Page 83 of 86 FirstFirst ... 33738182838485 ... LastLast
Results 1,641 to 1,660 of 1707

Thread: Experimental 2.77 Cycles Denoising build

  1. #1641
    Could someone please make a Linux build with at least the scramble option if not all the experimental additions from 3DLuvers builds? The last build that was available is badly outdated (effecting mostly the denoiser....). I cant believe that I'm the only one here using Linux .



  2. #1642
    Member
    Join Date
    Oct 2011
    Location
    Stockholm
    Posts
    204
    Originally Posted by 3DLuver View Post
    Hey, 2.8 is still in heavy dev. At present there is no way to have Eevee running in viewport and render with cycles, you have to change from one to the other. You may also find if rendering with cycles the realtime viewport window will interfere with the render output. If so just change the 3dviewport to text editor etc when rendering. Im sure the devs which are doing a great job with 2.8 will change these things as soon as possible.

    Have to say Eevee is looking better than i thought it would at this point, I know 2.79 is on the way but my eyes are on 2.8 for the time Blender becomes a big boys toy.
    Sorry if my comment sounded negative, I was genuinely wondering if the way it is designed right now would be the one in the 2.8 release, and if I was missing some piece of information to maintain a logical workflow (like a hotkey to disable cycles rendering in the viewport). Eevee looks really amazing already, I've been testing it for a while and it's really promising.



  3. #1643
    Member Froggy Jack's Avatar
    Join Date
    Jun 2017
    Location
    Sweden
    Posts
    13
    Originally Posted by bekic.bojan View Post
    Could someone please make a Linux build with at least the scramble option if not all the experimental additions from 3DLuvers builds? The last build that was available is badly outdated (effecting mostly the denoiser....). I cant believe that I'm the only one here using Linux .
    You are not dude But I can't really help you out...



  4. #1644
    Member
    Join Date
    Sep 2013
    Location
    Switzerland, Thun
    Posts
    156
    Recently I did some testing with the denoiser and it works very nice in general. Only details which originate from a roughness map aren't preserved very well. Is this something that could be improved?
    - Simon Liechti -
    Web | Artstation | Instagram



  5. #1645
    Member
    Join Date
    Oct 2012
    Location
    UK, England
    Posts
    531
    Does anyone know the debug code for de noise? e.g 256 is render settings debug code, but whats the debug code for the de noiser



  6. #1646
    Member lsscpp's Avatar
    Join Date
    May 2006
    Location
    Firenze
    Posts
    2,031
    Originally Posted by 3DLuver View Post
    Does anyone know the debug code for de noise? e.g 256 is render settings debug code, but whats the debug code for the de noiser
    set 'experimental' mode for cycles in render panel. Then you can enable denoise passes in renderlayer's pass panel
    That's all I know
    Everything's relative. Even saying "Everything's relative".



  7. #1647
    Member Ace Dragon's Avatar
    Join Date
    Feb 2006
    Location
    Wichita Kansas (USA)
    Posts
    28,146
    Just putting this here, has anyone discovered the possibility that the best results with denoising come through use of the Sobol sampler with Branched Path Tracing?

    At least in simple tests it performed far better with slightly blurry reflections (leaving the contact points sharp which does not happen with CMJ or with regular path tracing in general. Could it be the reason why some here have gotten amazing results while others had issues with blurriness?
    Sweet Dragon dreams, lovely Dragon kisses, gorgeous Dragon hugs. How sweet would life be to romp with Dragons, teasing you with their fire and you being in their games, perhaps they can even turn you into one as well.
    Adventures in Cycles; My official sketchbook



  8. #1648
    Member
    Join Date
    Sep 2012
    Posts
    2,572
    Yup... IIRC, Lukas (or?*) mentioned it somewhere already... along with why BPT & some more stuff in Cycles should be fixed.

    *many ppl involved & i am bad, bad, bad in connecting ideas with names without faces
    Last edited by burnin; 16-Jul-17 at 17:38.



  9. #1649
    Member
    Join Date
    Oct 2012
    Location
    UK, England
    Posts
    531
    Ive been playing with a new test branch that has SURE-based Filtering and Adaptive Sampling for Monte Carlo Reconstruction, it does a sure estimate of error on all separate R G B channels and sums variance results for Color,Depth,Norm also using neighbor pixel kernel values to generate the adaptive sample evaluation.

    It's still work in progress but I get better results with BDPT but im also using the patch not in master D2730 Improved triangle sampling for mesh lights, D2708 Cycles: Use correct multi sample MIS model for branched path tracing and some others cant remember off the top of my head.

    In the Last experimental branch i Posted on here those patch's as well as scramble, persistant data and a few others were included so if you want to test BDPT v PT for denoise test you can use that until im finished with my new build.



  10. #1650
    Member lsscpp's Avatar
    Join Date
    May 2006
    Location
    Firenze
    Posts
    2,031
    Originally Posted by 3DLuver View Post
    Ive been playing with a new test branch that has SURE-based Filtering and Adaptive Sampling for Monte Carlo Reconstruction, it does a sure estimate of error on all separate R G B channels and sums variance results for Color,Depth,Norm also using neighbor pixel kernel values to generate the adaptive sample evaluation.

    It's still work in progress but I get better results with BDPT but im also using the patch not in master D2730 Improved triangle sampling for mesh lights, D2708 Cycles: Use correct multi sample MIS model for branched path tracing and some others cant remember off the top of my head.

    In the Last experimental branch i Posted on here those patch's as well as scramble, persistant data and a few others were included so if you want to test BDPT v PT for denoise test you can use that until im finished with my new build.
    wow. how can we play too?
    hey wait a minute! SURE filtering? Who made the patch? What branch is this??
    Last edited by lsscpp; 17-Jul-17 at 07:23.
    Everything's relative. Even saying "Everything's relative".



  11. #1651
    Member
    Join Date
    Oct 2012
    Location
    UK, England
    Posts
    531
    I made the Patch, It's my New custom Adaptive Sample Experimental branch.



  12. #1652
    Member lsscpp's Avatar
    Join Date
    May 2006
    Location
    Firenze
    Posts
    2,031
    Originally Posted by 3DLuver View Post
    I made the Patch, It's my New custom Adaptive Sample Experimental branch.
    How cool is that! I didn't know you were a coder too. So, you worked to a new approach (SURE) for adaptive sampling? Can you explain it a bit?

    edit (after documenting...)
    so SURE is an alternative to LWR, right? (but Lukas lwr implementation is alredy in master...)
    And you made it work in conjunction with Adaptive rendering by M. Jaros?
    Last edited by lsscpp; 17-Jul-17 at 09:39.
    Everything's relative. Even saying "Everything's relative".



  13. #1653
    Member Ace Dragon's Avatar
    Join Date
    Feb 2006
    Location
    Wichita Kansas (USA)
    Posts
    28,146
    I do not know myself, but I assume 3DLuver's adaptive sampling will be more akin to Lukas' attempt at actual adaptive sampling rather than the denoising he worked on?

    Are you also going to submit the patch to the tracker when it's ready?
    Sweet Dragon dreams, lovely Dragon kisses, gorgeous Dragon hugs. How sweet would life be to romp with Dragons, teasing you with their fire and you being in their games, perhaps they can even turn you into one as well.
    Adventures in Cycles; My official sketchbook



  14. #1654
    Member Ace Dragon's Avatar
    Join Date
    Feb 2006
    Location
    Wichita Kansas (USA)
    Posts
    28,146
    Originally Posted by burnin View Post
    Yup... IIRC, Lukas (or?*) mentioned it somewhere already... along with why BPT & some more stuff in Cycles should be fixed.

    *many ppl involved & i am bad, bad, bad in connecting ideas with names without faces
    Well, further testing confirms my suspicion here, but BPT unfortunately is not very usable in scenes with many-closure materials and the convergence rate of certain highlights (on glass materials produced by mesh-lights) don't seem to see an effect from any of the sampling controls (the latter making BPT totally unusable if they need a lot of samples).

    Now I can tell that the all closure sampling is a key part of what makes the denoiser better with BPT, but it's also the downfall of that mode (for materials with many shading components, it might be necessary to have a system where not all of them need to get a sample with each pass and where not all of them need the same number within the pass).
    Last edited by Ace Dragon; 18-Jul-17 at 18:43.
    Sweet Dragon dreams, lovely Dragon kisses, gorgeous Dragon hugs. How sweet would life be to romp with Dragons, teasing you with their fire and you being in their games, perhaps they can even turn you into one as well.
    Adventures in Cycles; My official sketchbook



  15. #1655
    Member kivig's Avatar
    Join Date
    Feb 2009
    Location
    Riga, Latvia
    Posts
    200
    @lukasstockner97

    First of all - denoiser is amazing to have. It already saved me and my colleagues a lot in a recent animation and drastically improved Cycles usability. Thank you!

    If I may suggest an improvement - is it possible to denoise alpha?
    The noise becomes very apparent with shadow catchers as they are basically alpha-only.



  16. #1656
    Member cekuhnen's Avatar
    Join Date
    Mar 2003
    Location
    Grosse Pointe Woods, Michigan, USA
    Posts
    11,980
    Originally Posted by kivig View Post
    @lukasstockner97

    First of all - denoiser is amazing to have. It already saved me and my colleagues a lot in a recent animation and drastically improved Cycles usability. Thank you!

    If I may suggest an improvement - is it possible to denoise alpha?
    The noise becomes very apparent with shadow catchers as they are basically alpha-only.
    Very good point and suggestion!
    3D Design Generalist - Faculty Industrial + Interior Design
    Chair Interior Design - Wayne State University



  17. #1657
    Indeed, denoising alpha would be awesome.


    Also since I usually do animation and try to save as much rendertime as possible, I do usually almost -direct light setups. There adaptive sampling would make a big difference, since it's only the shadows where things get noisy, so it's good somebody is looking into that too.



  18. #1658
    I’ve started using the denoising options in 2.79 and it is truly awesome – great work Lucas!!

    However, I understand that cross frame filtering is not supported yet and as far as I know does not have a targeted release.

    But all is not lost. https://wiki.blender.org/index.php/U..._Documentation (and this thread) mentions that you can do it though it looks like with a few limitations…


    1. You need to output a Multiplayer OpenEXR frame
    2. You need to build a custom version of Blender that has the Keep Denoising Data feature enabled
    3. You need to build a custom version of the Standalone Cycles renderer


    Then you can generate the frames and use the cross-frame denoising options from the command line.

    Before I start trying to build custom version of Blender/Cycles stand alone (which for me is likely to take a large amount of time) is possible with the current code? Is there any reason people would not recommend it and just use what will be in 2.79?

    Thanks.


    • andrew



  19. #1659
    Member
    Join Date
    Oct 2012
    Location
    UK, England
    Posts
    531
    Just read this paper, This really could be a big turning point in realtime path tracing.

    http://research.nvidia.com/sites/def...f_preprint.pdf



  20. #1660
    Member Ace Dragon's Avatar
    Join Date
    Feb 2006
    Location
    Wichita Kansas (USA)
    Posts
    28,146
    Originally Posted by 3DLuver View Post
    Just read this paper, This really could be a big turning point in realtime path tracing.

    http://research.nvidia.com/sites/def...f_preprint.pdf
    It looks pretty good for such low sample counts, but I'm afraid you would still need a decent number of samples to help it resolve things like details and highlights in reflections (and the paper doesn't say if it can scale up).

    High-quality denoising of reflections with low to moderate sample counts appear to be the achilles-heel for most of the current algorithms (maybe the new neural-based algorithms will solve that). Though a major part of the problem is that scenes often need a lot of samples to render out highlights of small light sources (which may remain an issue until someone figures out how to use markov-chains minus the splotches).
    Sweet Dragon dreams, lovely Dragon kisses, gorgeous Dragon hugs. How sweet would life be to romp with Dragons, teasing you with their fire and you being in their games, perhaps they can even turn you into one as well.
    Adventures in Cycles; My official sketchbook



Page 83 of 86 FirstFirst ... 33738182838485 ... LastLast

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •