Page 2 of 6 FirstFirst 1234 ... LastLast
Results 16 to 30 of 82

Thread: Open Image Denoise node for DP Filter

  1. #16
    Registered User
    Join Date
    Aug 2016
    Location
    a place
    Posts
    1,853
    Quote Originally Posted by Greatmangaka View Post
    Would be nice to see this baby for LW2015 32bit��
    yes, it would still life in the old x33/64 2015.3 atm i think someone said they were looking at pbr shaders for 2015 as well. i guess if they come out it will be a little more useful too.

  2. #17
    Does a EULA exist for this plugin? I would like to try it at my job but we require proper EULAs to use software.

    Thanks.
    My opinions and comments do not represent those of my employer.
    www.ernestpchan.com
    www.zazzle.com/gopuggo

  3. #18
    Registered User
    Join Date
    Mar 2018
    Location
    Perth, Australia
    Posts
    87
    I did some experiments and I just can't get the results that the other renderers get (Blender, Luxrender).

    Using dp's OIDN filter I render a beauty pass, an Albedo pass (RAW_RGB) and a normal pass. I combine them using all HDR (and linear for output) and the result is less then stellar.... It cleans up tiny areas but most of the picture looks like it has been attacked by some heavy jpg artefacts.

    I read the documentation and saw that you need to clamp the albedo pass to max values of 1. Did that but still no difference. I tried all the passes with the binary itself after converting them to PFM's but the result is the same. At least I know that the filter is working as the original ;-)
    Also I used the Box anti alias filter without adaptive sampling as I don't know if the Mitchel etc filters (or adaptive sampling) will work with the denoiser.

    Here is the result of a simple scene....
    Click image for larger version. 

Name:	OIDN-result.jpg 
Views:	94 
Size:	1.39 MB 
ID:	144548
    Not very clean at all!

    Here are the Albedo and Normal buffes I used (Normal buffer shows up wrong as I have normalized it. In the denoiser I use the Float version.)
    Click image for larger version. 

Name:	Albedo.jpg 
Views:	71 
Size:	506.6 KB 
ID:	144549

    Click image for larger version. 

Name:	Normal.jpg 
Views:	62 
Size:	33.5 KB 
ID:	144552

    I get extremely jealouse seeing the results of the other packages.... This has to be able to work with LW?

    It is quite crazy if you see how low the quality can be and what the output of the denoiser is!! Amazing!
    Click image for larger version. 

Name:	Interior.jpg 
Views:	105 
Size:	284.7 KB 
ID:	144553

    Click image for larger version. 

Name:	Plate.jpg 
Views:	104 
Size:	130.6 KB 
ID:	144554
    Last edited by peebeearr; 03-26-2019 at 12:42 AM.

  4. #19
    Registered User
    Join Date
    Mar 2018
    Location
    Perth, Australia
    Posts
    87
    ..... and to be honest... I expected that people around here would be on this thing like flies on sh*t!

    There is a lot of complaining about slow rendering with LW and it seems that everyone has had his/her say on the subject.... this could be a very good solution to the problem (it even holds up in animation I have seen!!!) and there is very little people experimenting with this. That quite baffles me.

    Anyway, I hope Denis or someone more knowledgeable than me can shed a bit of light on this?
    Last edited by peebeearr; 03-26-2019 at 12:46 AM.

  5. #20
    Member
    Join Date
    May 2006
    Location
    France
    Posts
    4,050
    Quote Originally Posted by peebeearr View Post
    ..... I hope Denis or someone more knowledgeable than me can shed a bit of light on this?
    Thanks for testings and comments.
    Just a note about linear, make sure you need this conversion.

    Could be interesting to test a similar scene, materials, lighting and reconstruction,
    with different PBR renderers using MC GI, and then applying the filter

    If you can use the binary, you can also report to the Github site,
    since this is a recent beta version.

    Denis.

  6. #21
    pass:sword OFF's Avatar
    Join Date
    Feb 2003
    Location
    Russia
    Posts
    1,014
    Very interesting tests, thanks. For me, OIDN filter is a chance to return to work my second dual-processor server, which is now idle, since I work with Octane renderer.

  7. #22
    Registered User gdkeast's Avatar
    Join Date
    Jul 2017
    Location
    Hawaii
    Posts
    149
    Thank you for all the work you do in keeping your amazing plug-ins current and improving them too. Donation of appreciation sent.

  8. #23
    Registered User
    Join Date
    Mar 2018
    Location
    Perth, Australia
    Posts
    87
    Quote Originally Posted by dpont View Post
    Thanks for testings and comments.
    Just a note about linear, make sure you need this conversion.

    Could be interesting to test a similar scene, materials, lighting and reconstruction,
    with different PBR renderers using MC GI, and then applying the filter

    If you can use the binary, you can also report to the Github site,
    since this is a recent beta version.

    Denis.
    Hi Denis,

    I'll see if I can get Blender up and running and export the scene file to blender and see how that works. To be honest, I don't think anything is wrong with the way I do it but rather the way Lightwave renders for some reason.

    I just grabbed some sample pictures (PNG) from a blender forum and loaded them in LW and ran the OIDN node over it... the results were as expected... pretty good results from such low input values! Smooth surfaces rather than the stuff I got.

    Here are some results from the downloaded PNG's in LW.

    Source image
    Click image for larger version. 

Name:	Basement.png 
Views:	61 
Size:	1.08 MB 
ID:	144562

    Result with only the colour being filtered
    Click image for larger version. 

Name:	Basement-result.jpg 
Views:	63 
Size:	620.1 KB 
ID:	144556


    Source image
    Click image for larger version. 

Name:	Beamer.jpg 
Views:	81 
Size:	720.1 KB 
ID:	144563

    Result with only the colour being filtered
    Click image for larger version. 

Name:	Beamer-result.jpg 
Views:	85 
Size:	218.2 KB 
ID:	144564


    Source image
    Click image for larger version. 

Name:	Gym.jpg 
Views:	72 
Size:	3.53 MB 
ID:	144559

    Result with only the colour being filtered
    Click image for larger version. 

Name:	Gym_Result.jpg 
Views:	77 
Size:	956.2 KB 
ID:	144560


    Then I started to look closely at how Blender outputs low sampled images and I see that it starts with dots in a black void whereas Lightwave (on its lowest setting) barely has any black space in it. Lightwave seems to refine differently than the other renders.

    Here is a closeup of a Lightwave render on its lowest settings and a low SPP Blender render:
    Click image for larger version. 

Name:	LW-vs-Blender.jpg 
Views:	69 
Size:	533.2 KB 
ID:	144561

    I cannot get smooth results with native LW renders at all and I wonder why that is.
    It would be pretty crap if we can't use this because of the way LW outputs its pixels... but that's hopefully me thinking that the beer glass is ALWAYS half empty

  9. #24
    Registered User
    Join Date
    Mar 2018
    Location
    Perth, Australia
    Posts
    87
    Here is a compare with the lowest quality settings of Lightwave (everything set to 1 sample and NO AA)

    Click image for larger version. 

Name:	Beauty-Compare.jpg 
Views:	57 
Size:	907.7 KB 
ID:	144565

    No real smooth surfaces..... and lots of mangled noise that is not because it is JPG compressed. It's there in the original as well.
    Raising the quality settings does not provide a better picture.

    The OIDN version kinda looks like an oil painting almost

  10. #25
    Registered User
    Join Date
    Mar 2018
    Location
    Perth, Australia
    Posts
    87
    Hmmmm, it seems that Filter Radius was the culprit here....

    See below results :-)

    Raw LW
    Click image for larger version. 

Name:	Table-org.jpg 
Views:	115 
Size:	2.42 MB 
ID:	144566

    OIDN version
    Click image for larger version. 

Name:	Table-OIDN.jpg 
Views:	122 
Size:	913.8 KB 
ID:	144567

    This is with Albedo and Normal buffers applied... all in all much better results.

    I'll keep on testing....

  11. #26
    Quote Originally Posted by peebeearr View Post
    Hmmmm, it seems that Filter Radius was the culprit here....

    See below results :-)

    Raw LW
    Click image for larger version. 

Name:	Table-org.jpg 
Views:	115 
Size:	2.42 MB 
ID:	144566

    OIDN version
    Click image for larger version. 

Name:	Table-OIDN.jpg 
Views:	122 
Size:	913.8 KB 
ID:	144567

    This is with Albedo and Normal buffers applied... all in all much better results.

    I'll keep on testing....
    +

    This looks amazing. Will this work on a farm?

  12. #27
    Registered User
    Join Date
    Mar 2018
    Location
    Perth, Australia
    Posts
    87
    I have done some more experimenting and found that Anti Aliasing is the most important control to keep detail when you use OIDN. Cranking up rays, reflections/refraction samples is not really adding that much to the clean up process but it does inflict render times.

    Also I did NOT use Adaptive AA in my tests... that got me higher render times. Because of the low rays and samples settings you can really crank up anti aliasing and still get good render times.

    The most important thing to remember is to turn down Filter Radius in the Camera Properties. That really screws up the OIDN process. Interpolated radiosity is ok to use but in my tests it did not get better quality renders or better render times

    Another important point is to make sure that you don't use bump maps but instead ONLY Normal maps (mapped to the Material node, NOT anywhere else) as the filter uses normals to determine what to filter. (check your normals in the buffer view).

    Attached are frames of animations I made with DOF (the USB Speaker scene) and Motion Blur (the Room scene). They're PNG's so you can see the actual amount of artefacting.

    Frame 11 - Heavy DOF blur
    Click image for larger version. 

Name:	OIDN_Final_Render_011.png 
Views:	61 
Size:	561.4 KB 
ID:	144586

    Frame 35 - smooth result from 1 minute render
    Click image for larger version. 

Name:	OIDN_Final_Render_035.png 
Views:	70 
Size:	670.7 KB 
ID:	144587

    Frame 16 - Smooth MB
    Click image for larger version. 

Name:	OIDN_Final_Render_016.png 
Views:	56 
Size:	1.47 MB 
ID:	144584

    Frame 105 - Jeez how ugly this room is!!! I could have used a nicer scene....
    Click image for larger version. 

Name:	OIDN_Final_Render_105.png 
Views:	61 
Size:	1.51 MB 
ID:	144585

    Conclusion:
    I like it a lot! I think it is very useful in cleaning up frames. I have other examples of interior renders only lit by portal lights and to get the noise out of those you need A LOT of samples!! With OIDN, you don't have to and high (4000px+) resolutions are actually very fast! In fact, the higher the resolution the better the filter works... or in other words, the less you notice the slight artefacts the filter creates.
    As I only have a Quadro 4000 I can't use the NVIDIA denoise option in 2019... but with this who cares!!!! Wish they made it multithreading though... would scream with my 32 cores!

    Although the filter does create some artefacts, in the animations you don't really notice them.

    Oh, and I think this should not be a problem in render farms either.....


    Suggestion to Denis, can you have the OIDN filter node default to HDR instead of LDR?? Also perhaps an option to have the Albedo map clip at 1 (or 100 or whatever LW uses) I did not see any difference not clipping it but the manual says to clip it and default LW cranks values to over 100% by default. That would probably the most used option. Anyway, you just earned some beers my friend!

    Now that everyone knows how it all works I REALLY like to see some awesome interiors rendered at blazing speeds and ultra smooth.

    Cheers.

  13. #28
    Very interesting tests, thanks for experimenting, i should have a look when i get the chance.

  14. #29
    Member
    Join Date
    May 2006
    Location
    France
    Posts
    4,050
    Quote Originally Posted by peebeearr View Post
    ...can you have the OIDN filter node default to HDR instead of LDR?? Also perhaps an option to have the Albedo map clip at 1 (or 100 or whatever LW uses) I did not see any difference not clipping it but the manual says to clip it and default LW cranks values to over 100% by default...
    Or you can add a preset (in the Image Filter node editor panel),
    including the OIDenoiser node parameters you want
    and any king of transformations, compositing you may need for buffers.


    Denis.

  15. #30
    Member
    Join Date
    May 2006
    Location
    France
    Posts
    4,050
    I modified the readme file just for the link of the Open Image Denoise,
    which has been updated to version 0.8.2, works like previous,
    no major change.

    https://github.com/OpenImageDenoise/oidn/releases/

    Denis.

Page 2 of 6 FirstFirst 1234 ... LastLast

Bookmarks

Posting Permissions

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