Hi Nathan, Thanks for the exploration. I'm focusing on some other work so haven't updated the shadow algorithm these time. I myself also don't satisfy with current splitting method, and am considering using SDSM to replace it. SDSM must use computer shaders to compute the light space contributed values and then change distances dynamically according to the different values. Another important update for me is to implement exponent shadow map (ESM), as well as fix more and more bugs I made to the VDSM class. :-)
Will get back soon and hope you and others could keep in touch and give me more suggestions, Thanks, Wang Rui 2013/7/11 Nathan Monteleone <[email protected]> > Partially answering my own question here -- looks like I had my coordinate > spaces mixed up (splitting is done in view space not light space), so it's > not going to be as simple of a change as I thought... Will post after I > dig more. > > -Nathan > > > > > nbmont wrote: > > Hi Rui, > > > > I wanted to get your thoughts on something from this patch. I have been > trying it out with a relatively small patch of buildings as shadow casters > on a huge earth model (custom, not osg DatabasePager based). > > > > I am using vdsm with the PARALLEL_SPLIT option. The problem I have is > that the Z split is still calculated based on the bounding sphere for the > whole shadowed scene, not just the bounds for the shadow casters. So I > usually end up with the split in a very non-optimal place; usually all the > shadows end up in one split or the other. > > > > I think I'll try writing a custom ComputeShadowSplitCallback to see if I > can fix the problem that way -- however I think it will be more > computationally efficient to just calculate the light-space bounds before > splitting, and base the splits off those. From there just pass the > calculated near/far into computeShadowCameraSettings instead of redoing it > there. > > > > Not expecting you to go change the code for me -- just want to know if > you think this is a sane approach. If I can get something working I'll see > about submitting it. > > > > Thanks, > > Nathan > > > ------------------ > Read this topic online here: > http://forum.openscenegraph.org/viewtopic.php?p=55193#55193 > > > > > > _______________________________________________ > osg-submissions mailing list > [email protected] > > http://lists.openscenegraph.org/listinfo.cgi/osg-submissions-openscenegraph.org >
_______________________________________________ osg-submissions mailing list [email protected] http://lists.openscenegraph.org/listinfo.cgi/osg-submissions-openscenegraph.org
