LuxCoreRender Forums

"Bump Terminator"

We present a technique to hide the abrupt shadow terminator line when strong bump or normalmaps are used to emulate micro-geometry. Our approach, based on microfacet shadowingfunctions, is simple and inexpensive. Instead of rendering detailed and expensive height-fieldshadows we apply a statistical solution built on the assumption that normals follow a nearlynormal random distribution. We also contribute a useful approximate variance measure forGGX, which is otherwise undefined analytically.

source: A Microfacet-Based Shadowing Function to Solvethe Bump Terminator Problem

and much more... SIGGRAPH 2019 Ray-Tracing Roundup

I'm gonna sacrifice a goat and maybe a dog to have the gods of shaders create a treatment for this shadow terminator. Really, whenever I change lighting I inspect my renders and pray that this f*ing issue won't come up.

This seems really easy to implement. They provide a listing of their code and describe how to add it into their renderer(using it as a lighting term multiplier). Im sure the dev's will get to this soon to at least give it a go.

They do mention that this breaks reciprocity of light tho so this may be better to enable as an optional fix

patrickawalz wrote: ↑Sat Jun 22, 2019 4:23 pm They do mention that this breaks reciprocity of light tho so this may be better to enable as an optional fix

Does this mean it breaks the "bidirectional" nature of a BxDF?

Also, does this only apply to bump mapped surfaces or does it also work on any other surface? Because we have a general problem with the terminator, not only when bump mapping is used.

B.Y.O.B. wrote: ↑Sat Jun 22, 2019 4:43 pm

patrickawalz wrote: ↑Sat Jun 22, 2019 4:23 pm They do mention that this breaks reciprocity of light tho so this may be better to enable as an optional fix

Does this mean it breaks the "bidirectional" nature of a BxDF?

Also, does this only apply to bump mapped surfaces or does it also work on any other surface? Because we have a general problem with the terminator, not only when bump mapping is used.

Yes they found that applying this as a scaling factor to both the light and bsdf produced darker edges than they would like so they are only applying it to lights.

And yes this fix seems only for the terminator when a bump map is involved.

There is a new article I'll have to look up that is being presented at siggraph called "taming the shadow terminator" that might be a good solution to the general case but it isn't even in preprint yet

Thanks for the insight.
What do you think about this earlier post on the topic? viewtopic.php?f=9&t=119&p=1969#p1969

patrickawalz wrote: ↑Sun Jun 23, 2019 12:55 am There is a new article I'll have to look up that is being presented at siggraph called "taming the shadow terminator"

I've only seen the comparison picture from that and it's already mind blowing

B.Y.O.B. wrote: ↑Sun Jun 23, 2019 7:14 am Thanks for the insight.
What do you think about this earlier post on the topic? viewtopic.php?f=9&t=119&p=1969#p1969

It sounds like they are doing some version of dynamic epsilon along the same lines as to what we have already been doing since machineepsilon was introduced in the original lux years ago, albeit with some form of interpolation. But theas was more a feature description rather than a technical report, with no explanation of how they really achieved it. My money is on the upcoming 'taming the shadow terminator' talk as hopefully it will give some of the details on how it was done. It's a talk tho, so it might just be some PowerPoint slides in the end without any context, but at least one of its authors ( yining karl li) is an active blogger so it might be possible to contact him and get insights after the talk is presented.

Alright, looking forward to the presentation then.

Interesting: "Optimal Multiple Importance Sampling" (https://graphics.cg.uni-saarland.de/pub ... l-mis.html)