is padding input images better than resizing image?

I am working with images and was thinking of pre-training VGG19 and EffecientNETB0 model on my dataset. However, a query:-

Is padding the images then using EfficientNET better than having dynamic shapes with VGG19?

My data is of a very weird shape (it’s a panorama) so that’s why I was confused as to whether I should use VGG19 or EfficientNET because EN would require heavy padding and cropping while VGG won’t.

In short, is it more beneficial to pad images and use them, or does it yield more performance if I have a model that can accept data of any dimension?

Is there some theoretical explanation for what I should use?