Announcement

Collapse
No announcement yet.

Pixel shimmer hell

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

  • Pixel shimmer hell

    Hi folks,

    I'm doing an animation at the moment with a load of really detailed trees in ity where I start out wide and fly into the center of the trees so they have to be detailed enough to look okay up close - My problem is that the trees shimmer like bastards when I'm out wide - I know that it's a pixel creep issue but I can't seem to get rid of it- I've tried using area, mitchel and video and none of them avoid the problem. Even if I change the lighting so that it's only a single direct light with no shadows (I.e it's not a gi sampling issue - it's more of an image sampling issue) and I can't seem to get rid of the problem. At the moment we have motion blur turned down to a small amount 0.1 / 0.2 to match the live action plate it's going in to so hopefully we can sort it out without having to make the motion blur duration longer.

    Just so I'm goind the right way with the qmc sampler, I'm using a really small noise threshold (around 0.001) and the adaptive amount is at 0.99 - is this the correct way to go for tiny detail?

    Here's all of the settings:



    And there's a quicktime movie of the current tree element here

    http://www.joconnell.com/stuff/trees_vray.mov - unfortunately it's 19 megs cos I didn't want to introduce more noise again. You can see that theres a tonne of pixel creep and shimmering going on.

    If anyone can make any suggestions as to how to get rid of this it'd be very much appreciated.

  • #2
    well luckly for you this has been brought up before and there is a solution:

    1 - use adaptive subdivision with 0/2, 0/3 sampling and object outline and normals checked.
    2 - dont use sharpening filters like mitchel, rather use quadratic

    this should work. Let me know if you have more issues. I suggest rendering few frames without motion blur to see if it will.

    The reason for this, is for some reason qmc sampler cannot process such complex detail. But this is not only vray, mental ray has also such issue, but the sampling there is a bit different.
    The other solution would be to render double or triple the size with lower settings then shrink the footage down in post.
    Dmitry Vinnik
    Silhouette Images Inc.
    ShowReel:
    https://www.youtube.com/watch?v=qxSJlvSwAhA
    https://www.linkedin.com/in/dmitry-v...-identity-name

    Comment


    • #3
      Also your QMC settings are pretty low which will cause lots of noise.
      Natty
      http://www.rendertime.co.uk

      Comment


      • #4
        Excellent I'll go with those today - what's kind of irritating is that I shouldn't have to double or triple size it - surely the oversampling in the anti aliaser should allow me to do the same thing? I suppose doing that equates to having normal adaptive aa in the main render and then having the oversized render doing a fixed aa on top of that?

        I've never had a render like this using vray so you're right - my understanding of the qmc sampler was wrong. Cheers for the advice.

        Comment


        • #5
          Okay I've tried a lot of these options and they're giving a bit of an improvement but still not totally clicker free - at the moment I'm using adaptive subdivision 1,3 with a 0.001 threshold, outline and normal turned on again with a 0.001 threshold. I'm using the quadratic filter as suggested. For the (r)qmc sampler rollout, I have the adaptive amount at 0.6, the noise threshold at 0.001, the min samples at 48 and the global samples multiplier at 1.0.

          The adaptive settings seem to get me to a point and then the next stage seems to be doubling the render size and rescaling in after effects. The results look okay but when I go to a broadcast monitor I still get shimmer. I'm trying a triple size render now which'll hopefully sort a bit more but am I doing anything daft in my settings? I'd imagine stuff like normal threshold really won't make that much of a difference, more stuff like the adaptiove subdision threshold and the qmc sampler adaptive amount / noise threshold.

          Anything obvious i'm doing stupidly?

          Comment


          • #6
            Try up-ing your Global Subdivs to 2 or even higher.
            Natty
            http://www.rendertime.co.uk

            Comment


            • #7
              Will do - cheers for that.

              Would this be the same as upping the base level aa? As in using adaptiove subdivision with 1,3 instead of 0,3? I think the problem is really coming from the geometry being so detailed rather than anything to do with lighting or material settings.

              Thanks a million for the replys - they've helped a huge amount

              Comment


              • #8
                If only your broadcast monitor is flickering apply a 0.5 pixel vertical blur in post. This is because some details are only visible in one field and not in the other: with this blur you always keep them visible in both fields. In the posted quicktime it looks like one node in the network does not load an irr.map correctly: at the end some frames have darker leaves/fruit !!

                As already mentioned: adaptive sampling (0..3) with object edges and normals checked should work OK. Another way is to bake diffuse illumination and save this in the proxy-mesh. Calculate the irradiance maps with low-res proxies (for diffuse shadows) and replace them by the hi-res baked tree in the final pass !! With large meshes it is better to make flattened uv-maps for the baking in other program, because flattening a 100k poly mesh in Max can take ages

                Comment


                • #9
                  It's definitely a broadcast issue - I'm not even using gi in this example, just a simple omni light. The render has around 75 million polygons since you need to get in really close to the trees so it's suicide to try and use gi - going with nasty amounts of adaptive sampling in this case.I think it's that the details are so small that you just get pixel creep.

                  Comment


                  • #10
                    I know nothing. That said, could it be a bitmap filtering issue?
                    Surreal Structures
                    http://surrealstructures.com/blog

                    Comment


                    • #11
                      To check that I even stripped the scene back to one with no textures at all - it's pure geometry at this stage and all of the variation is coming from the leaves being at different angles - it's just a limitation of broadcast really. I've also won the all time estimated render time in work - 33 hours on a dual amd 2400 with 2 gigs of ram for a scene with 75 million polys, one direct light with a vrayshadow (non area) and one vray dome light. That said the trees only take up a third of the screen so it's more like 10 hours a frame - Back to shadow maps so

                      Comment


                      • #12
                        You didn't do this for a Cider Commercial just being shown in the uk did you?
                        Saw something similar last night on tv and remembered this... looked good!
                        well done if this was you?

                        Comment

                        Working...
                        X