How to define a layer that can handle different input shape? It looks like this:
# dynamic_input_shape_conv_layer is a tf.keras layer input_1 (None, 32, 32, 3
