I might consider to switch, since what I'm doing now is not a project I believe in and I'm only doing it to get the paper.
I'm working on a generalization of this concept to rendering, but I'm not sure I would publish the results or just write informally about it. The idea is to generalize light rays to depend on the geometry of the scene, allowing them to bend to reach more interesting parts (behind objects or the horizon) as a global function of the scene configuration.
I like the idea of VR/AR user interfaces where light rays bend interactively according to controller/gaze position. So for example by looking at the edge of an object the light rays bend so you see behind it. Or by looking off into the horizon the entire landscape folds over like in these photos. Possibly vomit-inducing and useless, but I still want to try it.