[vtkusers] Render function's in-deterministic performance

[Navneeth Subramanian]

Hi Siddharth,
Perhaps you are seeing a race-condition between the camera capture
thread & the vtk render thread ?
Can you force call ->Modified() on the vtkImageData object at the end
of the camera refresh to see if that helps.

best,
navneeth

> ------------------------------
>
> Message: 16
> Date: Tue, 6 Dec 2011 19:35:52 +0530
> From: Siddharth Vikal <siddharthvikal at gmail.com>
> Subject: [vtkusers] Render function's in-deterministic performance
> To: vtkusers at vtk.org
> Message-ID:
>        <CADM22yxyGDvhiWBogiqTV7M5B_ZExryetZCQzSxPn6jwGRhjPw at mail.gmail.com>
> Content-Type: text/plain; charset="iso-8859-1"
>
> Hi,
>
> Let me first describe the set-up. I'm using a quad core, linux box with vtk
> 5.4. I'm using Qt for GUI, thus QVTKWidget. I'm trying to visualize image
> frames (1280 x1280 pixels) being captured from a camera device. The
> anticipated frame rate is 30 fps. I've 'capture frame' functionality
> happening in a separate thread. A minimal vtk pipeline is set up using
> vtkImageData, vtkImageMapper, vtkActor2D. Each time a new frame arrives,
> update is callled on the vtk pipeline and the main thread running Qt, is
> notified to perform render operation using paint event of QVTKWidget. For
> synchronization, mutex is used. This setup seems to work just fine except
> from the performance predictability standpoint.I've put timers in place to
> measure time taken for each step.
>
> The problem is, the paint call (or eventually 'render' call) is not
> behaving deterministically. For some frames, render completes within 6-10
> milliseconds, but for others, inexplicably, it starts taking 50-70
> milliseconds!! Now with new frame arriving at every 33 milliseconds (30
> fps), I start loosing frames. I've tried the setup using vtkImageActor
> (therefore without vtkImageMapper, vtkActor2D) as well, but with similar
> behavior.
>
> I don't understand why is render taking 60-70 milliseconds? The data is
> ready to be rendered. The in-deterministic aspect is even more worrying.
>
> Can someone please help? It'll be greatly appreciated.
>
> regards
> Siddharth
> -------------- next part --------------
> An HTML attachment was scrubbed...
> URL: <http://www.vtk.org/pipermail/vtkusers/attachments/20111206/339c44b8/attachment-0001.htm>
>