LuxCoreRender 2.0


(Ace Dragon) #141

The Luxball denoising example shows a significant and highly noticeable blurring of some of the edges (notice the bottom half of the ball).

Does the Luxcore denoise make any use of normal and depth data? The Cycles denoiser tends to be pretty good at preserving edges like that (at that level of noise) unless the details are extremely small or packed together or are only found via a somewhat reflective surface).

If you’re trying to minimize the amount of data that needs to be fed into the algorithm, then you will likely need some very smart code to get a truly high quality result.


(LuxCore Addon Developer) #142

The stable release v2.0 is now available:
https://forums.luxcorerender.org/viewtopic.php?f=9&t=311

We’re also having a poll running to decide which features we implement next:
https://forums.luxcorerender.org/viewtopic.php?f=5&t=308

@Ace_Dragon: I answered you, but the forum switch did not carry over the post.


(Ace Dragon) #143

I just skimmed through the paper, I did notice what appears to be a few weaknesses that I would expect for an algorithm that does not take any form of pass or AOV information (namely, some smaller details as well as some fine detail and small highlights getting blurred). The same happens with the Cycles denoising (but you generally get better results with the relative filter checkbox off and more samples).

What I’m not sure about is how the denoising would handle regions with very similar colors or shading, but different values in terms of depth and normals (when considering how this type of information is not used at all here). My view on this type of thing is that it should only help if an algorithm had additional info that could help verify whether the noiseless approximation is correct.


(Dade) #144

What I’m not sure about is how the denoising would handle regions with very similar colors or shading, but different values in terms of depth and normals (when considering how this type of information is not used at all here). My view on this type of thing is that it should only help if an algorithm had additional info that could help verify whether the noiseless approximation is correct.

BCD uses something called “Ray Histogram Fusion” (RHF) to establish the zone to denoise together. It is something described in a paper before BCD: https://iie.fing.edu.uy/~mdelbra/rhf/delbracioTOG2014.pdf (you can find some code example in this paper: http://dev.ipol.im/~mdelbra/rhf/rhf_description_v1.1.pdf).

Depth and Normal related data lacks of many information RHF is able to capture.


(LuxCore Addon Developer) #145

A demo video of the upcoming denoiser.
Note that this is only Path CPU on a rather slow notebook CPU.
The render settings are also not entirely optimized as I realized after recording.


(LuxCore Addon Developer) #146

You can now test the denoiser in BlendLuxCore!
We have released v2.1alpha0 which includes the denoiser.


(chafouin) #147

Awesome, thanks for the hard work guys!


(English is not my native language) #148

Great! Thanks

A question not related to denoiser.
Is there a tutorial on how to correctly mapping procedural textures for this to be consistent throughout the animation? Should it be enough to configure as “Local” in 3D mapping on material nodes? Or should I use that node mode shown as “Luxcore texture nodes”?


(LuxCore Addon Developer) #149

If you want the procedural texture to “stick” to the object, use local 3D mapping.
Unfortunately, there is a bug in BlendLuxCore so they are not mapped correctly: https://github.com/LuxCoreRender/BlendLuxCore/issues/66

Or should I use that node mode shown as “Luxcore texture nodes”?

No, texture node trees are used when you want to use the same texture setup in multiple materials/volumes (they are a bit like Cycles node groups).


(English is not my native language) #150

Yes, that’s what I had noticed. In rendered preview the texture seemed to be consistent frame by frame, but then it changed between frames when Render animation.
Thanks.


(pafurijaz) #151

I downloaded now the last version the new alpha 2.1, with denoising…


(LuxCore Addon Developer) #152

We have released v2.1alpha1:
https://forums.luxcorerender.org/viewtopic.php?f=9&t=400

A short video about the new pause mode:


(sus_unn) #153

Nice to see this!!


(LuxCore Addon Developer) #154

@Dade is working on the light cache (part 1).

Working on Direct Light Sampling Cache. This is a rendering with white little spheres on cache entries receiving light and red glowing little spheres for entries in pitch black (from direct light sampling point of view):


(LuxCore Addon Developer) #155

It’s coming!


(Lsscpp) #156

NO WAY! It’s a joke, ins’t it? April’s fool in June maybe?


(Bernardo) #157

wow… this stuff looks extremely interesting!!


(Ace Dragon) #158

The thread talks about acceleration with direct lighting (so I’m not sure if the same speedup can be expected with the sampling of indirect light).

Though the move towards pre-processing and what resembles point-clouds is an interesting one when many other engines are moving away from such (like Renderman). This will mean additional controls users will have to worry about (an example being resolution settings).


(LuxCore Addon Developer) #159

We have released v2.1alpha2:
https://forums.luxcorerender.org/viewtopic.php?f=9&t=472#p4693


(cet77) #160

Wow… :slight_smile: Luxrender has become -very- impressive. :smiley: I’m happy…