badges 2

• @Anerdw I have updated the post with the necessary imports. However, in the last part of the post, you can see that the issue is not related to the tensorflow-models library. Rather, it is related to tf.keras.Input. That being said, I would be grateful, if you could help me with this issue. – Siladittya Commented Feb 4 at 15:07
• reverted title to be closer to original question again. – Daraan Commented Feb 7 at 15:37

Add a comment  | 

1 Answer 1

Sorted by: Reset to default 0

This is indeed a bit tricky as several things here are mixed on the tf and keras level.

At best you use the model factories and setups via configs

The main problem here is that you should not pass a layer here but an Input tensor created from tf_keras.Input. Below I pointed you out with two ways to set up your model:

import tf_keras

def create_model():

    input_specs = tf.keras.layers.InputSpec(shape=(None, None, IMG_SIZE, IMG_SIZE, 3))
    # Setup Backbone
    backbone = tfm.vision.backbones.ResNet3D(
        model_id=50,
        temporal_strides=[3, 3, 3, 3],
        temporal_kernel_sizes=[(5, 5, 5), (5, 5, 5, 5), (5, 5, 5, 5, 5, 5), (5, 5, 5)],
        input_specs=input_specs,
    )

    # Variant 1 use functional API yourself
    inputs = tf_keras.Input(shape=input_specs.shape[1:], name=input_specs.name)
    endpoints = backbone(inputs)
    x = endpoints[max(endpoints.keys())]  # <- Use your own function API from here

    ...  # set up your head and model manually

    # -- OR ---

    # Variant 2 Use a Classification Model with your backbone
    base_model = tfm.vision.classification_model.ClassificationModel(
        backbone=backbone,
        num_classes=NUM_CLASSES,
        input_specs=input_specs,
        dropout_rate=0.3,
        skip_logits_layer=False, # set to true to skip Droputout and final Dense Layer
        add_head_batch_norm=True, # adds a batchNorm by default
    )