Announcement

Collapse
No announcement yet.

Sampling tutorial for V-Ray 3.20

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

  • #61
    And further to that, a huge factor in deciding whether an object needs cleaning up via AA sampling OR shading sampling is it's surface area.

    Say for example you've got a wide flat plane with a glossy material. Each pixel of the plane sitting next to each other will have slight variation because A.) different parts of the object will reflect different parts of your scene and B.) When your sampling is quite low, you won't have taken enough samples for your glossy material for the final result to be accurate, so you'll get huge variation from one pixel to another. We see this in our renders as "noise" or "grain". So once this shading sampling is done, vray hands back those results to be then processed by the image sampler. If it sees lots of variation between the pixels, it presumes that it's noise and will need to be cleaned up, so we get the AA sampler starting to do lots of work to try and smooth out the bad results our shading sampling has done. This theory works for large, flat, simple things.

    Now lets say you've got an object that has loads of small, fine detail like fur or grass or trees. A lot of the details in these objects might end up only being a pixel or two wide in your final image. The quality of shape / outline / silhouette of your object will be determined by your anti aliasing sampling, so no matter what, if you've got loads of fine detail or geometric edges, you'll end up using lots of aa samples here to get smooth results. If we think of our shading noise from the earlier example, a big difference between this and a blade of grass or a bit of fur is that our fur object might only be one, two or three pixels wide in our final image. Previously we had a lot of pixels beside each other, all with highly varying results from our glossy material causing noise. In the case of grass, say if it's only three pixels wide in our final render, that gives you one line of pixels for the left edge of the grass, and since it's an edge it'll get high aa sampling, then you get the middle row of pixels which might have some noise from a glossy material, then you'll have the final right edge row of pixels which has reallyheavy aa and again will be very smooth. So on a blade of grass, it's only really the middle line of pixels that doesn't get massive aa, and since vray's sampling is always judging quality based on pixels beside each other, this middle line of pixels has nothing to compare itself too for starters, and second of all doesn't even cover a large enough amount of pixels or have a large enough surface area in our final render to visually detect any noise or grain so you might as well not bother with lots of shading sampling on these thin objects and just let aa handle.

    Long story short - big, wide, flat surfaces lots of shading sampling, especially if glossy. If big, wide, flat surface has a fine bump or displacement though, it's kind of like having lots of fine edges so slightly less shading sampling, more AA.

    Thin, detailed, fine geometric surfaces, lots of aa, less shading sampling.

    Comment


    • #62
      Originally posted by joconnell View Post
      And further to that, a huge factor in deciding whether an object needs cleaning up via AA sampling OR shading sampling is it's surface area.

      Say for example you've got a wide flat plane with a glossy material. Each pixel of the plane sitting next to each other will have slight variation because A.) different parts of the object will reflect different parts of your scene and B.) When your sampling is quite low, you won't have taken enough samples for your glossy material for the final result to be accurate, so you'll get huge variation from one pixel to another. We see this in our renders as "noise" or "grain". So once this shading sampling is done, vray hands back those results to be then processed by the image sampler. If it sees lots of variation between the pixels, it presumes that it's noise and will need to be cleaned up, so we get the AA sampler starting to do lots of work to try and smooth out the bad results our shading sampling has done. This theory works for large, flat, simple things.

      Now lets say you've got an object that has loads of small, fine detail like fur or grass or trees. A lot of the details in these objects might end up only being a pixel or two wide in your final image. The quality of shape / outline / silhouette of your object will be determined by your anti aliasing sampling, so no matter what, if you've got loads of fine detail or geometric edges, you'll end up using lots of aa samples here to get smooth results. If we think of our shading noise from the earlier example, a big difference between this and a blade of grass or a bit of fur is that our fur object might only be one, two or three pixels wide in our final image. Previously we had a lot of pixels beside each other, all with highly varying results from our glossy material causing noise. In the case of grass, say if it's only three pixels wide in our final render, that gives you one line of pixels for the left edge of the grass, and since it's an edge it'll get high aa sampling, then you get the middle row of pixels which might have some noise from a glossy material, then you'll have the final right edge row of pixels which has reallyheavy aa and again will be very smooth. So on a blade of grass, it's only really the middle line of pixels that doesn't get massive aa, and since vray's sampling is always judging quality based on pixels beside each other, this middle line of pixels has nothing to compare itself too for starters, and second of all doesn't even cover a large enough amount of pixels or have a large enough surface area in our final render to visually detect any noise or grain so you might as well not bother with lots of shading sampling on these thin objects and just let aa handle.

      Long story short - big, wide, flat surfaces lots of shading sampling, especially if glossy. If big, wide, flat surface has a fine bump or displacement though, it's kind of like having lots of fine edges so slightly less shading sampling, more AA.

      Thin, detailed, fine geometric surfaces, lots of aa, less shading sampling.
      wooooh !! Thanks again joconnel !! another really clear, clever, pragmatic explanation.... well done !
      thanks
      (Sorry for my bad english)

      Comment


      • #63
        Just putting another phrasing on vlado's ideas

        Comment


        • #64
          Hopefully all of that can be made automatic so that you won't have to worry about any of it

          Best regards,
          Vlado
          I only act like I know everything, Rogers.

          Comment


          • #65
            Vlado and his sneaky robots

            Comment


            • #66
              Originally posted by joconnell View Post
              Just putting another phrasing on vlado's ideas
              Yes, it's true but, you know, sometime, several explanations are necessary to understand the idea ! btw, for me !!

              Originally posted by alexyork View Post
              Vlado, I think we're gonna need another one of your fine video tutorials for this! A little "advanced sampling" chapter, perhaps.

              Also, eagerly awaiting your video(s) for Secondary Illumination and others.
              Of course, agree with alexyork !! Vlado, your videos tutorial are the holy grail for us
              (Sorry for my bad english)

              Comment


              • #67
                Originally posted by rikou View Post
                Yes, it's true but, you know, sometime, several explanations are necessary to understand the idea ! btw, for me !!
                Yep, I used to lecture before I started in vfx and it's very true that people understand the same idea in very different ways, so if you can explain the same thing with a few different metaphors then hopefully there'll be one that works for the listener

                Comment


                • #68
                  Originally posted by joconnell View Post
                  And further to that, a huge factor in deciding whether an object needs cleaning up via AA sampling OR shading sampling is it's surface area.

                  Say for example you've got a wide flat plane with a glossy material. Each pixel of the plane sitting next to each other will have slight variation because A.) different parts of the object will reflect different parts of your scene and B.) When your sampling is quite low, you won't have taken enough samples for your glossy material for the final result to be accurate, so you'll get huge variation from one pixel to another. We see this in our renders as "noise" or "grain". So once this shading sampling is done, vray hands back those results to be then processed by the image sampler. If it sees lots of variation between the pixels, it presumes that it's noise and will need to be cleaned up, so we get the AA sampler starting to do lots of work to try and smooth out the bad results our shading sampling has done. This theory works for large, flat, simple things.

                  Now lets say you've got an object that has loads of small, fine detail like fur or grass or trees. A lot of the details in these objects might end up only being a pixel or two wide in your final image. The quality of shape / outline / silhouette of your object will be determined by your anti aliasing sampling, so no matter what, if you've got loads of fine detail or geometric edges, you'll end up using lots of aa samples here to get smooth results. If we think of our shading noise from the earlier example, a big difference between this and a blade of grass or a bit of fur is that our fur object might only be one, two or three pixels wide in our final image. Previously we had a lot of pixels beside each other, all with highly varying results from our glossy material causing noise. In the case of grass, say if it's only three pixels wide in our final render, that gives you one line of pixels for the left edge of the grass, and since it's an edge it'll get high aa sampling, then you get the middle row of pixels which might have some noise from a glossy material, then you'll have the final right edge row of pixels which has reallyheavy aa and again will be very smooth. So on a blade of grass, it's only really the middle line of pixels that doesn't get massive aa, and since vray's sampling is always judging quality based on pixels beside each other, this middle line of pixels has nothing to compare itself too for starters, and second of all doesn't even cover a large enough amount of pixels or have a large enough surface area in our final render to visually detect any noise or grain so you might as well not bother with lots of shading sampling on these thin objects and just let aa handle.

                  Long story short - big, wide, flat surfaces lots of shading sampling, especially if glossy. If big, wide, flat surface has a fine bump or displacement though, it's kind of like having lots of fine edges so slightly less shading sampling, more AA.

                  Thin, detailed, fine geometric surfaces, lots of aa, less shading sampling.
                  Thanks for this detailed explanation.


                  Originally posted by vlado View Post
                  Hopefully all of that can be made automatic so that you won't have to worry about any of it

                  Best regards,
                  Vlado
                  Great ��
                  also much fun to understand the theory behind the pixels
                  for my blog and tutorials:
                  www.alfasmyrna.com

                  Comment


                  • #69
                    been playing around with this for a bit but still not really working as well as i hoped. I will package it up for you Vlado so you can have a go.

                    Side thought: for me right now, what I would LOVE would be some way of telling vray that certain objects or materials are less critical than others in terms of noise, and preferably in progressive (pretty much using progressive all the time now)
                    www.peterguthrie.net
                    www.peterguthrie.net/blog/
                    www.pg-skies.net/

                    Comment


                    • #70
                      Originally posted by peterguthrie View Post
                      what I would LOVE would be some way of telling vray that certain objects or materials are less critical than others in terms of noise, and preferably in progressive (pretty much using progressive all the time now)
                      i think that's what the 'subdivs multiplier' in the 'vray object properties' is for if that's what you mean?. in this image each sphere and light have identical settings apart from the sphere on the right has the 'subdiv multiplier' is lowered to 0.01. this is rendered using progressive

                      Click image for larger version

Name:	vraySampling.jpg
Views:	1
Size:	207.4 KB
ID:	858051

                      Comment


                      • #71
                        Originally posted by peterguthrie View Post
                        preferably in progressive (pretty much using progressive all the time now)
                        What are your settings for draft and final in progressive?
                        for my blog and tutorials:
                        www.alfasmyrna.com

                        Comment


                        • #72
                          Originally posted by Companioncube View Post
                          i think that's what the 'subdivs multiplier' in the 'vray object properties' is for if that's what you mean?. in this image each sphere and light have identical settings apart from the sphere on the right has the 'subdiv multiplier' is lowered to 0.01. this is rendered using progressive

                          [ATTACH=CONFIG]25477[/ATTACH]
                          did a quick test last night and seemed promising, i was under the impression the subdiv multiplier thing would only affect it in adaptive mode

                          Originally posted by pixela View Post
                          What are your settings for draft and final in progressive?
                          the same, thats the beauty of it
                          www.peterguthrie.net
                          www.peterguthrie.net/blog/
                          www.pg-skies.net/

                          Comment


                          • #73
                            Hey Peter, ever tried RT GPU as final renderer?
                            We do more and more final production renders with RT GPU and it rocks.

                            If you wish, I'd love to give your scene on top (the one you send to Vlado) a go to compare render-times, noise and look against RT GPU on GPU node.

                            Cheers
                            Stan
                            3LP Team

                            Comment


                            • #74
                              as i get it per-object subdivs setting is a multiplier of MinShade rate ?
                              Available for remote work.
                              My LinkedIn: https://www.linkedin.com/in/olegbudeanu/

                              Comment


                              • #75
                                Originally posted by pixela View Post
                                1. So is it possible to say, higher AA subdivisions are needed for geometry-based details and higher MSR for material based details such as glossiness etc?
                                correct.
                                2. For example a masonry wall with vray displacement (with a hi-res map)
                                As far as I understand, we need to raise AA subdivisions. Right?
                                that will depend from the projection on screen of the finest wall/texture details.
                                If you're close enough, perhaps not.
                                3. If we increase MSR and keep max subdivisions same,
                                then will we have more noise and less detail on geometrically-intricate areas?
                                Actually, the contrary: for EACH AA ray, ALL the MSR ones are cast.
                                So if you had an msr of 6 and a max subdivs of 10, you'd be casting 6*100 rays per pixel.
                                If you had an msr of 12, and the same max subdivs, you'd be casting 1200 rays per pixel, twice the original amount.
                                What would happen IF there was some finer detail, is that it'd run the risk of needing to reach all the AA subdivs set, so oversampling would occur there.
                                Vlado's at work to make sure this doesn't happen anymore, so it's not something we ought to bother too much about.
                                Lele
                                Trouble Stirrer in RnD @ Chaos
                                ----------------------
                                emanuele.lecchi@chaos.com

                                Disclaimer:
                                The views and opinions expressed here are my own and do not represent those of Chaos Group, unless otherwise stated.

                                Comment

                                Working...
                                X