steerable pyramid operates on image-like inputs and so should be able to operate on any input whose last two dimensions are height and width, operating independently across all the others. Refactor to make that work, documenting that's what we do.
After doing that, should be able to similarly refactor PortillaSimoncelli.