Hi Sebastian, > De: Sebastian Messerschmidt > > I know this question has been asked before, but I didn't extract > enough information out of the answers to understand it fully. > What I want to achive seems simple. I have separate renderpasses with > the same viewpoint to render different parts of the scene to different > rendertargets. > For example I need to split the opaque parts from the transparent/ > translucent ones. > Right now, I'm using nodemasks on the cameras an respective parts of > the scene. Unfortunally this brings two major problems: > 1. Everything has to traversed/culled more by every camera. So if the > scenegraph is large and complex with node masks only near the leaves, > the accumulated cull time will be a problem. > 2. If a geode has transparent and opaque parts it is impossible to do > it correctly, as drawables are "second" class nodes i.e. don't carry > nodemasks. > > So right now I'm splitting up geodes with different renderbins an > reorder them. Naturally this can have a huge impact on the scene, as > it get cluttered at the bottom effectively cutting my framerate in > certain scenes to half and below. > The idea I had in the last day would like this (in a rather abstract > way): > > Cull the scene once and let different cameras use different > renderbins to render to their target. > > > > Now my problem is, that I don't know where to start with this, or if > there are some things in OSG that might even allow this naturally.
I am currently converting FlightGear to deferred rendering. See http://wiki.flightgear.org/Project_Rembrandt for pointers and a log of the conversion (and videos) To address this particular problem, I have a first camera that traverse the scenegraph to initialize the MRT buffers and another camera that is used to render lighting. Both have the same view and projection matrix and share the same depth buffer. Ambient light, sun lighting, fog and bloom are rendered with additive blending by drawing screen aligned quads while point and spot lights are rendered like normal geometry (the light volume is then drawn). So I can remove transparent bins from the renderstage of the first camera and inject them in the renderstage of the second. I do that with the code below : As a global variable : osgUtil::RenderBin::RenderBinList savedTransparentBins; In the cull callback of the geometry camera (the one that traverse the scenegraph) : virtual void operator()( osg::Node *n, osg::NodeVisitor *nv) { osgUtil::CullVisitor* cv = static_cast<osgUtil::CullVisitor*>(nv); osg::Camera* camera = static_cast<osg::Camera*>(n); savedTransparentBins.clear(); cv->traverse( *camera ); // Save transparent bins to render later osgUtil::RenderStage* renderStage = cv->getRenderStage(); osgUtil::RenderBin::RenderBinList& rbl = renderStage->getRenderBinList(); for (osgUtil::RenderBin::RenderBinList::iterator rbi = rbl.begin(); rbi != rbl.end(); ) { if (rbi->second->getSortMode() == osgUtil::RenderBin::SORT_BACK_TO_FRONT) { savedTransparentBins.insert( std::make_pair( rbi->first, rbi->second ) ); rbl.erase( rbi++ ); } else { ++rbi; } } } In the cull callback of the camera that should display transparent objects : virtual void operator()( osg::Node *n, osg::NodeVisitor *nv) { simgear::EffectCullVisitor* cv = dynamic_cast<simgear::EffectCullVisitor*>(nv); osg::Camera* camera = static_cast<osg::Camera*>(n); cv->traverse( *camera ); // .... // Render saved transparent render bins osgUtil::RenderStage* renderStage = cv->getRenderStage(); osgUtil::RenderBin::RenderBinList& rbl = renderStage->getRenderBinList(); for (osgUtil::RenderBin::RenderBinList::iterator rbi = savedTransparentBins.begin(); rbi != savedTransparentBins.end(); ++rbi ){ rbl.insert( std::make_pair( rbi->first + 10000, rbi->second ) ); } info->savedTransparentBins.clear(); } For this to work correctly, we need the contribution of artists to flag the really transparent objects and put them in transparent bins. The other (non flagged) objects are put unconditionally in the normal RenderBin (with alpha test enabled) because many part of a model are opaque but use a texture atlas with alpha channel and the transparency detection from the loader plugging is not reliable enough, resulting in a lot of false positive (opaque object treated as transparent and incorrectly lit). HTH Regards, -Fred _______________________________________________ osg-users mailing list [email protected] http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
