Shooting and Calibrating PBR Textures

[dj_buckley]

Martin at Racoon Artworks has just published a great article over on his site.  It's a follow up to one he published a year ago.  Something I've often thought but never had the technical ability to verify but it seems those PBR ranges of between 30-50 might not be quite so 'safe' after all ....

https://www.racoon-artworks.de/blog_PBRshootingandcalibrating.php

I've proofread this for him, but if anyone more technical spots any errors then reach out to him.

[Juraj]

I need to read this twice to really understand it :- D And for anyone who wants cheaper alternative, get the colors from CultureHustle. It's not on same level, but the super-white is really..really white.

I've had this talk with him about two years ago, fore the past article, and came to same conclusion then. The 50-230 RGB range that Substance popularized seems overly narrow range exactly because they interpreted Albedo differently than most rendering software devs did. Albedo always being measure of total reflectance (thus Specular Reflection Albedo + Diffuse Reflection Albedo), while Albedo of most PBR shaders in non-metallic mode being just the Diffuse part, on which the Specular part is added up. And then Corona itself does shanenigans like substracting a bit from that former value, which is super inconsistent feature.

(My personal opinion is that the rendering software should either never modify Albedo and expect users to somehow get physically correct maps&values from Megascan, X-Rite, etc... OR, always substract the base non-metallic reflectance of 4perc. not only if users use higher than 230 value).

And while above 230 seem rather rare, below 50 (4perc. linear reflectivity) seems to be quite common for lot of dark materials. Corten for example being example I've shown at SOA in 2014 while showing first IR build of Corona :- ) At that time, PBR wasn't yet a thing, and this was my guess-work.

My suggested approach for people who want to quickly extrapolate these maps from common photos (so not calibrated scans, personal or 3rd party), is to use the kind of tonemapping that gets you closest to what cameras do. And Cameras use almost extreme S-Curve contrast by default. Not even default ACES OT tonemapper than Corona uses now is contrasty enough to mimic it. If material with ACES OT + 3 Contrast (for example) looks like a photo under neutral exposure, you guessed the Albedo roughly correct.

[racoonart]

Thanks for posting it here!

I have no idea about corona doing anything under the hood, but for me, the only future-proof solution is to just not screw with Basecolor at all. These are measurements from the physical world and if we start doing a bit of adjustment here and a bit of it there we'll be in the same trouble as we've been so far. Proper values plus proper tone-mapping and there's no need for guesswork. That's the whole point of me bitching about it for 2 years already :D - and ideally it shouldn't matter if we're talking about offline, realtime or whatever program it is used in. This, of course, will only work if we can all agree on that we don't want to make our own special rules for our own special softwares.

[Juraj]

(If you mean screwing as us artists) Yeah but in reality clients send you iPhone photo of some engineered floor that is like nothing else and you have one day to integrate it into photorealistic rendering.
Having some tricks under-sleeve that let's you guess-work the de-lighting process into satisfying enough solution can be very handy.

(If you mean screwing as soft devs heh) Corona started correcting for high albedo about 4-5 years ago, long before CoronaPhysical material came. If you make 255/255/255 + Full 1.52 Reflectance... it will change the Albedo to something lesser, I don't know what or how much exactly. Not exactly sure if it still does that with Physical Material too.

[dj_buckley]

I have no idea about corona doing anything under the hood

maybe now's a good time for the devs to confirm what, if anything, the Physical material is doing under the hood in that respect.

[racoonart]

(If you mean screwing as us artists) Yeah but in reality clients send you iPhone photo of some engineered floor that is like nothing else and you have one day to integrate it into photorealistic rendering.

I just meant from the software side. Jobs will always be messy, clients and users will do whatever they want. The client may display your work on a shitty random LCD but you're still color grading it on a proper calibrated display regardless, don't you? We should establish proper methodology based on physical evidence even if users choose to not calibrate their Albedos and do random adjustments by eye. If we keep telling them not go below 50,50,50 in the first place then it's just 'garbage in - garbage out' with a random chance of screwing up the "right way". If you don't have measurements then you'll have to guess from intuition - that's fine, but at least you can build better intuition on properly shot and calibrated reference material (like a good portion of megascans appears to be).

[piotrus3333]

4% is 59 in sRGB 8 bit albedo texture.
also there is nothing wrong with available reflectivity data. if data says 4%, just feed 4% into the shader. the shader is where all the approximations are happening so just compare different options available. simple furnace test will clearly show you where the shader takes necessary shortcuts.

https://patapom.com/blog/BRDF/images/MSBRDFWhiteFurnaceTest.jpg

[piotrus3333]

properly shot and calibrated reference material (like a good portion of megascans appears to be).

also funny thing about Megascans. I consider them too dark across the board, my adjustment is +0.04.
this problem was mentioned in first deep dive vid about Lumen in UE5 years ago. funny part is they noticed only while making demo scene with Lumen. did not lead to any changes though to my knowledge.