Re: [osg-users] laser scanner file
ok however, thanks 2008/5/22 Robert Osfield <[EMAIL PROTECTED]>: > Hi Fabio, > > Perhaps others can help you. I've already written a huge amount on > this thread. I'm afraid I have too many other tasks to keep giving > out free consultancy on performance optimization. > > Robert. > > On Thu, May 22, 2008 at 1:59 PM, fabio riot <[EMAIL PROTECTED]> wrote: >> it's mean that : >> if I move the camera near the model (viewing only same pieces of >> triangles (that is some geodes), and pruning the rest), should the >> first solution be better ? >> While in the second way, using only one geode, the frustum culling >> (not pruning geometry) is not efficent. Isn't it? >> >> But in my case when I move the camera near the model, the improvement >> in fps is the same. >> >> >> Should the first way better fot "frustum culling phase"? >>> I thought each geodes (leaf node) had own boundig volume used by the >>> OSG to chose which geometry render... >> >> >> They do, its the a quad/octr tree structure to your scene graph that >> is likely to help the cull traversal, but only if some of the scene >> graph will be culled by the cull traversal. >> >> >> >> >> 2008/5/22 Robert Osfield <[EMAIL PROTECTED]>: >>> On Thu, May 22, 2008 at 11:53 AM, fabio riot <[EMAIL PROTECTED]> wrote: >>>> Hi Robert, >>>> Going back to scene graph? >>>>> have I to create 1) a scene with one geode for each geometry? or 2) >>>>> only one >>>>> geode containing all N geometry? >>>> >>>> My question is: >>>> Why do the two solutions seem to behave in the same way? >>> >>> If they don't then cull performance isn't the bottleneck. It's also >>> not likely to make much difference to cull anyway unless you properly >>> build a quad tree of the scene graph. Large flat nodes be that >>> osg::Group or osg::Geode are inefficient to traverse. >>> >>>> Should the first way better fot "frustum culling phase"? >>>> I thought each geodes (leaf node) had own boundig volume used by the >>>> OSG to chose which geometry render... >>> >>> They do, its the a quad/octr tree structure to your scene graph that >>> is likely to help the cull traversal, but only if some of the scene >>> graph will be culled by the cull traversal. >>> >>> In you own example cull isn't likely to be the bottleneck anyway so >>> it's best to concentrate on what is the bottleneck - the sheer amount >>> of data you have and the way that you pass it to OpenGL. I've gone >>> over all of this before in this thread. >>> >>> Robert. >>> ___ >>> osg-users mailing list >>> osg-users@lists.openscenegraph.org >>> http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org >>> >> ___ >> osg-users mailing list >> osg-users@lists.openscenegraph.org >> http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org >> > ___ > osg-users mailing list > osg-users@lists.openscenegraph.org > http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org > ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] laser scanner file
it's mean that : if I move the camera near the model (viewing only same pieces of triangles (that is some geodes), and pruning the rest), should the first solution be better ? While in the second way, using only one geode, the frustum culling (not pruning geometry) is not efficent. Isn't it? But in my case when I move the camera near the model, the improvement in fps is the same. Should the first way better fot "frustum culling phase"? > I thought each geodes (leaf node) had own boundig volume used by the > OSG to chose which geometry render... They do, its the a quad/octr tree structure to your scene graph that is likely to help the cull traversal, but only if some of the scene graph will be culled by the cull traversal. 2008/5/22 Robert Osfield <[EMAIL PROTECTED]>: > On Thu, May 22, 2008 at 11:53 AM, fabio riot <[EMAIL PROTECTED]> wrote: >> Hi Robert, >> Going back to scene graph? >>> have I to create 1) a scene with one geode for each geometry? or 2) only >>> one >>> geode containing all N geometry? >> >> My question is: >> Why do the two solutions seem to behave in the same way? > > If they don't then cull performance isn't the bottleneck. It's also > not likely to make much difference to cull anyway unless you properly > build a quad tree of the scene graph. Large flat nodes be that > osg::Group or osg::Geode are inefficient to traverse. > >> Should the first way better fot "frustum culling phase"? >> I thought each geodes (leaf node) had own boundig volume used by the >> OSG to chose which geometry render... > > They do, its the a quad/octr tree structure to your scene graph that > is likely to help the cull traversal, but only if some of the scene > graph will be culled by the cull traversal. > > In you own example cull isn't likely to be the bottleneck anyway so > it's best to concentrate on what is the bottleneck - the sheer amount > of data you have and the way that you pass it to OpenGL. I've gone > over all of this before in this thread. > > Robert. > ___ > osg-users mailing list > osg-users@lists.openscenegraph.org > http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org > ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] laser scanner file
Hi Robert, Going back to scene graph? > have I to create 1) a scene with one geode for each geometry? or 2) only one > geode containing all N geometry? My question is: Why do the two solutions seem to behave in the same way? Should the first way better fot "frustum culling phase"? I thought each geodes (leaf node) had own boundig volume used by the OSG to chose which geometry render... THANKS. On Tue, May 6, 2008 at 3:32 PM, fabio riot <[EMAIL PROTECTED]> wrote: > About your advice...(how to render 1200 triangles file,ply) > Once I break the model into smaller geometry (for example N=200 > geometry blocks with 6 triangles), > have I to create a scene with one geode for each geometry? or only one > geode containing all N geometry? > > I thought the first solution was the best...but trying both I'haven't > noticed difference in frame rate Your biggest bottleneck will be the GPU, which is down to granularity and the way that you pass your data. The exact way you hang the Geometries off the scene graph won't matter too much right now as you aren't CPU limited, later on once you have the GPU side working efficient you might see CPU overhead being an issue, then we can look at optimizing the scene graph structure, but right now don't worry about it. 2008/5/9 Robert Osfield <[EMAIL PROTECTED]>: > Hi Fabio, > > On Fri, May 9, 2008 at 1:00 PM, fabio riot <[EMAIL PROTECTED]> wrote: >> Hi Robert >> Sorry, I haven't much experience in computer graphics...can you >> explain me what you mean with "GPU down in granularity"? > > I believe I wrote: > > "Your biggest bottleneck will be the GPU, which is down to granularity > and the way that you pass your data. " > > By which I mean the performance you get will mostly depend on > > 1) the granularity of the data i.e. how big Geometry chunks you use > 2) the use of VBO's, DisplayLists, glDrawElements etc. > > W.r.t granularity I recommend that you experiment with the size of > osg::Geometry (i.e. number of traingles > per Geometry) to see what sweet spot you get, it'll be driver/hardware > dependant so try it on a variety of > machines. > >> However I have tested the program in Windows XP 32 bit (precedent test >> was made in Vista 64 bit) >> and now I can use VBO and DL also for larger dataset (maybe the GPU >> driver aren't efficent in VISTA). >> The only side effect is with DL, for larger model ,some piece of >> geometry aren't rendered. > > I believe both the driver model and the driver implementations are > problematic under Vista, if you want best performance right now you > are probably best sticking to XP, or just move up to Linux, here the > compilers, OS, filesystem, threading support, memory management are > all better than any of the Windows OS's. > >> To improve frame rate is possible to use a decimate model during the >> movement (rotation ,zooming..) >> switching on complete model in motionless state? > > Just you can do this, such as by using a Switch which is manually > controlled, or perhaps a callback that monitors the view matrix and > automatically switches between low res and high res children. You'll > need to the decimentation of course, osgUtil::Simplifier could help > you out here. > > Robert. > ___ > osg-users mailing list > osg-users@lists.openscenegraph.org > http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org > ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] laser scanner file
Hi Robert Sorry, I haven't much experience in computer graphics...can you explain me what you mean with "GPU down in granularity"? However I have tested the program in Windows XP 32 bit (precedent test was made in Vista 64 bit) and now I can use VBO and DL also for larger dataset (maybe the GPU driver aren't efficent in VISTA). The only side effect is with DL, for larger model ,some piece of geometry aren't rendered. To improve frame rate is possible to use a decimate model during the movement (rotation ,zooming..) switching on complete model in motionless state? Thanks 2008/5/7 Robert Osfield <[EMAIL PROTECTED]>: > Hi Fabio, > > On Tue, May 6, 2008 at 3:32 PM, fabio riot <[EMAIL PROTECTED]> wrote: >> About your advice...(how to render 1200 triangles file,ply) >> Once I break the model into smaller geometry (for example N=200 >> geometry blocks with 6 triangles), >> have I to create a scene with one geode for each geometry? or only one >> geode containing all N geometry? >> >> I thought the first solution was the best...but trying both I'haven't >> noticed difference in frame rate > > Your biggest bottleneck will be the GPU, which is down to granularity > and the way that you pass your data. The exact way you hang the > Geometries off the scene graph won't matter too much right now as you > aren't CPU limited, later on once you have the GPU side working > efficient you might see CPU overhead being an issue, then we can look > at optimizing the scene graph structure, but right now don't worry > about it. > > W.r.t Granularity try various tests with different number of > triangles in each geometry, look for the sweet spot. > > W.r.t how you pass the data, VBO's will most likely be the best way. > Also you need to reduce the amount of memory you using - use > compressed versions of colours, share vertices as much as possible - > use osg::DrawElementUShort. > > >> beside with smaller geometry I'have noticed an improvement in >> rendering performance using VBO or IM (about 5 fps (N=200) versus 1fps >> (N=1)), >> but the best solution seem to be DL (fps 15 in all case N=1 or N=220). >> >> For larger dataset (30.000.000 triangles) the program is able to view >> the object, only using IM (switching off DL and VBO) > > What do you mean by IM? > ___ > osg-users mailing list > osg-users@lists.openscenegraph.org > http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org > ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] laser scanner file
IM=Immediate mode rendering 2008/5/7 Robert Osfield <[EMAIL PROTECTED]>: > Hi Fabio, > > On Tue, May 6, 2008 at 3:32 PM, fabio riot <[EMAIL PROTECTED]> wrote: > > About your advice...(how to render 1200 triangles file,ply) > > Once I break the model into smaller geometry (for example N=200 > > geometry blocks with 6 triangles), > > have I to create a scene with one geode for each geometry? or only one > > geode containing all N geometry? > > > > I thought the first solution was the best...but trying both I'haven't > > noticed difference in frame rate > > Your biggest bottleneck will be the GPU, which is down to granularity > and the way that you pass your data. The exact way you hang the > Geometries off the scene graph won't matter too much right now as you > aren't CPU limited, later on once you have the GPU side working > efficient you might see CPU overhead being an issue, then we can look > at optimizing the scene graph structure, but right now don't worry > about it. > > W.r.t Granularity try various tests with different number of > triangles in each geometry, look for the sweet spot. > > W.r.t how you pass the data, VBO's will most likely be the best way. > Also you need to reduce the amount of memory you using - use > compressed versions of colours, share vertices as much as possible - > use osg::DrawElementUShort. > > > > beside with smaller geometry I'have noticed an improvement in > > rendering performance using VBO or IM (about 5 fps (N=200) versus 1fps > > (N=1)), > > but the best solution seem to be DL (fps 15 in all case N=1 or N=220). > > > > For larger dataset (30.000.000 triangles) the program is able to view > > the object, only using IM (switching off DL and VBO) > > What do you mean by IM? > > ___ > osg-users mailing list > osg-users@lists.openscenegraph.org > http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org > ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] laser scanner file
Hi Robert
About your advice...(how to render 1200 triangles file,ply)
Once I break the model into smaller geometry (for example N=200
geometry blocks with 6 triangles),
have I to create a scene with one geode for each geometry? or only one
geode containing all N geometry?
I thought the first solution was the best...but trying both I'haven't
noticed difference in frame rate
beside with smaller geometry I'have noticed an improvement in
rendering performance using VBO or IM (about 5 fps (N=200) versus 1fps
(N=1)),
but the best solution seem to be DL (fps 15 in all case N=1 or N=220).
For larger dataset (30.000.000 triangles) the program is able to view
the object, only using IM (switching off DL and VBO)
Thanks
2008/4/21, Robert Osfield <[EMAIL PROTECTED]>:
> Hi Fabio,
>
>>
> As for techniques for handling large models, I'd suggest switching off
> display lists, and then breaking the model down into smaller geometry
> blocks, of something like 10,000 vertices/triangles each and use
> VBO's.
>
> Robert.
>
> On Mon, Apr 21, 2008 at 3:35 PM, fabio riot <[EMAIL PROTECTED]> wrote:
> > Hi all
> > I'm new in OSG, so I need some advices to address my work.
> > I need to visualize a large dimension laser scanner file (.ply).
> > it contains about 6.000.000 of points and 12.000.000 of triangles
> > (point's connections).
> > I'm not able to view the entire file with a single node scenegraph
> > (the program fails):
> >
> >
> > osg::Node* node = NULL
> > node = osgDB::readNodeFile("entire_file.ply"); //with my plugin
> > osg::ref_ptr root = new osg::Group();
> > root->addChild(node);
> > vieweretc
> >
> >
> > So I have broken the "laser cloud" in more little pieces
> > (at example 12 pieces each one with 500.000 points and about 1.000.000
> > triangles)
> > and I have created a scenegraph with 12 node
> >
> >
> > osg::ref_ptr root = new osg::Group();new osg::Group();
> > root->addChild(node1);
> > root->addChild(node2);
> >
> > .
> >
> > root->addChild(node12);
> >
> > where each node contains one pieces of data
> >
> > In this way I can load and view the all strucure...but the loading
> > time is very very long (60 min)
> >
> > I'would like to know if this is the correct way to tackle my problem
> > or if there's a clever way...Thanks
> > ___
> > osg-users mailing list
> > osg-users@lists.openscenegraph.org
> > http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
> >
> ___
> osg-users mailing list
> osg-users@lists.openscenegraph.org
> http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
>
___
osg-users mailing list
osg-users@lists.openscenegraph.org
http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] laser scanner file
Hi Brian
Thanks for your answer
I have tested your code
On my hardware (Quadro FX 3500M, 2Gb Ram ..Windows Vista, OSG (64bit compiled))
The frame rate is 7 fps (80% memory using)
But in my case loading ply file..the main problem isn't the rendering
but the loading time (that takes hours...), the file's size is about
250Mb.
May be I wrong in writing my ply file plugin...
it reads from ply file and sets a geometry class:
osg::ref_ptr geode = new osg::Geode();
osg::ref_ptr geometry = new osg::Geometry();
geode->addDrawable(geometry.get());
osg::ref_ptr colors=new osg::Vec4Array();
osg::ref_ptr verices = new osg::Vec3Array;
geometry->setVertexArray( vertices.get() );
geometry->setColorArray(colors.get());
geometry->setColorBinding(osg::Geometry::BIND_PER_VERTEX);
for(i=0;ipush_back( osg::Vec3( x, y, z) );
colors->push_back(osg::Vec4(r,g,b,1.0f));
}
for(i=0;i face=new osg::DrawElementsUInt
(osg::PrimitiveSet::TRIANGLES, 0);
face->push_back(index1);
face->push_back(index2);
face->push_back(index3);
geometry->addPrimitiveSet(face.get());
}
2008/4/21, Brian R Hill <[EMAIL PROTECTED]>:
> It's going to be dependent on your hardware and the osg structures you use.
>
> Try running this and the bring up the grid.ive in osgviewer. It generates
> 9.000.000 verts, 18.000.000 tris. On my box Win XP 4Gb Ram, Quadro FX 4600,
> it renders at 20 hz, using 1.3 Gb memory.
>
> Brian
>
>unsigned int num = 3000;
>osg::HeightField* grid = new osg::HeightField;
>grid->allocate(num,num);
>grid->setXInterval(1);
>grid->setYInterval(1);
>
>for(unsigned int rr=0;rr{
>for(unsigned int cc=0;cc{
>grid->setHeight(cc,rr,sin((float)(cc+rr)));
>}
>}
>osg::Geode * geode = new osg::Geode;
>geode->addDrawable(new osg::ShapeDrawable(grid));
>
>osgDB::writeNodeFile(*geode,"grid.ive");
>
>
> [EMAIL PROTECTED] wrote: -
>
>
> To: osg-users@lists.openscenegraph.org
> From: "fabio riot" <[EMAIL PROTECTED]>
> Sent by: [EMAIL PROTECTED]
> Date: 04/21/2008 10:35AM
> Subject: [osg-users] laser scanner file
>
> Hi all
> I'm new in OSG, so I need some advices to address my work.
> I need to visualize a large dimension laser scanner file (.ply).
> it contains about 6.000.000 of points and 12.000.000 of triangles
> (point's connections).
> I'm not able to view the entire file with a single node scenegraph
> (the program fails):
>
>
> osg::Node* node = NULL
> node = osgDB::readNodeFile("entire_file.ply"); //with my plugin
> osg::ref_ptr root = new osg::Group();
> root->addChild(node);
> vieweretc
>
>
> So I have broken the "laser cloud" in more little pieces
> (at example 12 pieces each one with 500.000 points and about 1.000.000
> triangles)
> and I have created a scenegraph with 12 node
>
>
> osg::ref_ptr root = new osg::Group();new osg::Group();
> root->addChild(node1);
> root->addChild(node2);
>
> .
>
> root->addChild(node12);
>
> where each node contains one pieces of data
>
> In this way I can load and view the all strucure...but the loading
> time is very very long (60 min)
>
> I'would like to know if this is the correct way to tackle my problem
> or if there's a clever way...Thanks
> ___
> osg-users mailing list
> osg-users@lists.openscenegraph.org
> http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
>
> This is a PRIVATE message. If you are not the intended recipient, please
> delete without copying and kindly advise us by e-mail of the mistake in
> delivery. NOTE: Regardless of content, this e-mail shall not operate to
> bind CSC to any order or other contract unless pursuant to explicit written
> agreement or government initiative expressly permitting the use of e-mail
> for such purpose.
>
>
> ___
> osg-users mailing list
> osg-users@lists.openscenegraph.org
> http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
>
___
osg-users mailing list
osg-users@lists.openscenegraph.org
http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] laser scanner file
Hi Robert,
My problem is that when I load the entire file.ply, the rendering has
a good frame rate, but the loading time take a long while (hours)...
hardware: nvidia quadro FX 3500M, RAM 2Gb
osg version 2.2.0
file dimension (.ply) is about 250Mb (6.000.000 points; 12.000.000 triangles)
My own ply loader is very simple, it reads in file.ply and loads
vertices and faces in geometry class:
osg::ref_ptr geode = new osg::Geode();
osg::ref_ptr geometry = new osg::Geometry();
geode->addDrawable(geometry.get());
osg::ref_ptr colors=new osg::Vec4Array();
osg::ref_ptr verices = new osg::Vec3Array;
geometry->setVertexArray( vertices.get() );
geometry->setColorArray(colors.get());
geometry->setColorBinding(osg::Geometry::BIND_PER_VERTEX);
for(i=0;ipush_back( osg::Vec3( x, y, z) );
colors->push_back(osg::Vec4(r,g,b,1.0f));
}
for(i=0;i face=new osg::DrawElementsUInt
(osg::PrimitiveSet::TRIANGLES, 0);
face->push_back(index1);
face->push_back(index2);
face->push_back(index3);
geometry->addPrimitiveSet(face.get());
}
However I'll try to follow your advices (using vbo) as soon as I can
Thanks
2008/4/21, Robert Osfield <[EMAIL PROTECTED]>:
> Hi Fabio,
>
> You say the problem fails.. but in what way?
>
> What hardware, OS, OSG versions are you working with?
>
> The OSG doesn't have a ply loader itself, is this your own loader.
>
> As for techniques for handling large models, I'd suggest switching off
> display lists, and then breaking the model down into smaller geometry
> blocks, of something like 10,000 vertices/triangles each and use
> VBO's.
>
> Robert.
>
> On Mon, Apr 21, 2008 at 3:35 PM, fabio riot <[EMAIL PROTECTED]> wrote:
> > Hi all
> > I'm new in OSG, so I need some advices to address my work.
> > I need to visualize a large dimension laser scanner file (.ply).
> > it contains about 6.000.000 of points and 12.000.000 of triangles
> > (point's connections).
> > I'm not able to view the entire file with a single node scenegraph
> > (the program fails):
> >
> >
> > osg::Node* node = NULL
> > node = osgDB::readNodeFile("entire_file.ply"); //with my plugin
> > osg::ref_ptr root = new osg::Group();
> > root->addChild(node);
> > vieweretc
> >
> >
> > So I have broken the "laser cloud" in more little pieces
> > (at example 12 pieces each one with 500.000 points and about 1.000.000
> > triangles)
> > and I have created a scenegraph with 12 node
> >
> >
> > osg::ref_ptr root = new osg::Group();new osg::Group();
> > root->addChild(node1);
> > root->addChild(node2);
> >
> > .
> >
> > root->addChild(node12);
> >
> > where each node contains one pieces of data
> >
> > In this way I can load and view the all strucure...but the loading
> > time is very very long (60 min)
> >
> > I'would like to know if this is the correct way to tackle my problem
> > or if there's a clever way...Thanks
> > ___
> > osg-users mailing list
> > osg-users@lists.openscenegraph.org
> > http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
> >
> ___
> osg-users mailing list
> osg-users@lists.openscenegraph.org
> http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
>
___
osg-users mailing list
osg-users@lists.openscenegraph.org
http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
[osg-users] laser scanner file
Hi all
I'm new in OSG, so I need some advices to address my work.
I need to visualize a large dimension laser scanner file (.ply).
it contains about 6.000.000 of points and 12.000.000 of triangles
(point's connections).
I'm not able to view the entire file with a single node scenegraph
(the program fails):
osg::Node* node = NULL
node = osgDB::readNodeFile("entire_file.ply"); //with my plugin
osg::ref_ptr root = new osg::Group();
root->addChild(node);
vieweretc
So I have broken the "laser cloud" in more little pieces
(at example 12 pieces each one with 500.000 points and about 1.000.000
triangles)
and I have created a scenegraph with 12 node
osg::ref_ptr root = new osg::Group();new osg::Group();
root->addChild(node1);
root->addChild(node2);
.
root->addChild(node12);
where each node contains one pieces of data
In this way I can load and view the all strucure...but the loading
time is very very long (60 min)
I'would like to know if this is the correct way to tackle my problem
or if there's a clever way...Thanks
___
osg-users mailing list
osg-users@lists.openscenegraph.org
http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
