Remove serialization for alias models

[jbischof]

Serialization is not needed for the aliases in order for test_saved_alias_model to pass. Since these are thin subclasses of ResNetV2Backbone, calling keras.utils.get_serialized_object points to the base class anyway.

>>> keras.utils.serialize_keras_object(keras_cv.models.ResNet50V2Backbone())
{'class_name': 'keras_cv.models>ResNetV2Backbone', 'config': {'stackwise_filters': ListWrapper([64, 128, 256, 512]), 'stackwise_blocks': ListWrapper([3, 4, 6, 3]), 'stackwise_strides': ListWrapper([1, 2, 2, 2]), 'include_rescaling': True, 'input_shape': (None, None, 3), 'stackwise_dilations': ListWrapper([1, 1, 1, 1]), 'input_tensor': None, 'block_type': 'block', 'name': 'res_net_v2_backbone', 'trainable': True}}

[ianstenbit]

Thanks! I need to improve my branch naming game to keep up 💯

[ianstenbit]

/gcbrun