Announcement

Collapse
No announcement yet.

About noise in general

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

  • About noise in general

    I don't know if it's just me but I've noticed on a number of occasions now that noise when using brute force GI seems to be getting a lot harder to get rid of in 2.0 than it used to be. It's hard to put one's finger on it and it took me all this time to become convinced that there was an issue there.

    Some old scenes of mine that used to render silky smooth in a reasonable amount of time are no longer doable in an acceptable time frame and come out a lot noisier (and with longer render times) with the same settings. Just the other day, I re-rendered an image I wasn't satisfied with (lots of tiny details, glossy reflections everywhere, complicated light paths) with much higher settings and, despite about 20 hours of rendering, came out nearly unchanged.

    I read everything about the environment variable, tried it, and it hasn't made a difference (then again, I never clamp my renders so I don't know if it would.)

    This is becoming an issue because with the kinds of render times I'm seeing, we are moving into the region of unbiased renderers. I fear the image I mentioned earlier would have come out cleaner from Maxwell given the same amount of time. And I would hate to have to use IR/LC again as I love the crispness of GI shadows under BF.

    Of course, it's very likely I'm the one doing something wrong. So if anyone has grappled with this or also thinks there is a fundamental difference between 1.5 and 2.0 in respect to noise, please let me know how you are dealing with it.
    Check my blog

  • #2
    Hey there,

    here's some info on the DMC changes:

    http://www.spot3d.com/vray/help/200R...ate_to_1_5.htm

    Regards,
    Thorsten

    Comment


    • #3
      Thanks Thorsten,
      Yes, I know that. I've created the environment variable but it doesn't solve my problem (I don't use subpixel mapping anyway so I think it may not be relevant).
      Check my blog

      Comment


      • #4
        I've had similar results. I did a render recently where I had to render the couches basically gray and color correct them in post because when they were black, the noise was so bad I couldn't work with it. It really didn't matter how much light or subdivs I threw at it either. I was also not using subpixel mapping.

        The biggest fix I found was using the gamma in the color mapping. At first I didn't think I needed it because I had it setup in max already, but in the end it was the fix I needed. When I used the max gamma at input 2.2 and output 1.0, and used the vray color mapping gamma at 2.2 then most of my woes went away. Previously I have never used the color mapping gamma. Naturally if you use it you don't want to hit the srgb button on frame buffer.

        The downside is now all my images look wrong when I bring them into the RAM player, no matter what overrides I use. I have to preview frames in after effects or things just look.... very very odd. I should really take a look at PDplayer but just haven't had the time since 2.0 came out.
        Last edited by Deflaminis; 24-05-2011, 12:34 AM.

        Comment


        • #5
          The trick is to use gamma 2.2 in color mapping and checking "dont affect colors". That way you render will look same as earlier, but it will be sampled as gamma 2.2 -> cleaner dark areas.
          Lasse Kilpia
          VFX Artist
          Post Control Helsinki

          Comment


          • #6
            Thanks guys. I'm already using a 2.2 gamma (LWF) and don't affect colours (I tried also to untick don't affect colours and SRGB button - no difference). Also, I don't seem to have a specific problem with dark areas. It's noise pretty much everywhere. I can just about eliminate it, but I have to boost the settings much higher than I used to in order to achieve the same result. Hence the 20-hour-plus render times I'm seeing at the moment (for 2K images too, not gigantic renders).

            EDIT, just an example, the image below rendered for 19 hours at this resolution (Max 2012, Vray 2.0, LC+BF, gamma 2.2, don't affect colours, SRGB on). From memory, the DMC sampler had subdivs of 40 and a colour threshold of 0.002. Noise threshold 0.002. I rendered another version with the environment variable = 1 but the result and render time were exactly the same. I've also made dozens of tests trying to raise and lower the values for the DMC subdivs, noise and colour thresholds, Brute Force subdivs... no luck. It's like hitting a glass ceiling.

            The image is heavily processed but you can see the heavy noise all over the place, including light areas, and is just not good enough in my opinion. I know I must be doing something wrong as I seem to be virtually alone in having this problem, but whatever it is, it used to work fine). And it's not just this scene either. Every interior right now is showing this problem as soon as I hit a certain amount of geometry/bouncing lights.

            Last edited by BBB3; 24-05-2011, 02:20 AM.
            Check my blog

            Comment


            • #7
              I see, that really is noisy.

              Are you using envFog also? Could that be the reason?
              Lasse Kilpia
              VFX Artist
              Post Control Helsinki

              Comment


              • #8
                I must be honest, the image is grainy yes, but it's way less drastic than I thought it would be based on your previous posts.. of course it's a matter of personal taste, and I can see why it wouldn't work for you.

                quite apart from that, have you tried to deal with the problem locally instead of relaying entirely on the DMC sampler? I used to have a kind of "universal settings" approach as well, leaving any local setting alone (lights subds, mats subds, bf and so on), and clearing all the noise only with the aa. in some cases though, it was much more efficient to tackle specific problematic areas such as the indirect illumination and taking load off the DMC sampler. render times can be more manageable as well.

                Comment


                • #9
                  Larsson: No I'm not. There is a fog pass but it was rendered separately against a black override material (and was pretty much noise free)

                  Hi Rivoli: Yes I did, especially by trying to put more of the load on the Brute Force subdivs. But somehow all the various approaches bring me to the same place: much longer time than I'm used to in order to reach a noise-free render. It's as though everything was working like before except for the fact that you need an extra 10 hours of render time.
                  Last edited by BBB3; 24-05-2011, 03:00 AM.
                  Check my blog

                  Comment


                  • #10
                    Have you tried playing with precalcing an IRmap/BF setup with reflections turned off in your scene and the quality set to high/very high? Use the VMC script to do this, then switch them back on for the final render.

                    An image at that resolution taking over 30 minutes is too long. 19 hours is insane.

                    Comment


                    • #11
                      Thanks cubicle. I'm not sure how I would do that. The BF is not a prepass, right, so how would you precalc it? Or do you mean BF for second bounce?
                      As for the irradiance map, it's not working well in this scene, even at high settings. A blurry irrmap gives me artifacts in the small details of the mouldings and a dense irrmap gives me blotchiness all over the place (plus smaller but still noticeable artefacts in the fine detail areas).
                      It's funny because I've been using Vray for some time now and it feels as though render times are getting longer for the same quality (though arguably my scenes have been getting more complex). But I agree, there is something wrong with 19 hours of render to obtain this result. Especially since the white walls are, in this case, pure diffuse.

                      EDIT: Sorry, you do mean IrrMap/BF in this order. Makes sense. I'll give it a try.
                      Check my blog

                      Comment


                      • #12
                        You are a master so it's difficult to suggest something should work, probably you've just tried but I try.

                        - did you try to delete or hide the glass and see if they are that raise rendertimes?
                        - with an override material (deleting glass to simulate transparent objects), did you get the same slow rendertimes?
                        - using irmap, you could use the GI multiplier in the vray properties for the objects with loss of GI details

                        I hope to help you..

                        Comment


                        • #13
                          Thanks Bardo. I did try to render without the glass and found that increased render time a bit, not hugely so but still significantly (but then again, I need all that glass.)
                          I did not try the gi multiplier for those difficults objects. I guess I could.
                          Check my blog

                          Comment


                          • #14
                            Have you tried rendering the scene in Solidrocks for comparison?

                            Comment


                            • #15
                              I don't own Solidrocks. I always trusted myself to pick the best settings for an image, but I'm obviously deluded
                              Check my blog

                              Comment

                              Working...
                              X