Add energy compensation

[vorg]

To be precise i mean multiscatter ggx preventing loosing energy from IBL for metallic materials

This is tested using Furnace Test with pure white environment map. As you can see rough material on the right is darker than glossy on the left.

[vorg]

Examples here https://doc.babylonjs.com/how_to/physically_based_rendering_master#energy-conservation

[vorg]

Original (?) source of discussion https://eheitzresearch.wordpress.com/240-2/

Excellent explanation of math https://patapom.com/blog/BRDF/MSBRDFEnergyCompensation/

ThreeJS PR https://github.com/mrdoob/three.js/pull/15644

Filament implementation https://google.github.io/filament/Filament.md.html#materialsystem/improvingthebrdfs/energylossinspecularreflectance

[vorg]

There is some work done here https://github.com/pex-gl/pex-renderer/issues/331

[dmnsgn]

A Multiple-Scattering Microfacet Model for Real-Time Image Based Lighting

// GLSL code for dielectrics
// Common code for single and multiple scattering
// Roughness dependent fresnel
vec3 Fr = max(vec3(1.0 - roughness), F0) - F0; vec3 kS = F0 + Fr * pow(1.0-ndv, 5.0);
vec2 f_ab = textureLod(uEnvBRDF, vec2(ndv, roughness), 0).xy;
vec3 FssEss = kS * f_ab.x + f_ab.y;
float lodLevel = roughness * numEnvLevels;
vec3 reflDir = reflect(-eye, normal);
// Prefiltered radiance
vec3 radiance = getRadiance(reflDir, lodLevel);
// Cosine-weighted irradiance
vec3 irradiance = getIrradiance(normal);
// Multiple scattering
float Ess = f_ab.x + f_ab.y;
float Ems = 1-Ess;
vec3 Favg = F0 + (1-F0)/21;
vec3 Fms = FssEss*Favg/(1-(1-Ess)*Favg);
// Dielectrics
vec3 Edss = 1 - (FssEss + Fms * Ems); vec3 kD = albedo * Edss;
// Composition
return FssEss * radiance + (Fms*Ems+kD) * irradiance;

[vorg]

Hasn't that been implemented already? Or is it only in some experiment?

[dmnsgn]

Hasn't that been implemented already? Or is it only in some experiment?

We are not doing multi-scattering in IBL no, as far as I am aware, and we're not computing the energyCompensation factor.

It has improved over the past years but still not energy conserving for metals (metallic: 1 and increasing roughness):