This is beyond awesome.
2 Likes
So you need deformation for your left nut? Is your right one already deformed?
6 Likes
That is indeed snazzy lookin’. I watched without audio, are there better performance comparisons than an arbitrary n frames per second?
There’s various solutions like this employed by disney, pixar etc - it’s much more useful for film where that close up realism is required. Still, I guess it’ll be good in games in 5-10 years. Right now there’s not much point blowing so much of the budget for calculating this on current-new-gen for something that’s very hard to detect in a realtime game 
Wow, during my masters course I would spend ages trying to get the deformations right through weight painting and deformers. It would be so awesome to have something like this work in Blender, let alone Unity
very impressive XD
I would definitely give your left nut to have this in unity now
2 Likes
You realize you can only make such a decision twice in your life?
3 Likes
The PDF of the paper gives some better performance comparisons. The “armadillo” model, for example, gets 86 fps using dual quaternions and only 3 fps using their method. The hand fairs a bit better with 95 vs 15 fps. The real-time examples they show (“Carl” and “Dana”) both get 800+ fps using dual quaternion.
There is also substantial memory overhead–looks like roughly half a megabyte per bone. It needs 32.5MB for both Dana and Carl.
It’s definitely feasible for certain parts of certain types of games, but not something that would be widely useful yet. It looks pretty neat, but I would probably rather keep all of my testicles.
Be careful what you wish for.
You have 2 left nuts?
3 Likes
An FPS measurement isn’t really useful, though. How about milliseconds doing just their unique stuff, and on what platform under what performance characteristics? And, how does it scale?
My thoughts are that it’s most likely to be useful for close-up, high detail stuff - lockpicking springs to mind as a potential application. So the fact that you can only do a hand at acceptable interactive frame rates might be ok.
Yeah, but that seems to be all they have in the paper and I imagine their initial goal was just the feasibility of the method. Broadly, there is a fairly significant performance difference.
They seem to be advertising a newer version on their site so they likely have done more optimizations and performance testing since then. If anyone contacted them with interest in licensing it, they would probably have some newer numbers.
It seems we are in the need of a right nut as well to get better performance.