Hello Tomas,
Tomas Hnilica wrote:
> Hello all OpenSG users,
>
> I am new to OpenSG so sorry for "stupid" quesitons in advance.
>
> I am working on multiplatform visualization application using FOX
> library for GUI. I have an openGL window there with basic scene
> functions like lights, cameras, scene manipulations,selection... (FOX
> has such class) so I am able to display and work with my models.
> The problem is, that the models can be very large - up to 15 milions of
> tetrahedrons - and the performance is very bad - there is no
> optimization, no scenegraphs, just traversing an array of "tetrahedron
> objects" and issuing openGL commands (of course many glBegin,glEnd).
>
> So I was looking for some optimization algorithms and have found OpenSG
> which looks very very good.
>
> So I would like to ask you:
> - will openSG speed up the graphics and how? In the StarterGuide i red,
> that graph operators will do the job - minimize glBegin,glEnd commands,
> optimize the scene graph, culling and so on...
as always, the answer is "it depends" ;) If all of your scene is on
screen with little overdraw there is not much that can be done (expect
getting faster hardware). But if parts of your scene are off-screen or
there is lots of overdraw or you do many switches between a set of
materials OpenSG can improve performance - it does frustrum culling,
occlusion culling and attempts to minimize OpenGL state changes.
> - is it better to use openSG since I only need graphics optimization or
> is it better to implement such optimizaion algorithms? I thnik openSG
> offers me matured functionality for displaying text, legends, selection
> highlighting, etc...On the other hand writing optimization algorithms
> would be something like reinventing the wheel...but more customized.
Hm, hard to say for me. Maybe if you tell us what optimizations you need
we can attempt to make a guess, but in the end you will likely have to
make that decision yourself, since you know your requirements and data
structures best.
> - How is it with memory usage? Since the models are big, i must be
> memory efficient. I think the way openSG users work with data is: read
> data and put them into nodes in the tree, when picking (or working on
> other way) traverse the tree. No additinal data structures are needed.
> Am I right? Currently I have data in classes that reflect the meaning of
> the data (ie not only to display, but also to calclate something...).
> How to split the data so the original purpose will remain and not to
> double the data for openSG graphs?
At the very bottom geometry data is basically held in an std::vector,
which is wrapped in something called a OSG::GeoProperty. Now, if your
non-rendering algorithms can work on that, you don't need to duplicate
data. Otherwise it depends on what data structures you need and how
difficult it is to keep things in sync, but I can't give you any tips
there from what you have told us, sorry.
> - LOD: in starterguide there is an example where 3 models had to be
> created. Is there any chance to use such optimization like LOD that
> creates the less-detailed models itself from origin(=full detail) data?
It can be done, but nobody has implemented it.
> - I tested ParaView (based on VTK) for performance and the graphics
> speed on my models was good. Do you think openSG can have comparable
> speed? Are there any comparsions available?
Sorry, I don't know ParaView so I can't comment on this.
Hope it helps,
Carsten
-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys - and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV
_______________________________________________
Opensg-users mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/opensg-users
