It's an academic presentation. No doubt it will eventually find its way into a shipping product though, just like content-aware resizing did a couple years ago.
In academia, Photoshop is hard to use and making a 3D model match up to the photograph is easy. In the real world I think that remains to be seen.
In any case, I do think that they should just cut the middeman and just put that blue 3d thing on the cover of the magazine. Make it skinny as hell, all "Bikini Ready By June With The 14 Day Diet BLUE SKELETAL MODEL REV 0.4.2010b Swears By!"
I can see other uses for the pose-matching technique (such as texturing models and the like), although I get the impression that this wasn't the novel part of this research.
It also seems like the 3D model part of this is kind of going the long way around. Why not just attach weights to the image for deformation purposes, instead of matching a 3D model to the image and then attaching weights to that?
This is probably related to why I quit my PhD in computer graphics and decided to get a real job instead.
Is it me or does the set up seem to take as much time as just manually painting the damn differences in? Computer Arts: automation is just more busy work in disguise.