Questions about fine-tuning KerasCV models

[Justim-Chung]

I have read some documentation on Keras, and I have a few questions about how to fine-tune Keras models. I hope to get some answers from the community.

In Keras.Applications, each pre-trained model has a corresponding preprocess_input layer. However, it seems that the KerasCV models do not require a preprocess_input layer. Is this correct?
The official Keras documentation specifically mentions that when fine-tuning, attention should be paid to the 'training' parameters of the batch_normalization layer. Should we also consider the training parameters of the batch_normalization layer when fine-tuning KerasCV models?
The official Keras documentation suggests setting the trainable parameter to freeze certain layers of the model during fine-tuning. However, in some Kaggle notebooks (such as this one), I have noticed that they do not freeze the layers of the model. Which approach is considered best practice?
Thank you all for your answers!