I have a model with an input (batch of images w/ shape (height, width, time)) that has a dynamically sized dimension (time), which is only determined at runtime. However, the Flatten layer requires fully defined spatial dimensions. Code snippet example:

import tensorflow as tf
from tensorflow.keras.layers import Dense, Flatten, Input

# Define an input with an undefined dimension (None)
input_tensor = Input(shape=(None, 256, 256, None, 13))

# Apply a Dense layer (which expects a fully defined shape)
x = Flatten()(input_tensor)
x = Dense(10)(x)

# Build the model
model = tf.keras.models.Model(inputs=input_tensor, outputs=x)

model.summary()

This raises the error:

ValueError: The last dimension of the inputs to a Dense layer should be defined. Found None.

How can I make it work using Flatten instead of alternatives like GlobalAveragePooling3D? I need to retain all pixel-level information.