Add 2D/3D to Sequence Transformation

[danieltudosiu]

Add the Ordering class that allows different processing to an input latent representation.

[danieltudosiu]

The Ordering class is dependent on the Generalized Hilbert repo for the 3D generalization of the Hilbert curves.

Sadly this repo was not transformed into a pip installable package so we need to either have it as a submodule (not sure how that would play with a pip install) or just copy and support their code in our codebase.

@ericspod @Warvito what is your take on this?

[Warvito]

I think for now it is not necessary to add the Hilbert Curve implementation, just the raster scan and the s-curve is enough.

[danieltudosiu]

@Warvito Hilbert curve is one of the s-curves (space-filling curves) that is available.

[Warvito]

Hi @danieltudosiu , I don't understood. From our code, we have 4 types of orderings: random, s_curve, raster, and hilbert (https://github.com/AmigoLab/vqvae/blob/fa505cd0a3a8ea1cf3d535c4e4c481e093fde854/src/networks/transformers/img2seq_ordering.py#L12). Just the hilbert depends on the gilbert package, right?

[danieltudosiu]

Sorry, you are right. Yes, I will go forward with the s-curves and row-major rasterisation.