Different Works in Progress

Post your tests, experiments and unfinished renderings here.
Post Reply
User avatar
lacilaci
Donor
Donor
Posts: 1969
Joined: Fri May 04, 2018 5:16 am

Re: Different Works in Progress

Post by lacilaci »

FarbigeWelt wrote: Thu Nov 01, 2018 8:20 pm
lacilaci wrote: Thu Nov 01, 2018 11:52 am There is no point digging any deeper into this. It works as advertised, I've been looking over the internet and this is simply what it is.
It is not using any extra information that would make it preserve details and no feature recognition so it will never work as any other denoiser in other renderers like vray, corona, cycles, optix etc.
There is no magical settings to make it better, it can be less blury/more noisy or less noisy/more blurry... that's it. And everyone coming from another renderer will be banging their head against the wall questioning what is going on.
You can take your noisy render to lightroom, darktable, gimp whatever and denoise there with just as good result.
This is what proper denoising looks like:
https://www.youtube.com/watch?v=4FZXKNBWxQs
and redshift example: https://www.youtube.com/watch?v=ofcCQdIZAd8
If I were happy with Lightroom or Photoshop denoiser I would not look closer at LuxCoreRendets denoiser.
I dont‘t agree with your statement noisy or blurry is the only options with LuxCoreRender but it seems quite right for Lightroom or Photoshop.
Are you sure, have you tried? And do you know that lightroom has a "denoiser" that is actually not so good and not even remotely suitable for monte carlo type of noise and yet with a single slider it can do a comparable job even on a very noisy render? Does no one see a problem here?
Attachments
bcd_vs_lightroom.jpg
User avatar
FarbigeWelt
Donor
Donor
Posts: 1046
Joined: Sun Jul 01, 2018 12:07 pm
Location: Switzerland
Contact:

Re: Different Works in Progress

Post by FarbigeWelt »

B.Y.O.B. wrote: Fri Nov 02, 2018 1:43 am
FarbigeWelt wrote: Thu Nov 01, 2018 10:08 pm Scales above 9 leads to Blender crash at the moment denoising starts after render halt.
Scales above 9 is an insane number.
You should rarely need more than the default of 3.
What this parameter does is it downsizes the image and factors this in during denoising, to remove low-frequency noise (splotches).
So a scale of 3 creates the original res, half res, and 1/4th res.
In this case, please set a reasonable range for the value of say 2 to 8 to avoid crashes and add explanation to tool tip in next version. THX.
Light and Word designing Creator - www.farbigewelt.ch - aka quantenkristall || #luxcorerender
MacBook Air with M1
User avatar
FarbigeWelt
Donor
Donor
Posts: 1046
Joined: Sun Jul 01, 2018 12:07 pm
Location: Switzerland
Contact:

Re: Different Works in Progress

Post by FarbigeWelt »

lacilaci wrote: Fri Nov 02, 2018 5:30 am
FarbigeWelt wrote: Thu Nov 01, 2018 8:20 pm
lacilaci wrote: Thu Nov 01, 2018 11:52 am There is no point digging any deeper into this. It works as advertised, I've been looking over the internet and this is simply what it is.
It is not using any extra information that would make it preserve details and no feature recognition so it will never work as any other denoiser in other renderers like vray, corona, cycles, optix etc.
There is no magical settings to make it better, it can be less blury/more noisy or less noisy/more blurry... that's it. And everyone coming from another renderer will be banging their head against the wall questioning what is going on.
You can take your noisy render to lightroom, darktable, gimp whatever and denoise there with just as good result.
This is what proper denoising looks like:
https://www.youtube.com/watch?v=4FZXKNBWxQs
and redshift example: https://www.youtube.com/watch?v=ofcCQdIZAd8
If I were happy with Lightroom or Photoshop denoiser I would not look closer at LuxCoreRendets denoiser.
I dont‘t agree with your statement noisy or blurry is the only options with LuxCoreRender but it seems quite right for Lightroom or Photoshop.
Are you sure, have you tried? And do you know that lightroom has a "denoiser" that is actually not so good and not even remotely suitable for monte carlo type of noise and yet with a single slider it can do a comparable job even on a very noisy render? Does no one see a problem here?
In my opinion your BCD image looks better than lightroom image. And yes, there is a workpath issue if I had to post process every image of an animation. However, in earlier posts I refered to photoshop for removing fireflies, and also used color or luminance denoiser of lightroom. As said, I was never realy happy with denoised but less sharp results.

Currently, there is a short animation rendering. I was able to reduce total render time for HD by factor ~3 enabling denoiser and adusting halt time. Initial time 40 min per frame, current time 12 min per frame, preserving details. Guessing on GPU denoiser would run 5 to 10 times faster on my system, it will be a prefered solution.
Light and Word designing Creator - www.farbigewelt.ch - aka quantenkristall || #luxcorerender
MacBook Air with M1
User avatar
lacilaci
Donor
Donor
Posts: 1969
Joined: Fri May 04, 2018 5:16 am

Re: Different Works in Progress

Post by lacilaci »

FarbigeWelt wrote: Fri Nov 02, 2018 7:44 am
lacilaci wrote: Fri Nov 02, 2018 5:30 am
FarbigeWelt wrote: Thu Nov 01, 2018 8:20 pm

If I were happy with Lightroom or Photoshop denoiser I would not look closer at LuxCoreRendets denoiser.
I dont‘t agree with your statement noisy or blurry is the only options with LuxCoreRender but it seems quite right for Lightroom or Photoshop.
Are you sure, have you tried? And do you know that lightroom has a "denoiser" that is actually not so good and not even remotely suitable for monte carlo type of noise and yet with a single slider it can do a comparable job even on a very noisy render? Does no one see a problem here?
In my opinion your BCD image looks better than lightroom image. And yes, there is a workpath issue if I had to post process every image of an animation. However, in earlier posts I refered to photoshop for removing fireflies, and also used color or luminance denoiser of lightroom. As said, I was never realy happy with denoised but less sharp results.

Currently, there is a short animation rendering. I was able to reduce total render time for HD by factor ~3 enabling denoiser and adusting halt time. Initial time 40 min per frame, current time 12 min per frame, preserving details. Guessing on GPU denoiser would run 5 to 10 times faster on my system, it will be a prefered solution.
you win I give up
User avatar
FarbigeWelt
Donor
Donor
Posts: 1046
Joined: Sun Jul 01, 2018 12:07 pm
Location: Switzerland
Contact:

Re: Different Works in Progress

Post by FarbigeWelt »

lacilaci wrote: Fri Nov 02, 2018 8:00 am you win I give up
Well, okay.
But, the story is not finished yet. There are further parts to look at, and they seem to ask for a few improvements.
Light and Word designing Creator - www.farbigewelt.ch - aka quantenkristall || #luxcorerender
MacBook Air with M1
User avatar
FarbigeWelt
Donor
Donor
Posts: 1046
Joined: Sun Jul 01, 2018 12:07 pm
Location: Switzerland
Contact:

Closer look at Denoiser, last part

Post by FarbigeWelt »

This is the last part of the closer look at denoiser, at least for a while.

With Scales above 1 there appears a yellow border around the blue laser spot.
But Scales of 1 does not make much sense and leads to color bleeding.
Yellow border is hardly visible without zooming in, so its not really an issuse in this image.

Image resolution 0.5 HD.
Denoised HDT 1.0, SWR 6, Scales 2, 4, 8 comparison
Denoised HDT 1.0, SWR 6, Scales 2, 4, 8 comparison
I've choosen Scales 3 and 4 for the HD renders because 3 is default and 4 looks sharper.
One image is based on less samples, I did not rerender the image because the 'borderline' smoothing with Scales 3 is similar to the one in 0.5 HD image.
Denoised HD, HDT 1.0, SWR 6, Scales 3, 4 comparison
Denoised HD, HDT 1.0, SWR 6, Scales 3, 4 comparison
Actually, render should have stopped at noise threshold of 10. But I have no idea how this threshold is related to convergence level, one image reached 0.998 with 5100 Samples. Because of, intentionally, nasty material properties the image should keep noisy, no matter how many Samples have been rendered.
Light and Word designing Creator - www.farbigewelt.ch - aka quantenkristall || #luxcorerender
MacBook Air with M1
User avatar
FarbigeWelt
Donor
Donor
Posts: 1046
Joined: Sun Jul 01, 2018 12:07 pm
Location: Switzerland
Contact:

Re: Different Works in Progress

Post by FarbigeWelt »

kintuX wrote: Thu Nov 01, 2018 8:46 pm @FarbigeWelt
Thank you for everything you're posting in this thread. The world is truly colorful :D Keep it up!
Hi kintuX

Thank you much for the feedback.

Regards
FarbigeWelt
Light and Word designing Creator - www.farbigewelt.ch - aka quantenkristall || #luxcorerender
MacBook Air with M1
User avatar
FarbigeWelt
Donor
Donor
Posts: 1046
Joined: Sun Jul 01, 2018 12:07 pm
Location: Switzerland
Contact:

Projected Image Missing, spot light, animation

Post by FarbigeWelt »

Known Issue: Projected image of a spot light is missing or changes its brightness depending on frame number.
see also viewtopic.php?f=4&t=592&p=5894#p5887
According to B.Y.O.P., this effect cannot be reproduced on some systems and root cause is currently unknown.
In animations with projected images same frames are affected reproducible.

Workaround trials

Code: Select all

Increasing importance of spot light	>helps sometimes
Disable auto brightness 		>is no solution
Rerender frame				>may help for stills (F12) but not for animations
Different tonemapper			>is no solution
Disable denoiser			>is no solution, effect happens with BiDir metropolis too
Does anybody missing projected images, with the animation provided here, too?
Denoiser Tests_2.zip
Scene: Denoiser Test 2, settings Test_f
(678.58 KiB) Downloaded 172 times

Code: Select all

LuxCoreRender 2.1 beta 1
Blender 2.79b, MS Windows 10 Pro;			Remove Fireflies enabled
CPU: 3.5 GHz, i7-4770K, Intel;				Histogram Distance Threshold 1.0
GPU: 1.0 GHz, 2xHawaii, AMD, 4GB			Search Windows Radius 6
RadeonSW Adrenalin 18.9.3;				Scales 4
OpenGL/OpenCL 24.20.11000;				Patch Radius 1
	
Render Engine: GPU Sobol, Adaptive Strength 0.95;	Composite 0.8 denoise 0.2 raw
Total Path Depth 10;					Frame 01, Save time 13:38
Diffuse 5, Glossy 5, Specular 20;			Frame 40, Save time 14:57
;							Animation render time ~01h23m  
Scene: VRAM 833 MB, 8 Lights, Projector Importance 2.0;
Resolution 1440x810, 0.75xHD, 1.16 MPixels;		Per frame
Border no, Crop no, Filter none, Clamped 500.0;		Halt ~500 Samples
Reinhard, Pre 1.0, Post 0.8, Burn 6.00;			Warm up time ~00m20s
Exposure 0.2, Gamma 1.0;				Render time 01m05s, 01m11s
;							Total time 01m54s, 02m01s
;							Denoise time 00m39s, 00m40s
Animated Gif: 256 bit, dithered, 30% loss (Adobe Photoshop Open image sequence 10 frames per s, Export>Export for Web)
File size of image and animated Gif depends much on noise level, the reason here for the different resolutions.
Legend: den denoised, 0.8 cmp composite of 0.2 raw 0.8 denoised
Denoiser-Test_f_0720x405_3.5 MB_0.8-cmp
Denoiser-Test_f_0720x405_3.5 MB_0.8-cmp
Denoiser-Test_f_0720x405_4.7 MB_raw
Denoiser-Test_f_0720x405_4.7 MB_raw
The attachment Denoiser-Test_f_0720x405_3.5 MB_0.8-cmp.gif is no longer available
EDIT: Tip, use Courier font e.g. in Notepad for formatting code text.
Light and Word designing Creator - www.farbigewelt.ch - aka quantenkristall || #luxcorerender
MacBook Air with M1
User avatar
B.Y.O.B.
Developer
Developer
Posts: 4146
Joined: Mon Dec 04, 2017 10:08 pm
Location: Germany
Contact:

Re: Different Works in Progress

Post by B.Y.O.B. »

Can you test if the same happens with a textured point light?
User avatar
FarbigeWelt
Donor
Donor
Posts: 1046
Joined: Sun Jul 01, 2018 12:07 pm
Location: Switzerland
Contact:

Re: Different Works in Progress

Post by FarbigeWelt »

B.Y.O.B. wrote: Sat Nov 03, 2018 4:04 pm Can you test if the same happens with a textured point light?
Yes, I used the same scene but this time with Point Light(IES and image) instead of Spot Light. Blue and green shades among others are from the point light. Point Light is stable for at least 40 frames.
Denoiser-Test_g_0720x405_4.7-MB_raw.gif
But denoiser does not like this kind of image as it looks like.
Image0001.png
Light and Word designing Creator - www.farbigewelt.ch - aka quantenkristall || #luxcorerender
MacBook Air with M1
Post Reply