Page 67 of 83 FirstFirst ... 1757656667686977 ... LastLast
Results 1,321 to 1,340 of 1657
  1. #1321
    Originally Posted by Geographic View Post
    ... Though in this latest build i get 'lighter' denoise boxes, not seen it before. ...
    Yes, in the last two builds there are increasingly white dots.
    For example in the glass scene from this thread (don't remember the page)



  2. #1322
    Member
    Join Date
    Jul 2008
    Location
    Russia, Voronezh
    Posts
    277
    All, (glass too) worked fine, and it worked on 2 GPUs.


    Lukas, THANK YOU VERY MUCH for your work!!!


    I'm an archviz artist and speed rendering in cycles was a real pain until your patch (and i used BI). With your denoiser, and AO mixer it is possible to use cycles in archviz, because the speed and quality became comparable to vray and corona. Thank you, man!
    Attached Images Attached Images



  3. #1323
    Member mib2berlin's Avatar
    Join Date
    May 2008
    Location
    Germany
    Posts
    3,476
    Hi, some news:
    Most part of the denoiser is now accepted from Brecht, formerly Cycles core developer and Cycles originator,
    so not much time left to have it in master.

    Cheers, mib
    OpenSUSE Leap 42.1/64 i5-3570K 16 GB Blender 2.7 Octane 3.03
    GTX 760 4 GB, GTX 670 2 GB Driver 375.26 | Blender for Octane



  4. #1324
    Sorry if I make a silly question but I've searched this for long time without finding how to do it :

    How can I save both images, the denoiser output image and the non-denoised one ?

    (The only way I can do that is with two RenderLayers, one with denoiser on and the other one with denoiser off... but it renders the image again from zero for the second renderlayer so it's 2 times slower... )

    Thanks a lot in advance !



  5. #1325
    Member mib2berlin's Avatar
    Join Date
    May 2008
    Location
    Germany
    Posts
    3,476
    Hi oneloops, I fear you cant.
    In the first incarnation of the Denoiser was two step denoising included, you first render the image then start denoising.
    It was not enough time to include this for Blender 2.79.

    Cheers, mib
    OpenSUSE Leap 42.1/64 i5-3570K 16 GB Blender 2.7 Octane 3.03
    GTX 760 4 GB, GTX 670 2 GB Driver 375.26 | Blender for Octane



  6. #1326
    Originally Posted by oneloops View Post
    Sorry if I make a silly question but I've searched this for long time without finding how to do it :

    How can I save both images, the denoiser output image and the non-denoised one ?

    (The only way I can do that is with two RenderLayers, one with denoiser on and the other one with denoiser off... but it renders the image again from zero for the second renderlayer so it's 2 times slower... )

    Thanks a lot in advance !

    Try:

    http://pasteall.org/blend/index.php?id=46318

    you get a viewer output noised and an exr with two layers: one noised and one denoised. The formula is (direct+indirect)*color, for diffuse pass, glossy pass, trans and SSS, and add the lights and environment



  7. #1327
    Originally Posted by wikifry View Post
    Try:

    http://pasteall.org/blend/index.php?id=46318

    you get a viewer output noised and an exr with two layers: one noised and one denoised. The formula is (direct+indirect)*color, for diffuse pass, glossy pass, trans and SSS, and add the lights and environment
    Thanks a lot ! It works



  8. #1328
    Member Ace Dragon's Avatar
    Join Date
    Feb 2006
    Location
    Wichita Kansas (USA)
    Posts
    27,903
    Originally Posted by Ace Dragon View Post
    Has anyone tried the filter on scenes that are lit largely by indirect light (especially those that are difficult to light)?

    I have such a scene and at low to medium sample values, I tend to get a lot of regions where pixels contain negative color values (not NaN's, but patches that are a lot darker than the minimum brightness in the unfiltered tile). I know that the darkening is expected in many cases due to how things would look if sampled to convergence, but I'm pretty sure the filter shouldn't produce negative values.

    Also, sometimes when the filter processes a single firefly, the entire box of pixels around it (depending on the radius) will completely change in both tone and brightness. I don't think a single pixel should have that much influence.
    And now, here's a clear example of how the firefly handling kills the patch in terms of use in many situations
    DenoiseTest.png

    Fireflies right now seem to completely throw off the denoiser, leading to the creation of not only a bunch of bright squares, but a lot of dark pixels too (ie. much darker than the minimum color value of any pixel in the tile before denoising).

    The solution to this could be the idea that bright pixels with no immediate neighbors get removed immediately and a way to measure the minimum pixel brightness in a tile and forbid any darkening below that value. On top of that, the algorithm could use a concept of pixel influence, or how much a pixel's value contributes to the denoised result.

    Any thoughts Lukas?
    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



  9. #1329
    Member
    Join Date
    Jan 2014
    Location
    parkbench
    Posts
    26
    Is the firefly problem bigger when using filmic blender? Reasoning being the lighting values needed to fill a scene become way higher, and any subsequent fireflies can have enormous values that throws the denoiser off?



  10. #1330
    Member Ace Dragon's Avatar
    Join Date
    Feb 2006
    Location
    Wichita Kansas (USA)
    Posts
    27,903
    Originally Posted by wowsers View Post
    Is the firefly problem bigger when using filmic blender? Reasoning being the lighting values needed to fill a scene become way higher, and any subsequent fireflies can have enormous values that throws the denoiser off?
    I wouldn't say for sure it would be worse with brighter lights, because the issue pretty much involves any pixel that might be a bit brighter than all of its neighbors (even if it's not really bright in general).

    The only way to get around it to my knowledge might be clamping, but that can kill a bit of the indirect lighting detail by way of dimming it.
    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



  11. #1331
    Just use higher samples. Caustics are a pain in any renderer.
    I can only give suggestions, personal opinions and constructive critique, but it is your decision what you do with it.



  12. #1332
    Member Ace Dragon's Avatar
    Join Date
    Feb 2006
    Location
    Wichita Kansas (USA)
    Posts
    27,903
    Still, it seems erroneous to me that the denoisier would respond to flreflies by producing pixels with negative color values (which shouldn't happen in any case unless the scene itself has materials and lights with negative colors, no matter how noisy things are).
    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



  13. #1333
    Member lsscpp's Avatar
    Join Date
    May 2006
    Location
    Firenze
    Posts
    2,002
    Originally Posted by Ace Dragon View Post
    Still, it seems erroneous to me that the denoisier would respond to flreflies by producing pixels with negative color values (which shouldn't happen in any case unless the scene itself has materials and lights with negative colors, no matter how noisy things are).
    this might be a bug independent from denoiser. when i'll have time i'll post an example of negative values i obtain using blur filters in compositor

    edit:
    here's a screenshot
    http://imgur.com/a/zvQwa
    and here a file:
    http://pasteall.org/blend/index.php?id=46370

    in a more sophisticated scene, the bug appears also using the glare node
    Last edited by lsscpp; 26-Apr-17 at 03:41.
    Everything's relative. Even saying "Everything's relative".



  14. #1334
    Originally Posted by 3DLuver View Post
    Hey,

    Lukas updated the temp denoise branch so this is the experimental version from that with as many additions as i could make.

    Lukas 20 April update:

    https://mega.nz/#!9sRCUBTL!kcvASWpmg61IsOAYYRCmJngFc1sKyGR8g64OZxD vVyA

    Build includes:

    Opencl&Cuda_GPU Denoise system (This is Lukas latest Denoise code system)
    GPU Denoise Multi GPU support (Even in Viewport, Def works for Cuda but not tested multiple Opencl GPU)
    Scramble Distance added for solbol and multi jitter (works on CPU & GPU) Also added to Supported features render tab
    Blue Noise Dithered Sobol with scramble
    Thread DiverganceSort Reduction patch (Gives 30% speedup in classromm and 8% in barcellona)
    Latest Master
    Cycles: Implement persistent storage for all scene data (Patch D2613)
    Cuda & Opencl Supported
    Disney BSDF
    Compositor SMAA Node
    Material Node editor New Noise lyapuno/ mandelbulb Procedural Additions
    This is working REALLY well for me.
    500vs3000 samples
    3000vs500.jpg



  15. #1335
    Member
    Join Date
    Jan 2014
    Location
    parkbench
    Posts
    26
    Both Catrom and Mitch try to keep edges, so extreme blurs (really difference between light and dark) might cause it to "overshoot" colours into the negative.

    It does however display the same behaviour of darkening out smaller details more than large ones, if you set the blur to 7x7 it just barely stops creating black spots on the largest light reflection.
    http://imgur.com/40KIauu

    I imagine if the denoiser uses a similar edge preserving algorithm when bluring, fireflies would be a huge pain to deal with. And the bigger the difference between the firefly value and the surrounding colour the wore it would get...

    edit:
    Basically blurs that are edge preserving work sort of like a lanczos curve thing, it subtracts values around the outside of the bright area to make edges more visible, another blur type that is edge preserving but doesn't thread into the subtractive territory is "Hanning" blur. Hanning blur looks close to regular edge preserving algorithms, but it doesn't use negative values to enhance edges (which also makes it slightly more blurry). I can't speak on how good it looks because I've never used it myself, it might look bad in this usecase.

    Here's a link to an image with some different blur types.
    http://matplotlib.org/1.4.3/_images/...on_methods.png
    Last edited by wowsers; 26-Apr-17 at 07:55.



  16. #1336
    Member lsscpp's Avatar
    Join Date
    May 2006
    Location
    Firenze
    Posts
    2,002
    Originally Posted by wowsers View Post
    Both Catrom and Mitch try to keep edges, so extreme blurs (really difference between light and dark) might cause it to "overshoot" colours into the negative.

    It does however display the same behaviour of darkening out smaller details more than large ones, if you set the blur to 7x7 it just barely stops creating black spots on the largest light reflection.
    http://imgur.com/40KIauu

    I imagine if the denoiser uses a similar edge preserving algorithm when bluring, fireflies would be a huge pain to deal with. And the bigger the difference between the firefly value and the surrounding colour the wore it would get...
    would it be the matter of just avoid catrom or mitch algorithm in the denoise blur?
    Everything's relative. Even saying "Everything's relative".



  17. #1337
    Member
    Join Date
    Jan 2014
    Location
    parkbench
    Posts
    26
    The biggest issue is having the blur being somewhat edge preserving.



  18. #1338
    Member lsscpp's Avatar
    Join Date
    May 2006
    Location
    Firenze
    Posts
    2,002
    Originally Posted by wowsers View Post
    The biggest issue is having the blur being somewhat edge preserving.
    btw some sort of blur is probably happening in the glare filter node since I had those same issue there
    Everything's relative. Even saying "Everything's relative".



  19. #1339
    Member
    Join Date
    Oct 2012
    Location
    UK, England
    Posts
    501
    Originally Posted by wowsers View Post
    The biggest issue is having the blur being somewhat edge preserving.
    I did suggest to lukas a while back that give each material and object it's own color ID for a mask system to seperate everything to it's own denoise layer, then you can make sure edges are not the issue. But not sure if he got what i meant, i first used the same idea in UE4 as a post pro mask system so i could do a quick pixel color check that allowed to me do specific post pro that only effects specific objects or materials. Worked well for that.



  20. #1340
    Member Ace Dragon's Avatar
    Join Date
    Feb 2006
    Location
    Wichita Kansas (USA)
    Posts
    27,903
    The code review process is ongoing
    https://lists.blender.org/pipermail/...il/095977.html

    Also to note, no new builds should be done with this as they contain no changes the user will notice.

    Hopefully within the next month, 3D Luver can take a rest knowing the denoiser is in master.
    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 67 of 83 FirstFirst ... 1757656667686977 ... 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
  •