Fix the limit for interpolation of R0 with respect to metallic and the calculation of the cos theata in the Fresnel Shlick term in SSR

[n0tank3sh]

The default of Ro is assumed to be around 1.5 after applying the formula which is ((u -1)/(u +1))^2 => ((1.5 -1)/(1.5+1))^2=>0.04. The element with the highest refractive index found in nature is considered to be Germanium. It has a range of refractive index from 4.05 to 4.1. After applying the formula to the maximum of the range ((4.1 - 1)/(4.1 +1))^2=>(Approx)0.37. So, the rational limit should be this.
Fix #75302

[RedMser]

The rationale for 0.37 should probably be mentioned in a code comment, since it currently looks like an arbitrarily chosen magic number.

[logzero]

FYI: The refractive index of Germanium listed there is for infrared. For visible light diamond with 2.4 would be the limit.

[n0tank3sh]

@logzero FYI: Fresnel equation is about reflection, not refraction.

[jcostello]

@MightiestGoat Before and after?

[n0tank3sh]

@jcostello wdym?

[jcostello]

@MightiestGoat how it looks before and after the fix

[armmah]

I tried using proper f0 and f90 like in frostbite paper as well, as it stood out to me on first glance. However as could be expected it makes no difference for dielectrics, and only makes a difference on metals. The problem on #75302 mentions non-metals only.

I checked your changes (with 0.37 instead of 1.0) and as expected only changes the behavior for metals.

[n0tank3sh]

@armmah I started working on the non-metal part as soon as jcostello demanded before and after the fix picture.

[armmah]

Oh nice catch on the reflect vector. All of the experiments I've done I have been using either view_dir or -view_dir, didn't think to try with reflect over normal.

It ends up solving the problem in a proper physically accurate and consistent way.

P.S. on the tall cube the visual was similar, but your ray dir looks much nicer on some other scenarios I have set up in the material tests demo project.

[n0tank3sh]

@armmah Thank you

[fire]

Do you think it's possible to have only one commit?

[n0tank3sh]

@fire I think of having two commits one fixing the interpolation of R0 and another fixing the Shlick term. Is it reasonable to you? I squash all the commit @fire

[clayjohn]

There are two things going on in this PR. The change of R0 is one thing, it only changes the behaviour for metals and does not help #75302. Essentially it adds in some schlick reflectance for metallic metals.

The second change is using ray_dir instead of -view_dir which kind of addresses the issue in #75302 but it doesn't quite work. For non-metallics, reflections should be most visible at a glancing angle and then they should smoothly fade in strength as the view direction approaches the normal. With this change the reflection is strongest midway between a perpendicular view and parallel view. At a glancing angle reflections disappear.

Nearly parallel, reflection is fading away as it should

Midpoint, reflection is still strong

Glancing angle, reflections should be at their strongest, but they are faded out

[clayjohn]

Taking a look it seems like removing the reversal of the normal makes it work. I.e. the correct code is :

//normal.y = -normal.y;
float m = clamp(1.0 - dot(normal, -view_dir), 0.0, 1.0);

[n0tank3sh]

@clayjohn Thank that makes sense but why not use view_dir instead of -view_dir?

[clayjohn]

@clayjohn Thank that makes sense but why not use view_dir instead of -view_dir?

-view_dir is the "eye vector" i.e. the vector from the fragment to the eye. Which is what we need for this. What we are measuring is the difference in angle between the normal and the view direction. If the camera is looking straight down at a plane the eye vector will be the same as the normal (because both are normalized) and the dot product will return a 1. If the camera is perpendicular to the normal then the dot product will return 0. This is what we want. If we were to use the view direction then when looking straight down the view vector would be -1 and would also scale to 0 as the view becomes perpendicular to the normal

[n0tank3sh]

Yes, I understood that we are calculating the cos theta of the two vector vectors. view_dir is a little confusing as to know which side the view_dir pointing to. Thank you for clearing it out.

[clayjohn]

Looks good to me. In the long term we should probably just write out f0 from the main pass instead of metallic. Then we can take specular color into account (even if its just grayscale)

[YuriSizov]

Thanks!

[YuriSizov]

Cherry-picked for 4.0.3.