The delusional world of PBR

Become a member of the CGSociety

Connect, Share, and Learn with our Large Growing CG Art Community. It's Free!

REPLY TO THREAD
 
Thread Tools Search this Thread Display Modes
  3 Weeks Ago
Originally Posted by kgs7165: One more reason why current PBR workflow is not actually "physically based".

Textures.

Think about how major vfx / animation / game studios make their diffuse textures. Take whatever on google. cgtextures.com, even you took photos with your DSLR in ambient lighting, these pictures have specular on the images. Not polarized(you get this sometimes for digi double work). and how you usually make spec / gloss map? Starting with making diffuse into grey scale. Sounds familiar right? So in rendering, you're adding spec on top of spec. of course you can CC the source images look more diffussy- but on that point, not really physically accurate.

Sure I was thinking the same thing about subsurface scattering textures.. of course a plain diffuse texture has already subsurface scattering in it so generally you add SSS to on top of SSS and that's way more difficult to remove it by color correcting the texture .. but that doesn't mean the 'tech' to better approach this ain't already available.. and I'm sure top studios in a way or another are already using it.

http://wp.doc.ic.ac.uk/rgi/project/...t-field-camera/

Another thing you may not know is about microfacet BSDFs (resposable for your blurry reflections/refractions).. a truly physical BSDF is multi-scattered however most of the BSDFs outthere are just single scattering.. they are still called 'physically based' just because microfacet theory itself is considered to have a physical foundation but you still lose a lot of energy in most cases. That's why generally the 'metalness' parameter in a PBR workflow is over-used, you add a tiny bit of diffuse to compensate for that energy loss even on pure rough metals and glasses. But again this doesn't mean multi-scattered BSDFs ain't already available (Cycles too have them ie).

https://www.rombo.tools/2017/03/15/...cattering-bsdf/

ciao, max

Last edited by RomboStudios : 3 Weeks Ago at 04:33 PM.
 
  3 Weeks Ago
>BSDFs outthere are just single scattering

RomboTools uses the Smith model ? Any idea what the perf impact is over single-scatter ? There is also no analytical derivation for this model right ?
 
  3 Weeks Ago
Originally Posted by RomboStudios: most of the BSDFs outthere are just single scattering..

I know this was the case back in 2000-2007 ... but is this really still the case?

Last edited by ACiD80 : 3 Weeks Ago at 01:19 AM.
 
  3 Weeks Ago
Originally Posted by ACiD80: I'd suggest you read the link "kgs7165" posted.
I'll repost the link for you: https://www.chaosgroup.com/blog/the...iased-rendering

You'll find that pretty much all renderer used in production (need to be fast) are biased.

Just to clarify, are you claiming that Arnold is a biased renderer? I think that's an important thing for you to say outright.

If so, then I don't think we have much further to discuss on this topic. I'm not interested in debating the fine details of what counts and biased/unbiased based on an article by a lead contributor to VRay. I think the vfx industry at large has an acceptance of what unbiased and biased mean, in the functional terms of everyday rendering requirements. It's a term we can use in the office and people understand the implications and why we would want to be as unbiased as possible.

I think anyone who says that vfx hasn't had an extremely strong swing to unbiased rendering in the last 4 years, specifically in regards to Arnold, is wrong. If you claim that Arnold and other unbiased renderers are actually technically biased, they just minimise it, then I think you're missing the point.

If I seem passionate on this point it's because I am literally, working at the moment on a project with people who have won both technical Oscars for their contributions to CG lighting, and VFX Oscars and nominations for their output (I am the studio side supervisor, so I'm their client for reference) and they talk me through alternative methods when I want to change shit in comp, or with certain AOVs, or shaders, because their commitment to PBR and to being as unbiased as possible has changed the way they work. You may say that, technically, they're actually using biased renderers. I will say that they would take offense to that because of what it implies ... which is just weird because (as I've said many times) I think stuff can look good in almost any renderer if you want it too.

I hope that gives some context to why I say this is a deeper change in the industry and that some people take this quite seriously.
__________________
Critcal feedback example #62: "Well instead of the Stalinist purges and the divorce and the investigation ... it could be about losing a balloon."
 
  3 Weeks Ago
Originally Posted by axiomatic: Just to clarify, are you claiming that Arnold is a biased renderer? I think that's an important thing for you to say outright.

Yes & no...
it probably can be configured to be unbiased(?) but no one does that.

Originally Posted by axiomatic: I'm not interested in debating the fine details of what counts and biased/unbiased based on an article by a lead contributor to VRay
So, you're always right... Even when you're wrong. Nice argument. I see you are very keen on learning and being correctly informed.
The vray devs have a no-bullshit reputation, that's a big part why they are so loved.
The dev's of one of the most popular renderers out there (which won an academy award) kind of know what they are talking about.

Originally Posted by axiomatic: I think the vfx industry at large has an acceptance of what unbiased and biased mean, in the functional terms of everyday rendering
requirements. It's a term we can use in the office and people understand
the implications and why we would want to be as unbiased as possible.

That's quite a few no's there
More something like mass- confusion and misinformation about the topic.
That's why its important to try and get it right here... The "whatever, i'm always right" attitude isn't really adding to that.

Originally Posted by axiomatic: I think anyone who says that vfx hasn't had an extremely strong swing to unbiased rendering in the last 4 years, specifically in regards to
Arnold, is wrong.

Yes and no... (again)
Unbiased rendering first got lots of public attention when renderers like maxwell came out.
Marketing departments started hyping things like spectral rendering, physical correct rendering, unbiased rendering etc...
Spreading a lot of half-truths and even misinformation to brainwash people into thinking these tings are the biggest must-haves of all time without really explaining what it is or does.
So people didn't really know/understand what unbiased rendering really means but what they saw happening on the screen that was visually different was maxwell's 'new' way of rendering so in their minds they started to link it to that.
That's why so many people still think that progressive path tracing = unbiased rendering, for example (which it isn't).
Some think linear workflow is unbiased rendering (again, nope)... some think physical based rendering is unbiased rendering (nope)... spectral rendering (nope)...

All because someone started hyping a term without explaining it and thus people automatically linking it to what they saw happening differently on the screen and spreading that around the net.
Just google it and see how many different wrong explanations you get.

That's also why (especially you being a supervisor) when someone that is a dev of one of the most popular renderers out there takes the time and effort to try and fix all the misinformation out there you should maybe take a minute or 3 to inform yourself correctly.

Computers are getting faster and so renderers can use more unbiased techniques where it really helps, so that's nice. I think we all can agree on that.

Originally Posted by axiomatic: If you claim that Arnold and other unbiased renderers are actually technically biased, they just minimise it, then I think you're missing
the point.

What?! :suprised:
A renderer that uses 'minimized biased techniques' is still, obviously' biased.
And you're telling me I'm missing the point... right

Last edited by ACiD80 : 3 Weeks Ago at 02:03 PM.
 
  3 Weeks Ago
Originally Posted by ACiD80: So, you're always right... Even when you're wrong. Nice argument. I see you are very keen on learning and being correctly informed.

You're now attacking me personally, which is exactly why I previously decided to delete my comment and stop arguing. I regret that I let Roberto convince me to explain my position.

For what it's worth, I've repeatedly said that I agree with you regarding the over-hyping of unbiased rendering and the advantages over biased methods that it provides, although I find your assertion that unbiased rendering is just popular because of marketing hype to be somewhat absurd. My point has always been that unbiased rendering, as a methodology, is something artists pursue because of the same type of mindset that has them talking about PBR. These things encapsulate a change in the old approach to CG in VFX, and I believe it's having a huge impact on the industry.

When I say I'm not interested in debating the fine details of the differences between unbiased and biased, it's because of exactly the kind of response you've provided. You can argue that Arnold is an biased renderer, I could spend time trying to argue the opposite, but it isn't important to my point. What is important to me is that, even when using something like VRay, people want it to render as unbiased as they can within the allowable limits. That is a mindset and it's dominant in the industry - the industry has changed so that it's pro the methodology of unbiased even if it's unattainable.

I should add that I feel that I've tried to make a number of concessions in this discussion. I think my problem during this discussion has been that my argument isn't really on point enough. It's clear you guys want to argue the inaccuracies of these terms, while I want to merely point out that those inaccuracies are not as important as the simple fact that these terms are being used, and are being persued, and I think that is in and of itself a significant change that impacts all levels of the vfx pipeline.

But, in an effort to avoid more of your sarcasm and derision, I will be as clear as I can and agree with you that unbiased rendering is more hype than substance. I'm happy to leave uncontended your comments that Arnold is a biased renderer as I cannot point to evidence to the contrary. And I can also agree m with you that PBR is both over used and rather inaccurate terminology, given that CG rendering is still far from physically accurate.

Aand last but not least ... I appreciate the insight you've given into the topic, and it's been an interesting discussion in that respect, but this has become more adversarial than I'd like so I'm going to respectfully decline to continue commenting further.
__________________
Critcal feedback example #62: "Well instead of the Stalinist purges and the divorce and the investigation ... it could be about losing a balloon."
 
  3 Weeks Ago
Originally Posted by axiomatic: You're now attacking me personally, which is exactly why I previously decided to delete my comment and stop arguing. I regret that I let
Roberto convince me to explain my position.

You can't just expect everyone to just agree with you especially when you are choosing to ignore facts and label them as 'debating the fine details' when they are exactly explaining what is being talked about but they happen to disagree with you.

Originally Posted by axiomatic: You can argue that Arnold is an biased renderer, I could spend time trying to argue the opposite, but it isn't important to my point.
You were the one asking if I think Arnold is biased or not? ...
 
  3 Weeks Ago
How much of an arsehole do you have to be?

I've acknowledged every single point you've made, and have flat out agreed with almost all of them. Where I disagree I've even acknowledged you have valid points. On the other hand you haven't addressed a single one of my own observations about the fundamental changes in the industry that I believe result from the change in mindset - you've completely ignored everything I've said and instead keep going over the same points, about how you're right, and about you know more about this than me. You think I'm trying to argue with you when I'm trying to engage in productive conversation. This is bizaare.

But, jesus christ, now I remember why I took a break from these forums. It's so stressful having to have discussions with people like you. All self rightious and full of desire to proove their superiority. Makes me want to delete my account. I'm out of here.
__________________
Critcal feedback example #62: "Well instead of the Stalinist purges and the divorce and the investigation ... it could be about losing a balloon."
 
  3 Weeks Ago
Guys keep it easy.

Practically in CG, Biased vs Unbiased arised in the times where pathtracing was getting a new hype and was compared against other stuff like finalgathering, photon mapping, hierarchical diffusepoint based bleeding and other techs to provide Global Illumination. The artifacts output of an 'unbiased' renderer was and still is noise where for the other can be any sort of splotch, light leak and so on. So in this regard Arnold is 'unbiased'. I think this is what Axiomatics is trying to say and he's perfectly right in this regard. However then practically for what the video posted above shows, - Arnold is a fully biased renderer. Much more than others. It does fully sample for example only the first reflection bounce, it endorses sample clamping, SSS is not pathtraced etc. And why all of this ? Because it perfectly works ! Here biased means being smart. And this is what ACID80 is probably saying. So you are both correct (and of course both wrong).

About microfacets physically based models, it's some 50 years they are here coming from optical science fields and later generalized into BSDFs around 2000. However practically in CG nobody used them until 10 years ago or so. Speculars were light loops(ie. not raytraced) and reflections were fuzzy/blurry reflections based on simple qmc random directions, no importance or multiple importance sampling. Multiple scattering is from 2014, It has maybe 1% perf impact over single scattering in biased renderers and it means practically something only for rough materials. Something new we're introducing in Rombo based on a mix of latest research papers is a gBSDF, aka a generalized BSDF where we don't have anymore to specify a GGX or Beckmann distributions but we have a visual parameter called Slope to fully model our rough refl distributions, the cool thing is that the parameter is texturable so we can have different distributions (that visually means the shape or lobe of our refls) on the same surface. Think about. You want to model a fully polished metal with scratches. In the actual PBR workflow we supply a roughness and a reflection map to model that and most of the times.. also a metalness map. Why ? Because without that it doesn't visually work so great so we wanna believe that a scratch in a metal exhibits a diffuse behaviour.. but it's not necessarly.. it just exhibits a different reflection shape because not super polished.

In this regard I think that any artist outthere keep pushing for visual realism until something new makes it better and easier and feeling being part of a whole workflow that somehow make sense and brings efforts together. However don't forget at that point that nothing was and nothing will be ever 'physical' in the naive meaning you wanna see it. It simply is 'physically based' that simply means a model that can reproduce and predict natural behaviours and phenomenons. And even when better models will be adopted beauty render outputs should be still seen as 'render data', it is actually an 'art' to turn it into 'visual data' that can beentertaining etc also outside the scientific community for the general people.

Last edited by RomboStudios : 3 Weeks Ago at 05:14 PM.
 
  3 Weeks Ago
Originally Posted by RomboStudios: Guys keep it easy.

Practically in CG, Biased vs Unbiased arised in the times where pathtracing was getting a new hype and was compared against other stuff like finalgathering, photon mapping, hierarchical diffusepoint based bleeding and other techs to provide Global Illumination. The artifacts output of an 'unbiased' renderer was and still is noise where for the other can be any sort of splotch, light leak and so on. So in this regard Arnold is 'unbiased'. I think this is what Axiomatics is trying to say and he's perfectly right in this regard.

This is where I disagree and pls if I'm wrong I'm all open to get corrected and learn new stuff.

Path Tracing isn't always unbiased, in fact most of the time it isn't.
For example:
1. Path Tracing can use adaptive sampling (which most renderers do afaik), which is biased.
2. Since a few years many of them clamp the indirect samples to fight fireflies, which is biased.
3. Even limiting the amount of bounces turns out to be biased.
4. etc...
And if you look at arnold, I'm pretty sure it applies 1 & 2... I'm not sure about 3 (not sure if you can turn them off)

Last edited by ACiD80 : 3 Weeks Ago at 05:35 PM.
 
  3 Weeks Ago
It's exactly what I'm saying if you kept reading the very same paragraph
 
  3 Weeks Ago
Originally Posted by RomboStudios: It's exactly what I'm saying if you kept reading the very same paragraph

But you also said this, which is confusing me:
Quote: The artifacts output of an 'unbiased' renderer was and still is noise where for the other can be any sort of splotch, light leak and so on. So
in this regard Arnold is 'unbiased'
To me it looks like you are suggesting just because arnorld doesn't use interpolated sampling that this automatically makes this part unbiased?
If, so, that's what I don't agree with that because the samples get 'influenced/altered' using (before mentioned) biased techniques before they contribute to the noise and thus the resulting noise is not a biased result.

Last edited by ACiD80 : 3 Weeks Ago at 06:02 PM.
 
  3 Weeks Ago
People, at the time where first usable pathtracers were coming out, were simply saying they preferred 'unbiased' renderers because noise is getting converged at some point where other kind of artifacts are not. In this regard I'm trying to understand Axiomatic on why he perceives Arnold as unbiased. For me biased vs unbiased does not mean anything at all, everything is of course biased. Cheers.

Last edited by RomboStudios : 3 Weeks Ago at 06:09 PM.
 
  3 Weeks Ago
Originally Posted by RomboStudios: People, at the time where first usable pathtracers were coming out, were simply saying they preferred 'unbiased' renderers because noise is getting converged at some point where other kind of artifacts are not. In this regard I'm trying to understand Axiomatic on why he perceives Arnold as unbiased. For me biased vs unbiased does not mean anything at all. Cheers.

Well, this has been my point all the time, this has been perceived as unbiased and many people still think this, but wrongfully so.
 
  3 Weeks Ago
Kris if the distiction of biased vs unbiased doesn't make too much sense nowdays also right vs wrong remains kinda a bit limited. We found as humans and thinkers and scientists that the big True doesn't exist but that doesn't mean we are all wrong all the time. Nowdays what makes things 'true' is the coherency of the system or in case of human life and natural languages, being coherent and consistent about our own point of view. In this sense I don't feel axiomatic is 'wrong', he just uses a word that doesn't make too much sense but he's notably coherent on everything is saying. That's also what makes kinda 'true' a PBR approach.. as language, intentions, tools and actions.. it brings things done for what is the state-of-the-art visual level that people are expecting. For anything left.. guys keep rendering and have an happy halloween !
 
reply share thread



Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

vB code is On
Smilies are On
[IMG] code is On
HTML code is Off
CGSociety
Society of Digital Artists
www.cgsociety.org

Powered by vBulletin
Copyright 2000 - 2006,
Jelsoft Enterprises Ltd.
Minimize Ads
Forum Jump
Miscellaneous

All times are GMT. The time now is 06:14 AM.


Powered by vBulletin
Copyright ©2000 - 2017, Jelsoft Enterprises Ltd.