Loading Saved Model Format Load Times for Object Detection Model API

1. The entire URL of the documentation with the issue

https://tensorflow-object-detection-api-tutorial.readthedocs.io/en/latest/auto_examples/plot_object_detection_saved_model.html#load-the-model

2. Describe the issue

It would be nice if a little more was said about the expected load times and
why such times are in the range of 30-60 seconds for example, when loading a saved model.

I am confused (maybe because of my lack of knowledge about saved model format)
why it takes up to 30 seconds to load a saved exported model for example. Is this typical?
Is there a way to speed it up? I notice there is an example of using the model builder
and loading from a checkpoint but this requires having the TF2 object detection API installed which
requires so many dependencies and should not be required for deployments.

Is there some way to speed up loading of an exported model?
I am caching it in memory anyway but I want to understand it more.

Loading biggers models like Resnet50, inception, and so on from tf.keras.applications dont take this long.

Thanks for the help.