After looking at the abstract for the algorithm, I think I can grab some ideas from it to improve the SABR shader.
I need to hunt for the full whitepaper, though.
It does look amazing for low res "cartoony" shapes that have very low color bit depth... and they even say themselves that it kinda sucks for anything meant to look realistic (like, even remotely). But they have some good ideas on ways to decide which edge to extend in upsampling. The technique looks calculation intensive, at least when taken to the extreme shown in the supplemental material -- whenever an algorithm starts talking about optimizing splines and graphs, I start to suspect it's not realtime. I don't even want to think about the amount branching that would be required for a full implementation.
5xBR is circa 2011/2012, and this abstract was supposedly written in 2011, so it's possible 5xBR is based on it in part. Maybe, maybe not.
EDIT: The technique proposed isn't really suitable for realtime rendering.