How to add padding to an image using Detectron2 while preserving correct segmentation location?

Question

I have a training dataset consisting of images of shape (1080, 1920, 3) I would like to augment my data by various methods, including cropping and rotating. The problem is, I am unable to reshape the images back to (1080, 1920, 3). Is there a way to reshape from a smaller crop size (e.g. (640, 640, 3)) to (1080, 1920, 3) ?

If not, I will just add padding to the images. But is there a way to do it with Detectron2 so that the location of my annotated objects also get transformed?

Thank you for the help!