Denoise renders and save a LOT of time! (Nvidia not Reqd)

145791013

Comments

  • Leonides02Leonides02 Posts: 1,379
    kameneko said:

     

    You're right: I've tried again on a less noisy picture, and it works fine!

    I don't know if it's better of worse, I'd say it's slightly less aggressive than the Nvidia one! :D

    Agreed. To my eyes, the "less aggressive" is better. Sometimes the Nvidia one makes even a mostly noise-free image look like it was made with water colors! LOL

  • TaozTaoz Posts: 9,941
    Hellboy said:

    Taoz, thanks for the tif support!! :D

    YW! :)

  • JoepingletonJoepingleton Posts: 746
    edited April 2019

    WOW thank you so much for this information.  I stumbled across this thread and had to try it out.  This image was rendered at 3840X2160 @ 10 Iray iterations (1 min render time).  Then I ran it through denoiser.exe (4 sec), The results blew my mind.

    Here is another test rendered at 4300X1950 @ 10 Iray iterations (1 min render time)

    Of course, these would be better with longer render times, but this looks like it may be a good cheat for creating faster animations from DAZ Studio and Iray.

    denoiseTest01.jpg
    1920 x 540 - 564K
    denoiseTest02.jpg
    1075 x 487 - 199K
    Post edited by Joepingleton on
  • Leonides02Leonides02 Posts: 1,379

    WOW thank you so much for this information.  I stumbled across this thread and had to try it out.  This image was rendered at 3840X2160 @ 10 Iray iterations (1 min render time).  Then I ran it through denoiser.exe (4 sec), The results blew my mind.

    Here is another test rendered at 4300X1950 @ 10 Iray iterations (1 min render time)

    Of course, these would be better with longer render times, but this looks like it may be a good cheat for creating faster animations from DAZ Studio and Iray.

    Very impressive! Did you use Nvidia or Intel?

  • JoepingletonJoepingleton Posts: 746
    edited April 2019

    WOW thank you so much for this information.  I stumbled across this thread and had to try it out.  This image was rendered at 3840X2160 @ 10 Iray iterations (1 min render time).  Then I ran it through denoiser.exe (4 sec), The results blew my mind.

    Here is another test rendered at 4300X1950 @ 10 Iray iterations (1 min render time)

    Of course, these would be better with longer render times, but this looks like it may be a good cheat for creating faster animations from DAZ Studio and Iray.

    Very impressive! Did you use Nvidia or Intel?

    Thank you, I just used the Intel version

    Post edited by Joepingleton on
  • SotoSoto Posts: 1,440

    Aw... said is expired :(

  • Leonides02Leonides02 Posts: 1,379
    Hellboy said:

    Aw... said is expired :(

    Huh?

  • Alas I still don’t understand how to use this I’m confused about the ins and outs of the process, where to input what to get the end product. I betcha I can’t even convey even what I’m talking about LOL!!!

     

  • TaozTaoz Posts: 9,941
    edited November 2019
    Hellboy said:

    Aw... said is expired :(

    The latest version won't expire (it will go on the net to check for updates instead):

    https://taosoft.dk/software/freeware/dnden/

     

    Post edited by Taoz on
  • edited May 2019

    whoops, I did an email reply to the forum notification - sorry if something weird shows up

    Post edited by ibr_remote aka infinity10 on
  • edited May 2019

    OK, what I needed to mention was that I can get the taosoft Dnd app to work, all paths properly designated in the .ini file, and am using the Declan Russell nVidia and Intel denoiser versions downloaded from his blog (latest versions), and I am running a 64-bit windows 10 home OS 

    I run DnD with admin privileges, but when I try to drag over my rendered image, the cursor turns into a Not Allowed icon (black circle with a diagonal line across it).

    oh well, I tried.

    Post edited by ibr_remote aka infinity10 on
  • TaozTaoz Posts: 9,941

    Alas I still don’t understand how to use this I’m confused about the ins and outs of the process, where to input what to get the end product. I betcha I can’t even convey even what I’m talking about LOL!!!

    If you're on windows you can use the DragNDrop app, this page explains how to install and use it:

    https://taosoft.dk/software/freeware/dnden/

  • Oh my...  I used Declan Russel's Sequence.Bat file suggestion and well....all my system32 PNGs are now denoised.  OKAY, I am quittting this denoising solution while I haven't killed my OS.

  • TaozTaoz Posts: 9,941

    I run DnD with admin privileges, but when I try to drag over my rendered image, the cursor turns into a Not Allowed icon (black circle with a diagonal line across it).

    You should not run it as admin, dragNdrop will not work then.  

  • edited May 2019
    Taoz said:

    I run DnD with admin privileges, but when I try to drag over my rendered image, the cursor turns into a Not Allowed icon (black circle with a diagonal line across it).

    You should not run it as admin, dragNdrop will not work then.  

    Oh. I see. In that case, I need to retry.

    UPDATE: intel denoiser with DnD works very fast, and works well !  nVidia denoiser is slow - probably unusable for me.

    Post edited by ibr_remote aka infinity10 on
  • edited May 2019

    This image pre-Intel Denoiser via DragNDrop was done with Max 15000 samples, Max time = 3600 secs, Rendering quality 0.1, Rendering convergence ration = 10.8 per cent; render time was 50.23 seconds on my PC with specs nVidia GTX760 (192-bit), Driver version = 430.39, Intel Core i7-4790 CPU @3.60GHZ, RAM = 15.96 GB.  It is naturally very noisy.

    50-23_15000_3600_0-10_10.8.png
    1280 x 720 - 2M
    Post edited by ibr_remote aka infinity10 on
  • edited May 2019

    Here is the Intel Denoised version using Drag N Drop interface:

    (The strip lighting and the shadows falling on the upper kitchen cabinets need anti-aliasing.)  The total production time is pretty fast, to be honest !!

    Product is the High Rise Kitchen and Dining by Tesla3DCorp, with background HDRI from Dimension Theory.

    5_15000_7200_0-10_10-8_denoisedIntel.png
    1280 x 720 - 855K
    Post edited by ibr_remote aka infinity10 on
  • LenioTGLenioTG Posts: 2,118
    edited May 2019

     

    Here is the Intel Denoised version using Drag N Drop interface:

    (The strip lighting and the shadows falling on the upper kitchen cabinets need anti-aliasing.)  The total production time is pretty fast, to be honest !!

    Yes, this is a revolution! ^^

    I even managed to "restore" part of my first comics on my Patreon, since it understands where the text is and it doesn't blur anything out! :D

    Post edited by LenioTG on
  • TaozTaoz Posts: 9,941
    Taoz said:

    I run DnD with admin privileges, but when I try to drag over my rendered image, the cursor turns into a Not Allowed icon (black circle with a diagonal line across it).

    You should not run it as admin, dragNdrop will not work then.  

    Oh. I see. In that case, I need to retry.

    UPDATE: intel denoiser with DnD works very fast, and works well !  nVidia denoiser is slow - probably unusable for me.

    What Nvidia card do you have? With my GTX 1070 it usually just takes a few seconds.

  • My PC specs for Daz Studio Pro isntallation is in my post with the pre-Intel Denoise image

     

    Taoz said:
    Taoz said:

    I run DnD with admin privileges, but when I try to drag over my rendered image, the cursor turns into a Not Allowed icon (black circle with a diagonal line across it).

    You should not run it as admin, dragNdrop will not work then.  

    Oh. I see. In that case, I need to retry.

    UPDATE: intel denoiser with DnD works very fast, and works well !  nVidia denoiser is slow - probably unusable for me.

    What Nvidia card do you have? With my GTX 1070 it usually just takes a few seconds.

     

  • PadonePadone Posts: 3,700
    edited May 2019

    Just be aware that the denoiser will not wotk fine without providing the albedo and normal buffers. That is, without extra information the denoiser will mainly just blur the image so details will be lost. This is nothing more than applying a denoiser in GIMP or Photoshop. To get real denoising you need the albedo and normal buffers too.

    This is from the OID page ..

     Using auxiliary feature images like albedo and normal helps preserving fine details and textures in the image thus can significantly improve denoising quality.

    https://openimagedenoise.github.io/documentation.html

    Post edited by Padone on
  • LenioTGLenioTG Posts: 2,118
    Padone said:

    Just be aware that the denoiser will not wotk fine without providing the albedo and normal buffers. That is, without extra information the denoiser will mainly just blur the image so details will be lost. This is nothing more than applying a denoiser in GIMP or Photoshop. To get real denoising you need the albedo and normal buffers too.

    This is from the OID page ..

     Using auxiliary feature images like albedo and normal helps preserving fine details and textures in the image thus can significantly improve denoising quality.

    What are you talking about?

    Are these features included in that git hub link?

    It's already nowhere near "just blur".

  • PadonePadone Posts: 3,700
    TGFan4 said:

    It's already nowhere near "just blur".

    This is an example by pdxtobin of what happens to details when you don't use the albedo and normal buffers. It happens in the 4.11 denoiser too because it is just passing the beauty canvas.

  • Matt_CastleMatt_Castle Posts: 2,571
    edited May 2019
    Padone said:

    Just be aware that the denoiser will not wotk fine without providing the albedo and normal buffers. That is, without extra information the denoiser will mainly just blur the image so details will be lost. This is nothing more than applying a denoiser in GIMP or Photoshop. To get real denoising you need the albedo and normal buffers too.

    It is different to applying a denoiser in GIMP or Photoshop, as it is specifically trained to work on ray-tracing noise. If you try passing a noisy but non ray-traced image through these denoisers, they may barely work at all. (For example, I tried an image from a rasterising engine with noisy DOF and ambient occlusion passes - they didn't pick up on the noise from either)

    In any case, does Daz even have the option to export an albedo canvas? I could only find the option for a Diffuse canvas, but while "diffuse" and "albedo" are sometimes used interchangeably, when I tried using the diffuse canvas, the results were definitely worse.

     

    Post edited by Matt_Castle on
  • Taoz said:

    Alas I still don’t understand how to use this I’m confused about the ins and outs of the process, where to input what to get the end product. I betcha I can’t even convey even what I’m talking about LOL!!!

    If you're on windows you can use the DragNDrop app, this page explains how to install and use it:

    https://taosoft.dk/software/freeware/dnden/

     

    I guess my confusion is this can be applied to an already completed render correct? So if I have a completed render saved as .png I can use that as the source file to run it through the denoiser? Thank you for the link it's very appreciated.

  • PadonePadone Posts: 3,700
    edited May 2019
    In any case, does Daz even have the option to export an albedo canvas?

    AFAIK it should be possible via LPE but no clues from DAZ and no success at the time being .. Perhaps a "DIY" albedo could be done by turning all the materials to emitters so to emulate a flat shader, but it's just a guess.

    https://www.daz3d.com/forums/discussion/205886/light-path-expressions-iray

    Post edited by Padone on
  • TaozTaoz Posts: 9,941
    Taoz said:

    Alas I still don’t understand how to use this I’m confused about the ins and outs of the process, where to input what to get the end product. I betcha I can’t even convey even what I’m talking about LOL!!!

    If you're on windows you can use the DragNDrop app, this page explains how to install and use it:

    https://taosoft.dk/software/freeware/dnden/

     

    I guess my confusion is this can be applied to an already completed render correct? So if I have a completed render saved as .png I can use that as the source file to run it through the denoiser? Thank you for the link it's very appreciated.

    Yes, that's how you're supposed to use it.

  • Leonides02Leonides02 Posts: 1,379
    Padone said:
    TGFan4 said:

    It's already nowhere near "just blur".

    This is an example by pdxtobin of what happens to details when you don't use the albedo and normal buffers. It happens in the 4.11 denoiser too because it is just passing the beauty canvas.

    In general, this has been my experience, as well. The denoiser even in BETA isn't (AFAIK) using anything other than the beauty pass. That's why skin and clothes get hit with the "painterly" effect. I've done a fair amount of research and there seems to be no easy way to export an albedo from Daz Iray. We won't have true denoising until then, I don't think.

  • Matt_CastleMatt_Castle Posts: 2,571
    Padone said:

    Perhaps a "DIY" albedo could be done by turning all the materials to emitters so to emulate a flat shader, but it's just a guess.

    A sound theory, although even if it works well enough, trying to do that using Iray would mean having to render the albedo separately, at which point you might as well argue you should just render twice as long in the first place.

    However, taking the core idea, I've tried an attempt at faking an albedo pass using emissive materials in an OpenGL export, and the results of what the denoisers can achieve even with just that imperfect albedo AOV (with no proper reflection, refraction, depth of field, etc, and no normal AOV) do show staggering promise, with a huge leap in retained fine detail.

    It's really disappointing that even though Iray actually supports albedo AOVs, Daz doesn't provide it as an export canvas, because being able to feed the denoiser all of the parts of the picture (pun only slightly intended) would allow better results in less time.

     

  • TooncesToonces Posts: 919

    I took a screen shot of the pic and ran it through the external Intel denoiser (post-denoise result attached). The girl on the left didn't appear to lose any detail.

    DenoiseTest.png
    893 x 906 - 898K
Sign In or Register to comment.