NASNetLarge input shape error

[buckeye17]

Summary

I'm able to fit several Keras applications (VGG16, InceptionResNetV2, Xception, etc) without error, but when I attempt to fit NASNetLarge with the line of code shown below, I get the error message shown below. Clearly the error message isn't true based on my line of code since I have include_top=False. In fitting these different applications, I'm simply importing the corresponding function and renaming the function in my line of code given below.

Environment

• Python version: 3.8.12
• Keras version: 2.7.0
• Keras-applications version: 2.7.0
• Keras backend with version: tensorflow-gpu 2.7.0

Logs or source codes for reproduction

base_model = NASNetLarge(classes=n_classes, include_top=False, input_shape=(img_height, img_width, 3), weights='imagenet')

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
~\AppData\Local\Temp/ipykernel_51428/1637328170.py in <module>
      2 # base_model = VGG19(classes=n_classes, include_top=False, input_shape=(img_height, img_width, 3), weights='imagenet')
      3 # base_model = InceptionResNetV2(classes=n_classes, include_top=False, input_shape=(img_height, img_width, 3), weights='imagenet')
----> 4 base_model = NASNetLarge(classes=n_classes, include_top=False, input_shape=(img_height, img_width, 3), weights='imagenet')
      5 
      6 # from tensorflow.keras.applications.inception_resnet_v2 import InceptionResNetV2

~\Anaconda3\envs\deep_learning\lib\site-packages\keras\applications\nasnet.py in NASNetLarge(input_shape, include_top, weights, input_tensor, pooling, classes)
    461           backend that does not support separable convolutions.
    462   """
--> 463   return NASNet(
    464       input_shape,
    465       penultimate_filters=4032,

~\Anaconda3\envs\deep_learning\lib\site-packages\keras\applications\nasnet.py in NASNet(input_shape, penultimate_filters, num_blocks, stem_block_filters, skip_reduction, filter_multiplier, include_top, weights, input_tensor, pooling, classes, default_size, classifier_activation)
    172 
    173   # Determine proper input shape and default size.
--> 174   input_shape = imagenet_utils.obtain_input_shape(
    175       input_shape,
    176       default_size=default_size,

~\Anaconda3\envs\deep_learning\lib\site-packages\keras\applications\imagenet_utils.py in obtain_input_shape(input_shape, default_size, min_size, data_format, require_flatten, weights)
    347     if input_shape is not None:
    348       if input_shape != default_shape:
--> 349         raise ValueError('When setting `include_top=True` '
    350                          'and loading `imagenet` weights, '
    351                          f'`input_shape` should be {default_shape}.  '

ValueError: When setting `include_top=True` and loading `imagenet` weights, `input_shape` should be (331, 331, 3).  Received: input_shape=(256, 256, 3)