Randomly flip rotate, crop, etc., matching pairs of images in Tensorflow? (Image augmentation pipeline)

Question

I want to be able to work with datasets of paired image training examples (like a source image and segmentation map) and add geometric transforms into my pipeline that I can