[vtkusers] Large number of actors

[Clinton Stimpson]

On Wednesday, June 08, 2011 03:48:38 pm Nicolas Rannou wrote:
> Hi Clinton,
> 
> I've been reading about the vtkMultiBlockDataset but I am not quite sure
> about how it could improve the performance.

Its an alternative similar to
vtkPolyData->vtkStripper->vtkAppend->vtkMapper->vtkActor

vtkPolyData->
vtkPolyData->  vtkMultiBlockDataSet -> vtkCompositePolyDataMapper2 -> vtkActor
vtkPolyData->
...

The difference is you don't do a vtkAppend which duplicates the data and 
increases your memory usage.

> 
> From my understanding, it would store/order the polydatas in a tree
> structure. But the main issue are the actors and mappers... (not the
> polydata itself...) After storing my polydatas within
> vtkMultiBlockDataset, I still have to generate N actors from my N
> polydatas which belong to my block of interest...
> 
> I tried to create 1 block per polydata.
> 
> 1-add polydatas in the multi block dataset using
> vtkMultiBlockDataset::AddBlock()
> 2-extract/append polydatas
> vtkCompositeDataGeometryFilter or vtkMultiBlockDataset ::GetBlock()
> 3-map it and visualize it through 1 actor and one mapper.
> 
> vtkCompositeDataGeometryFilter generates 1 polydata from all polydatas,
> which makes picking impossible... vtkMultiBlockDataset ::GetBlock() will
> return 1 polydata, which will not improve the performance since we have to
> create 1 mapper and 1 actor for each polydata present in this block.
> 
> Is there a better practice to render all polydatas which belong to a block
> through multiple actors. Am I missing something?

If you're going to use multiple actors, then don't use a vtkMultiBlockDataSet.

If you really want multiple actors, you could look at vtkMapper::SetStatic to 
speed up the multiple actors/mappers.
But then you'll have to do extra work if pipeline changes need to be accounted 
for.

Clint

> 
> Thanks
> 
> On Jun 8, 2011, at 3:11 PM, Clinton Stimpson wrote:
> > On Wednesday, June 08, 2011 01:00:48 pm Nicolas Rannou wrote:
> >> Hi all,
> >> 
> >> Our application requires to render up to 5000x1000 PolyDatas.
> >> Each polydata is made of 500 triangles.
> >> 
> >> It looks like it is impossible to store all the related actors in memory
> >> (on a machine with 48Gb RAM).
> >> 
> >> We have been playing around with things such as "vtkStripper",
> >> "ImmediateRendering", "ReleaseDataFlag", "BackFaceCulling", etc.
> >> 
> >> For 1000 PolyDatas:
> >> 
> >> The best option in term of memory looks like turning
> >> "ImmediateRendering" ON. -> 1000 actors
> >> 
> >> 	-> 60 Mb of memory usage
> >> 	-> slow interaction
> >> 
> >> The best option in term of interaction seems to be using the following
> >> pipeline: vtkPolyData->vtkStripper->vtkAppend->vtkMapper->vtkActor
> >> 
> >> 	->1 actor
> >> 	-> 90mo in memory
> >> 	-> good interaction
> >> 
> >> Using only one actor is not an option for us since we want to be able to
> >> pick and change properties of each object (individually).
> >> 
> >> Would somebody have any advise?
> > 
> > In this situation, I think you can get both interaction speed and memory
> > savings by putting your many vtkPolyData objects into a
> > vtkMultiBlockDataSet and rendering that with an actor or a few actors.
> > 
> > You could put some properties, such as colors, in the vtkPolyData instead
> > of the vtkActor.  And for other properties, you could use the same
> > actor/mapper for those objects that share the properties.  Or perhaps
> > use vtkPainter to customize rendering of these objects.

-- 
Clinton Stimpson
Elemental Technologies, Inc
Computational Simulation Software, LLC
www.csimsoft.com