Announcement

Collapse
No announcement yet.

Why IrMap over Brute force for animation rendering?

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

  • #91
    does having retrace turned on have any effect when using IR? or is it automatically ignored

    Comment


    • #92
      Originally posted by hoppergrass View Post
      does having retrace turned on have any effect when using IR? or is it automatically ignored
      This is an interesting question. Since Retrace is performed on the BF side, what happens if you have IrrMap+LC as GI solution?

      Also, time wise, is there a comprehensive comparison of animation fly-through using IrrMap+LC versus BF+LC? Would be VERY nice to see something along the lines "video A took XX time to render using IrrMap+LC and video B too YY time using BF+LC". Maybe someone already done it and just I'm not aware?

      Cheers.

      Comment


      • #93
        I've been trying to follow along to this thread. There are a few things that have gone over my head, but i'm slowly trying to work through the animation workflow when i have time.

        In response to davious3d's request to see a render time comparison,,.
        I have attached 2 links below (hopefully they work) which show a comparison between IM+LC and BF+LC.
        Zdepth has been turned off so that the final results are clear.

        No precalculating was done for either animation. I simply hit render and sent it to my local render farm.


        Irradiance Map + Light Cache
        https://www.dropbox.com/s/s6ohbaoftp..._test.mov?dl=0

        - each frame rendered at approximately 15mins. (I think that the render times should be much faster on this... but as I really only work on stills, i don't have a lot of experience with optimising scenes for animation).
        - the quality of the animation is rubbish. Lots of flickering. I anticipate that this could be minimised if precalculating was done etc, but based on the results, i don't have much of a desire to troubleshoot it.


        Brute Force + Light Cache
        https://www.dropbox.com/s/9q1r7chzv5..._test.mov?dl=0

        - each frame rendered at approximately 30mins!!!.
        - quality of the animation is SIGNIFICANTLY better, however render times is DOUBLE!!!
        Obviously i'm new to 'animation', and i have a lot to learn, but based on my first test.... I don't have any interest in using IM+LC.... However if i'm going to use BF+LC, I' need to improve render times.

        I hope you find this a helpful comparison.

        I will try to read over and understand some of the later comments that were shared on this thread.
        Now that i've done the comparision, i'm not really sure what to try next. But i guess i just keep reading up on it, and sending scenes to render in my downtime!!!
        Long road ahead before i offer this to a client!!! gulp.
        Last edited by 4754simon; 03-05-2017, 07:11 PM.

        Comment


        • #94
          the animations don't play very well directly from the links. probably best to download links and play directly from your hard drive.

          Comment


          • #95
            The animation being "rubbish" for the IRMap is why it took so little to calculate.
            By the time you made IRMap work well enough for such a scene (that is, if you chose its settings just right, after trial and error across the SEQUENCE, as individual frames change in content, and one may succeed where another fails.), your precalc time would shoot through the roof, along with the needed ram to store and interpolate the samples.
            Hence the Brute Force method: fire, forget.

            In one word, the comparison isn't quite viable yet as we lack the settings,and the irmap solution isn't usable at all, so it can't possibly work as a yardstick.
            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


            • #96
              Thanks for the tests, Simon, but you can't compare IrrMap+LC *without* caching at all! It will inevitably flicker (as it did). The workflow (for the test you made) isn't complicate *at all*, but you need to cache it nevertheless. Once cached, even your render times WILL decrease per frame (I suspect they will fall within 10 minutes per frame). Of course you have to account for the caching (which will take roughly 5-10 minutes per frame, but you would only have to do it for every 15 or 20 frame in your case).

              Here what I would do (and if you can do it, the better so we will have a more complete comparison )

              1 - Set Irradiance Map to "Multiframe Incremental" or "Incremental Add to Current Map". Click "SAVE", choose a name and location.
              2 - Set Light Cache to "Fly-Through". There's also the option to "Use Camera Path" - I normally DON'T use it as I want to calculate it every 15 or 20 frames, but not entirely sure this should be ON or OFF. Click "SAVE", choose a name and location.
              3 - Go to "Global Switches" tab and tick "Don't Render Final Image"
              4 - Go to "Common" tab and set "Every Nth Frame" to 15 or 20 in your case
              5 - Render. This will go a tad slow (5 to 10 min) on the first frame and faster with each subsequent frame
              6 - When it's over, go to the "GI" tab and on the Irradiance Map, choose "From File" and make sure you select the file named on step 1
              7 - On the Light Cache, also choose "From File" and also select the file you named on step 2
              8 - Untick "Don't Render Final Image" on the "Global Switches" tab
              9 - Go to "Common" tab and set "Every Nth Frame" back to 1.
              10 - Render

              I know. It looks like too much, but it's actually easy

              Comment


              • #97
                Originally posted by davius3d View Post
                Thanks for the tests, Simon, but you can't compare IrrMap+LC *without* caching at all! It will inevitably flicker (as it did). The workflow (for the test you made) isn't complicate *at all*, but you need to cache it nevertheless. Once cached, even your render times WILL decrease per frame (I suspect they will fall within 10 minutes per frame). Of course you have to account for the caching (which will take roughly 5-10 minutes per frame, but you would only have to do it for every 15 or 20 frame in your case).

                Here what I would do (and if you can do it, the better so we will have a more complete comparison )

                1 - Set Irradiance Map to "Multiframe Incremental" or "Incremental Add to Current Map". Click "SAVE", choose a name and location.
                2 - Set Light Cache to "Fly-Through". There's also the option to "Use Camera Path" - I normally DON'T use it as I want to calculate it every 15 or 20 frames, but not entirely sure this should be ON or OFF. Click "SAVE", choose a name and location.
                3 - Go to "Global Switches" tab and tick "Don't Render Final Image"
                4 - Go to "Common" tab and set "Every Nth Frame" to 15 or 20 in your case
                5 - Render. This will go a tad slow (5 to 10 min) on the first frame and faster with each subsequent frame
                6 - When it's over, go to the "GI" tab and on the Irradiance Map, choose "From File" and make sure you select the file named on step 1
                7 - On the Light Cache, also choose "From File" and also select the file you named on step 2
                8 - Untick "Don't Render Final Image" on the "Global Switches" tab
                9 - Go to "Common" tab and set "Every Nth Frame" back to 1.
                10 - Render

                I know. It looks like too much, but it's actually easy
                Hi, what do you mean "caching"? how to do it?
                Best regards,
                Jackie Teh
                --

                3ds max design 2023, V-Ray 7 [7.00.03 build 32836]
                AMD Threadripper 1950X @3.40 GHz | 64GB RAM | Nvidia RTX 3070 ti
                Website: https://www.sporadicstudio.com
                Email: info@sporadicstudio.com
                YouTube: https://www.youtube.com/c/SporadicStudio

                Comment


                • #98
                  Thanks davius3D.
                  I vaguely remember watching collegues doing this nearly 10 years ago, so i'm vaguely familiar with the process.

                  I will try to set it up today and render over the weekend. All going well, i will try to update this feed with the comparison.

                  Comment


                  • #99
                    Caching is exactly the steps I described. It's pre-calculating the GI solution, interpolate it, save it and then use it to render the final frames. This is good (IMHO) for animations where only the camera moves. When you have animated characters and/or animated lights, caching becomes problematic, convoluted and, with recent updates of VRay, inefficient too.

                    @4754simon if you can, re run your tests using the steps I described. I'm curious on the possible outcome

                    Comment


                    • Following on from my previous post, i have completed the tests (BF+LC vs IM+LC).
                      I have re-rendered the IM+LC animation with pre-calculated passes (as per davius3D's suggestion).

                      Despite the recommendations at the start of this post... i (personally) can see no benefit in producing animations (camera movement only) using Brute Force.

                      Following on from davius3D's suggestions...
                      - I pre-calculated IM and LC maps. I rendered every 20th frame. (It was so quick, i probably could have precalculated every 10th).
                      - I then loaded each pre-calculated map back into the settings, and sent it to render.
                      - The images rendered within about 14-15mins per frame. (There wasn't much time saved to render each frame without calculating the pre-pass, however it still renders the animation in half the time of BF+LC).

                      I have attached both links. (NOTE: best to download and play on your local machine)

                      Although its a bit annoying having to calculate the passes (because you increase the risk of making a mistake when the deadline is looming), there are NO FLICKERS in the animation.
                      The quality seems to match BF+LC.
                      However the one kicker... speed to produce the animation is halved.

                      IM+LC
                      https://www.dropbox.com/s/28pdezxuk5...r_IM2.mov?dl=0


                      BF+LC
                      https://www.dropbox.com/s/ilhv3vm3a4...r_BF2.mov?dl=0

                      Comment


                      • The first thing i saw was precisely the flickering on the rosemary plant.
                        But if it floats your boat, and the scenes you have are of this complexity, hey, it wasn't left in V-Ray by accident: use it.
                        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


                        • I would definitely prefer to use the BF+LC workflow that you discussed earlier in the thread. It is certainly a workflow i would like to adopt. However if i want to offer animation as a service, i need to figure out how to half the render times to match IM+LC speed. When deadlines become a reality, i need to get animations out the door.

                          Admittedly, my knowledge of animation is very limited at this stage, but i will continue watching this thread with interest, and hopefully i can stumble across a workflow that works for me.

                          Comment


                          • Well, i can guarantee you i am not an elitist je*k: I (and the Chaos team, most importantly!) really do believe, based on numbers, that the BF/LC as suggested is by far the best, hassle-less, quicker-through-the-door solution we have to offer, bar none.
                            As said before, yes, corner cases, or extremely simple setups, will lend themselves to a quicker, and yet quite usable, IRMap solution, but those scenes are becoming fewer, and much, much farther apart.

                            Let me take you through a thought experiment with the scene you posted.
                            For all the geometric simplicity (i mean topologically. not as a slight to your scene prepping.) across most of it, all flat-ish surfaces and gently curving into one another, the introduction of the Rosemary Plant pretty much made IRMap unusable.
                            If i was your client, and noticed flickering, we'd be having an argument now (it's there, no it's not, well i see it also on the oven, i can't use this, so i won't pay, and then you need to fix it...), which isn't free, nor instantaneous.
                            If you then stuck to IRMap and expected it to be pixel perfect (which you can try. crank up that max rate!) and stable, you ram usage and render times would literally skyrocket, and yet it may (and often would) still fail in places.
                            That approach would require two things: great experience in reading a scene so to nail the right settings for it (it is all screen-based. so how far is that tiny detail from screen? what's your max rate gonna be? frame range interval? Subdivs? Interpolation mode and settings?), or a LOT of time wasted testing sequences, and the time to re-render for final.

                            BF/LC would be out of the door, the first time rendering, every single time, regardless of scene contents.

                            So, a straight up render time comparison between an arbitrarily loosely interpolated, and fallacious, technique, and a supersampled, exact one, cannot be had this way.
                            A "Production" isn't the rendertime for a frame, but the whole process, start to end, including testing, troubleshooting, debugging, and scene changes to exhaustion (another thing IRMap grieves you into, and BF unshackles you from.).

                            I hope i summed up the thoughts nicely enough for the debate to continue: i'd love to hear more qualms about the method, and see where i can help at least with providing some perspective.
                            Before and After which, anyone's free to do as they please, anyway.
                            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


                            • Originally posted by 4754simon View Post
                              However the one kicker... speed to produce the animation is halved.

                              IM+LC
                              https://www.dropbox.com/s/28pdezxuk5...r_IM2.mov?dl=0


                              BF+LC
                              https://www.dropbox.com/s/ilhv3vm3a4...r_BF2.mov?dl=0
                              add animated people, cars, gusts of wind and than compare the time spent on rendering an animation.
                              Marcin Piotrowski
                              youtube

                              Comment


                              • thanks for the comments.
                                Yes. The scene only has an animated camera, so i would expect there to be huge complications if i started animating objects within the scene.

                                With regard to the flickering rosemary. If you look closely that both animations, the flickering is exactly the same on the rosemary plant. I think DOF is creating the illusion that it is flickering.
                                Also, the shadowing around the edges with the IM+LC seems to be more stable (if you look very closely). NOTE: you have to download the anims to your local machine to view.... Viewing it on Dropbox is rubbish.

                                That being said, I am certainly open to using BF+LC. I would MUCH prefer to use this method. I just need to figure out how to get render times down.
                                I will keep trying to find more resources on the technique.

                                Comment

                                Working...
                                X