I first saw this technique here, note the date: http://blog.wolfire.com/2009/11/volumetric-heat-diffusion-skinning/
The Autodesk implementation has a better voxelization and weight falloff solution though. Having implemented this technique before, the most difficult part it to get a reasonable performance out if it. On Tuesday, March 18, 2014, Raffaele Fragapane <raffsxsil...@googlemail.com> wrote: > http://www.dgp.toronto.edu/~mdelasa/voxelization/ <- proposal and video > http://www.delasa.net/data/sca2013_voxelization.pdf <- actual paper > > It's not particularly hard conceptually, but personally I wouldn't do it > in ICE. It'd likely be below acceptable performance for a large enough set > of voxels, and you'd be in trouble to bring GPGPU performance into ICE > nodes if you were willing to write one (I got really flaky results > personally, but it's not impossible). > > It's not terribly hard to do in C++ nor to thread it decently in CUDA, > there are some very obvious potential kernels to it, and plenty stuff you > could start from for OCTrees structures, sorting, walking and so on already > available to not re-invent the wheel. > > You are missing a (moderately) clever bit though as you need to find > boundaries and geodesic distance, if you just walked outwards from a bone > inclusive voxel indiscriminately you would be only marginally better off > than if you simply used plain geometric distance methods, but it's not an > order of magnitude more complex than what you'd think of by intuition. > > It's not really a hugely impressive feature, don't get me wrong, but it's > not a day's worth of work to offer a proper generic implementation that > will scale to respectable complexity. > > > On Wed, Mar 19, 2014 at 1:43 PM, Bk > <p...@bustykelp.com<javascript:_e(%7B%7D,'cvml','p...@bustykelp.com');> > > wrote: > >> Unless there is some clever part that Im missing, It's not a hard thing >> to implement. Fill a volume with voxels. Spread out( iterate) influence >> from bones across the volume particle to particle by neighbours. >> Get closest voxel to surface and find resulting influence for each bone. >> Normalise and write to envelope . >> I'd be bold enough to suggest it could be done in ICE in a day. >> >> >> On 19 Mar 2014, at 02:35, Eric Thivierge >> <ethivie...@gmail.com<javascript:_e(%7B%7D,'cvml','ethivie...@gmail.com');>> >> wrote: >> >> Unfortunately have to agree with Luc-Eric on this point. Blender probably >> has a heat map algo or something. I haven't seen the geodesic anywhere but >> the AD research vid from around last Siggraph. >> >> -------------------------------------------- >> Eric Thivierge >> http://www.ethivierge.com >> >> >> On Tue, Mar 18, 2014 at 10:31 PM, Luc-Eric Rousseau >> <luceri...@gmail.com<javascript:_e(%7B%7D,'cvml','luceri...@gmail.com');> >> > wrote: >> >>> On Tue, Mar 18, 2014 at 3:05 PM, Paulo César Duarte >>> <paulocdua...@gmail.com<javascript:_e(%7B%7D,'cvml','paulocdua...@gmail.com');>> >>> wrote: >>> > Agree, and the geodesic voxel binding skin algorithm, Blender already >>> have >>> > at least 1 year ago or more. In other words, no innovation, only >>> > implementation of existing tools. >>> >>> Got a link to that? Geodesic voxel binding is research by Autodesk. >>> >>> >> > > > -- > Our users will know fear and cower before our software! Ship it! Ship it > and let them flee like the dogs they are! > -- Technical Director @ DreamWorks Animation
