Effective Model Tuning

Hello, I am training a Unet model for Image Segmentation. I wanted to know what is the best way to carry out tests. I want to evaluate the effect of pre-filtering images. data augmentation, adding an extra layer etc.

I am training my model for 100 epochs. What is the most efficient to evaluate models? Do I have to retrain the entire model and compare the validation scores? This is really time consuming, is there a smarter way to do this?