Announcement

Collapse
No announcement yet.

What's the best way for sampling?

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • What's the best way for sampling?

    Hello,

    I use Vray for Rhino, so my interface isn't the same like for 3dsmax, but maybe someone could give me general suggestions.

    My problem is, that I found no way, to get a good sampling of the geometry details and a material based subdiv control.

    Example:
    * Left image - classical setup 1/4 and material subdivs at high values (f.ex. 60) - materials looks fine, but edges and geometric details are to low sampled
    * middle image - I increase the image sampler to get a better geometry sampling, but lost the surface quality. The material subdivs dons't help so much to increase the quality, only the general noise threshold.
    * right image - Vlado's universal setup, only the noise level helps to increase the quality of the whole image, but a per material control is inpossible

    What could I do? I'm searching for a way to work like at the classical setup, with material subdivs and to get higher geometric sampling? Is it possible to use a higher image sampling without to lost the material based control?

    Ciao, Micha


    www.simulacrum.de ... visualization for designer and architects

  • #2
    Micha, looking at your first image to the left, you may want to try sub-pixel sampling...

    Those hard edges are typical of highlight reflections of a very strong light source.

    Let us know.

    ciao

    gio

    Comment


    • #3
      Originally posted by Giovanni
      Micha, looking at your first image to the left, you may want to try sub-pixel sampling...

      Those hard edges are typical of highlight reflections of a very strong light source.

      Let us know.

      ciao

      gio
      I agree, we have done some thurow testing and concluded that, with subpixel mapping and color clamping off, the noise is greatly increased, because everything is being dispersed in the scene with float values, which can be very dim sometimes, almost not visible to human eye, but they do cause a lot of noise.
      Also, you dont need to raise the subdivisions that hight. Instead try lowering noise threshold.
      Our typical solution which works in animations is 1/12, 1/24 with 0.005, 0.002.
      Dmitry Vinnik
      Silhouette Images Inc.
      ShowReel:
      https://www.youtube.com/watch?v=qxSJlvSwAhA
      https://www.linkedin.com/in/dmitry-v...-identity-name

      Comment


      • #4
        Thanks Gio, attached the result with 1/4 and mat subdiv 66 and subpixel mapping. The subpixel help to get a better geometry sampling, but it looks quite random and don't reach the quality of the image sampler increase (and the render time is long).



        @Morbid Angle: Thanks, it's my experience too, that an high image sampler (this example scene looks great with 1/32) and low threshold helps. But I try to don't use the threshold option, because I would be happy, if I could control the problem effects only - a single material subdiv for critical materials or the image sampler for small details.
        Interesting is, if I use 1/32, than the classical mat subdivs don't show any effect (only very high values like 300), the noise threshold seems to be the best control for this high image sampler.

        So, it seems to be inpossible to stay at the classical setup with material subdivs, if the image sampler count is increased? Now way to increase the image sampling without to lost the local subdiv controls?

        At Vray for Rhino the QMC noise threshold and the image sampler noise threshold are permanet conected. Could it help to adjust the thresholds? I don't have any experience with independent noise threshholds. I hope, we get this control at VfR too. Could it be a way here?

        Now, I try to combine the suggestions: a little bit higher image sampling 1/6 and subpixel sampling - not bad. The light increase image sampler seems to help to increase the subpixel quality. Render time is medium, better than subpixel+1/4 only. It seems to be, the subpixel sampling like a higher image sample count.



        Morbid Angle, I found my old test with 1/32 and 0.005 at my harddisk - best quality. The price is that only one quality control options works - the noise threshold. It's a kind of universal setup.

        www.simulacrum.de ... visualization for designer and architects

        Comment

        Working...
        X