
Aluminum | 0.91, 0.92, 0.92 | #e8eaea |

Titanium | 0.76, 0.73, 0.69 | #c1baaf |

Iron | 0.77, 0.78, 0.78 | #c4c6c6 |

Platinum | 0.83, 0.81, 0.78 | #d3cec6 |

Gold | 1.00, 0.85, 0.57 | #ffd891 |

Brass | 0.98, 0.90, 0.59 | #f9e596 |

Copper | 0.97, 0.74, 0.62 | #f7bc9e |

[Table [fNormalMetals]: $\fNormal$ for common metals] All materials have a Fresnel reflectance of 100% at grazing angles so we will set $\fGrazing$ in the following way when evaluating the specular BRDF $\fSpecular$: $$$$\fGrazing = 1.0$$$$ Figure [grazing_reflectance] shows a red plastic ball. If you look closely at the edges of the sphere, you will be able to notice the achromatic specular reflectance at grazing angles. ![Figure [grazing_reflectance]: The specular reflectance becomes achromatic at grazing angles](images/material_grazing_reflectance.png) **Conductors** The specular reflectance of metallic surfaces is chromatic: $$$$\fNormal = baseColor \cdot metallic$$$$ Listing [fNormal] shows how $\fNormal$ is computed for both dielectric and metallic materials. It shows that the color of the specular reflectance is derived from the base color in the metallic case. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ vec3 f0 = 0.16 * reflectance * reflectance * (1.0 - metallic) + baseColor * metallic; ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ [Listing [fNormal]: Computing $\fNormal$ for dielectric and metallic materials in GLSL] #### Roughness remapping and clamping The roughness set by the user, called perceptualRoughness here, is remapped to a perceptually linear range using the following formulation: $$$$\alpha = perceptualRoughness^2$$$$ Figure [roughness_remap] shows a silver metallic surface with increasing roughness (from 0.0 to 1.0), using the unmodified roughness value (bottom) and the remapped value (top). ![Figure [roughness_remap]: Roughness remapping comparison: perceptually linear roughness (top) and roughness (bottom)](images/material_roughness_remap.png) Using this visual comparison, it is obvious that the remapped roughness is easier to understand by artists and developers. Without this remapping, shiny metallic surfaces would have to be confined to a very small range between 0.0 and 0.05. Brent Burley made similar observations in his presentation [#Burley12]. After experimenting with other remappings (cubic and quadratic mappings for instance), we have reached the conclusion that this simple square remapping delivers visually pleasing and intuitive results while being cheap for real-time applications. Last but not least, it is important to note that the roughness parameters is used in various computations at runtime where limited floating point precision can become an issue. For instance, _mediump_ precision floats are often implemented as half-floats (fp16) on mobile GPUs. This cause problems when computing small values like $\frac{1}{perceptualRoughness^4}$ in our lighting equations (roughness squared in the GGX computation). The smallest value that can be represented as a half-float is $2^{-14}$ or $6.1 \times 10^{-5}$. To avoid divisions by 0 on devices that do not support denormals, the result of $\frac{1}{roughness^4}$ must therefore not be lower than $6.1 \times 10^{-5}$. To do so, we must clamp the roughness to 0.089, which gives us $6.274 \times 10^{-5}$. Denormals should also be avoided to prevent performance drops. The roughness can also not be set to 0 to avoid obvious divisions by 0. Since we also want specular highlights to have a minimum size (a roughness close to 0 creates almost invisible highlights), we should clamp the roughness to a safe range in the shader. This clamping has the added benefit of correcting specular aliasing[^frostbiteRoughnessClamp] that can appear for low roughness values. [^frostbiteRoughnessClamp]: The Frostbite engine clamps the roughness of analytical lights to 0.045 to reduce specular aliasing. This is possible when using single precision floats (fp32). ### Blending and layering As noted in [#Burley12] and [#Neubelt13], this model allows for robust blending between different materials by simply interpolating the different parameters. In particular, this allows to layer different materials using simple masks. For instance, figure [materialBlending] shows how the studio Ready at Dawn used material blending and layering in _The Order: 1886_ to create complex appearances from a library of simple materials (gold, copper, wood, rust, etc.). ![Figure [materialBlending]: Material blending and layering. Source: Ready at Dawn Studios](images/material_blending.png) The blending and layering of materials is effectively an interpolation of the various parameters of the material model. Figure [material_interpolation] show an interpolation between shiny metallic chrome and rough red plastic. While the intermediate blended materials make little physical sense, they look plausible. ![Figure [material_interpolation]: Interpolation from shiny chrome (left) to rough red plastic (right)](images/material_interpolation.png) ### Crafting physically based materials Designing physically based materials is fairly easy once you understand the nature of the four main parameters: base color, metallic, roughness and reflectance. We provide a [useful chart/reference guide](./Material%20Properties.pdf) to help artists and developers craft their own physically based materials. ![Crafting physically based materials](images/material_chart.jpg) In addition, here is a quick summary of how to use our material model: All materials : **Base color** should be devoid of lighting information, except for micro-occlusion. **Metallic** is almost a binary value. Pure conductors have a metallic value of 1 and pure dielectrics have a metallic value of 0. You should try to use values close at or close to 0 and 1. Intermediate values are meant for transitions between surface types (metal to rust for instance). Non-metallic materials : **Base color** represents the reflected color and should be an sRGB value in the range 50-240 (strict range) or 30-240 (tolerant range). **Metallic** should be 0 or close to 0. **Reflectance** should be set to 127 sRGB (0.5 linear, 4% reflectance) if you cannot find a proper value. Do not use values under 90 sRGB (0.35 linear, 2% reflectance). Metallic materials : **Base color** represents both the specular color and reflectance. Use values with a luminosity of 67% to 100% (170-255 sRGB). Oxidized or dirty metals should use a lower luminosity than clean metals to take into account the non-metallic components. **Metallic** should be 1 or close to 1. **Reflectance** is ignored (calculated from the base color). ## Clear coat model The standard material model described previously is a good fit for isotropic surfaces made of a single layer. Multi-layer materials are unfortunately fairly common, particularly materials with a thin translucent layer over a standard layer. Real world examples of such materials include car paints, soda cans, lacquered wood, acrylic, etc. ![Figure [materialClearCoat]: Comparison of a blue metallic surface under the standard material model (left) and the clear coat model (right)](images/material_clear_coat.png) A clear coat layer can be simulated as an extension of the standard material model by adding a second specular lobe, which implies evaluating a second specular BRDF. To simplify the implementation and parameterization, the clear coat layer will always be isotropic and dielectric. The base layer can be anything allowed by the standard model (dielectric or conductor). Since incoming light will traverse the clear coat layer, we must also take the loss of energy into account as shown in figure [clearCoatModel]. Our model will however not simulate inter reflection and refraction behaviors. ![Figure [clearCoatModel]: Clear coat surface model](images/diagram_clear_coat.png) ### Clear coat specular BRDF The clear coat layer will be modeled using the same Cook-Torrance microfacet BRDF used in the standard model. Since the clear coat layer is always isotropic and dielectric, with low roughness values (see section [Clear coat parameterization]), we can choose cheaper DFG terms without notably sacrificing visual quality. A survey of the terms listed in [#Karis13] and [#Burley12] shows that the Fresnel and NDF terms we already use in the standard model are not computationally more expensive than other terms. [#Kelemen01] describes a much simpler term that can replace our Smith-GGX visibility term: $$$$V(l,h) = \frac{1}{4(\LoH)^2}$$$$ This masking-shadowing function is not physically based, as shown in [#Heitz14], but its simplicity makes it desirable for real-time rendering. In summary, our clear coat BRDF is a Cook-Torrance specular microfacet model, with a GGX normal distribution function, a Kelemen visibility function, and a Schlick Fresnel function. Listing [kelemen] shows how trivial the GLSL implementation is. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ float V_Kelemen(float LoH) { return 0.25 / (LoH * LoH); } ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ [Listing [kelemen]: Implementation of the Kelemen visibility term in GLSL] **Note on the Fresnel term** The Fresnel term of the specular BRDF requires $\fNormal$, the specular reflectance at normal incidence angle. This parameter can be computed from an index of refraction of an interface. We will assume that our clear coat layer is made of polyurethane, a common compound [used in coatings and varnishes](https://en.wikipedia.org/wiki/List_of_polyurethane_applications#Varnish), or similar. An air-polyurethane interface [has an IOR of 1.5](http://www.clearpur.com/transparent-polyurethanes/), from which we can deduce $\fNormal$: $$$$\fNormal(1.5) = \frac{(1.5 - 1)^2}{(1.5 + 1)^2} = 0.04$$$$ This corresponds to a Fresnel reflectance of 4% that we know is associated with common dielectric materials. ### Integration in the surface response Because we must take into account the loss of energy caused by the addition of the clear coat layer, we can reformulate the BRDF from equation $\ref{brdf}$ thusly: $$$$f(v,l)=\fDiffuse(n,l) (1 - F_c) + \fSpecular(n,l) (1 - F_c) + f_c(n,l)$$$$ Where $F_c$ is the Fresnel term of the clear coat BRDF and $f_c$ the clear coat BRDF ### Clear coat parameterization The clear coat material model encompasses all the parameters previously defined for the standard material mode, plus two parameters described in table [clearCoatParameters]. Parameter | Definition ----------------------:|:--------------------- **ClearCoat** | Strength of the clear coat layer. Scalar between 0 and 1 **ClearCoatRoughness** | Perceived smoothness or roughness of the clear coat layer. Scalar between 0 and 1 [Table [clearCoatParameters]: Clear coat model parameters] The clear coat roughness parameter is remapped and clamped in a similar way to the roughness parameter of the standard material. Figure [clearCoat] and figure [clearCoatRoughness] show how the clear coat parameters affect the appearance of a surface. ![Figure [clearCoat]: Clear coat varying from 0.0 (left) to 1.0 (right) with metallic set to 1.0 and roughness to 0.8](images/material_clear_coat1.png) ![Figure [clearCoatRoughness]: Clear coat roughness varying from 0.0 (left) to 1.0 (right) with metallic set to 1.0, roughness to 0.8 and clear coat to 1.0](images/material_clear_coat2.png) Listing [clearCoatBRDF] shows the GLSL implementation of the clear coat material model after remapping, parameterization and integration in the standard surface response. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ void BRDF(...) { // compute Fd and Fr from standard model // remapping and linearization of clear coat roughness clearCoatPerceptualRoughness = clamp(clearCoatPerceptualRoughness, 0.089, 1.0); clearCoatRoughness = clearCoatPerceptualRoughness * clearCoatPerceptualRoughness; // clear coat BRDF float Dc = D_GGX(clearCoatRoughness, NoH); float Vc = V_Kelemen(clearCoatRoughness, LoH); float Fc = F_Schlick(0.04, LoH) * clearCoat; // clear coat strength float Frc = (Dc * Vc) * Fc; // account for energy loss in the base layer return color * ((Fd + Fr * (1.0 - Fc)) * (1.0 - Fc) + Frc); } ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ [Listing [clearCoatBRDF]: Implementation of the clear coat BRDF in GLSL] ### Base layer modification The presence of a clear coat layer means that we should recompute $\fNormal$, since it is normally based on an air-material interface. The base layer thus requires $\fNormal$ to be computed based on a clear coat-material interface instead. This can be achieved by computing the material's index of refraction (IOR) from $\fNormal$, then computing a new $\fNormal$ based on the newly computed IOR and the IOR of the clear coat layer (1.5). First, we compute the base layer's IOR: $$IOR_{base} = \frac{1 + \sqrt{\fNormal}}{1 - \sqrt{\fNormal}}$$ Then we compute the new $\fNormal$ from this new index of refraction: $$f_{0_{base}} = \left( \frac{IOR_{base} - 1.5}{IOR_{base} + 1.5} \right) ^2$$ Since the clear coat layer's IOR is fixed, we can combine both steps to simplify: $$f_{0_{base}} = \frac{\left( 1 - 5 \sqrt{\fNormal} \right) ^2}{\left( 5 - \sqrt{\fNormal} \right) ^2}$$ We should also modify the base layer's apparent roughness based on the IOR of the clear coat layer but this is something we have opted to leave out for now. ## Anisotropic model The standard material model described previously can only describe isotropic surfaces, that is, surfaces whose properties are identical in all directions. Many real-world materials, such as brushed metal, can, however, only be replicated using an anisotropic model. ![Figure [anisotropic]: Comparison of isotropic material (left) and anisotropic material (right)](images/material_anisotropic.png) ### Anisotropic specular BRDF The isotropic specular BRDF described previously can be modified to handle anisotropic materials. Burley achieves this by using an anisotropic GGX NDF: $$$$D_{aniso}(h,\alpha) = \frac{1}{\pi \alpha_t \alpha_b} \frac{1}{((\frac{t \cdot h}{\alpha_t})^2 + (\frac{b \cdot h}{\alpha_b})^2 + (\NoH)^2)^2}$$$$ This NDF unfortunately relies on two supplemental roughness terms noted $\alpha_b$, the roughness along the bitangent direction, and $\alpha_t$, the roughness along the tangent direction. Neubelt and Pettineo [#Neubelt13] propose a way to derive $\alpha_b$ from $\alpha_t$ by using an _anisotropy_ parameter that describes the relationship between the two roughness values for a material: \begin{align*} \alpha_t &= \alpha \\ \alpha_b &= lerp(0, \alpha, 1 - anisotropy) \end{align*} The relationship defined in [#Burley12] is different, offers more pleasant and intuitive results, but is slightly more expensive: \begin{align*} \alpha_t &= \frac{\alpha}{\sqrt{1 - 0.9 \times anisotropy}} \\ \alpha_b &= \alpha \sqrt{1 - 0.9 \times anisotropy} \end{align*} We instead opted to follow the relationship described in [#Kulla17] as it allows creation of sharp highlights: \begin{align*} \alpha_t &= \alpha \times (1 + anisotropy) \\ \alpha_b &= \alpha \times (1 - anisotropy) \end{align*} Note that this NDF requires the tangent and bitangent directions in addition to the normal direction. Since these directions are already needed for normal mapping, providing them may not be an issue. The resulting implementation is described in listing [anisotropicBRDF]. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ float at = max(roughness * (1.0 + anisotropy), 0.001); float ab = max(roughness * (1.0 - anisotropy), 0.001); float D_GGX_Anisotropic(float NoH, const vec3 h, const vec3 t, const vec3 b, float at, float ab) { float ToH = dot(t, h); float BoH = dot(b, h); float a2 = at * ab; highp vec3 v = vec3(ab * ToH, at * BoH, a2 * NoH); highp float v2 = dot(v, v); float w2 = a2 / v2; return a2 * w2 * w2 * (1.0 / PI); } ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ [Listing [anisotropicBRDF]: Implementation of Burley's anisotropic NDF in GLSL] In addition, [#Heitz14] presents an anisotropic masking-shadowing function to match the height-correlated GGX distribution. The masking-shadowing term can be greatly simplified by using the visibility function instead: $$$$G(v,l,h,\alpha) = \frac{\chi^+(\VoH) \chi^+(\LoH)}{1 + \Lambda(v) + \Lambda(l)}$$$$ $$$$\Lambda(m) = \frac{-1 + \sqrt{1 + \alpha_0^2 tan^2(\theta_m)}}{2} = \frac{-1 + \sqrt{1 + \alpha_0^2 \frac{(1 - cos^2(\theta_m))}{cos^2(\theta_m)}}}{2}$$$$ Where: $$$$\alpha_0 = \sqrt{cos^2(\phi_0)\alpha_x^2 + sin^2(\phi_0)\alpha_y^2}$$$$ After derivation we obtain: $$$$V_{aniso}(\NoL,\NoV,\alpha) = \frac{1}{2((\NoL)\hat{\Lambda}_v+(\NoV)\hat{\Lambda}_l)} \\ \hat{\Lambda}_v = \sqrt{\alpha^2_t(t \cdot v)^2+\alpha^2_b(b \cdot v)^2+(\NoV)^2} \\ \hat{\Lambda}_l = \sqrt{\alpha^2_t(t \cdot l)^2+\alpha^2_b(b \cdot l)^2+(\NoL)^2}$$$$ The term $\hat{\Lambda}_v$ is the same for every light and can be computed only once if needed. The resulting implementation is described in listing [anisotropicV]. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ float at = max(roughness * (1.0 + anisotropy), 0.001); float ab = max(roughness * (1.0 - anisotropy), 0.001); float V_SmithGGXCorrelated_Anisotropic(float at, float ab, float ToV, float BoV, float ToL, float BoL, float NoV, float NoL) { float lambdaV = NoL * length(vec3(at * ToV, ab * BoV, NoV)); float lambdaL = NoV * length(vec3(at * ToL, ab * BoL, NoL)); float v = 0.5 / (lambdaV + lambdaL); return saturateMediump(v); } ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ [Listing [anisotropicV]: Implementation of the anisotropic visibility function in GLSL] ### Anisotropic parameterization The anisotropic material model encompasses all the parameters previously defined for the standard material mode, plus an extra parameter described in table [anisotropicParameters]. Parameter | Definition ----------------------:|:--------------------- **Anisotropy** | Amount of anisotropy. Scalar between -1 and 1 [Table [anisotropicParameters]: Anisotropic model parameters] No further remapping is required. Note that negative values will align the anisotropy with the bitangent direction instead of the tangent direction. Figure [anisotropyParameter] shows how the anisotropy parameter affect the appearance of a rough metallic surface. ![Figure [anisotropyParameter]: Anisotropy varying from 0.0 (left) to 1.0 (right)](images/materials/anisotropy.png) ## Subsurface model [TODO] ### Subsurface specular BRDF [TODO] ### Subsurface parameterization [TODO] ## Cloth model All the material models described previously are designed to simulate dense surfaces, both at a macro and at a micro level. Clothes and fabrics are however often made of loosely connected threads that absorb and scatter incident light. The microfacet BRDFs presented earlier do a poor job of recreating the nature of cloth due to their underlying assumption that a surface is made of random grooves that behave as perfect mirrors. When compared to hard surfaces, cloth is characterized by a softer specular lobe with a large falloff and the presence of fuzz lighting, caused by forward/backward scattering. Some fabrics also exhibit two-tone specular colors (velvets for instance). Figure [materialCloth] shows how a traditional microfacet BRDF fails to capture the appearance of a sample of denim fabric. The surface appears rigid (almost plastic-like), more similar to a tarp than a piece of clothing. This figure also shows how important the softer specular lobe caused by absorption and scattering is to the faithful recreation of the fabric. ![Figure [materialCloth]: Comparison of denim fabric rendered using a traditional microfacet BRDF (left) and our cloth BRDF (right)](images/screenshot_cloth.png) Velvet is an interesting use case for a cloth material model. As shown in figure [materialVelvet] this type of fabric exhibits strong rim lighting due to forward and backward scattering. These scattering events are caused by fibers standing straight at the surface of the fabric. When the incident light comes from the direction opposite to the view direction, the fibers will forward-scatter the light. Similarly, when the incident light from the same direction as the view direction, the fibers will scatter the light backward. ![Figure [materialVelvet]: Velvet fabric showcasing forward and backward scattering](images/screenshot_cloth_velvet.png) Since fibers are flexible, we should in theory model the ability to groom the surface. While our model does not replicate this characteristic, it does model a visible front facing specular contribution that can be attributed to the random variance in the direction of the fibers. It is important to note that there are types of fabrics that are still best modeled by hard surface material models. For instance, leather, silk and satin can be recreated using the standard or anisotropic material models. ### Cloth specular BRDF The cloth specular BRDF we use is a modified microfacet BRDF as described by Ashikhmin and Premoze in [#Ashikhmin07]. In their work, Ashikhmin and Premoze note that the distribution term is what contributes most to a BRDF and that the shadowing/masking term is not necessary for their velvet distribution. The distribution term itself is an inverted Gaussian distribution. This helps achieve fuzz lighting (forward and backward scattering) while an offset is added to simulate the front facing specular contribution. The so-called velvet NDF is defined as follows: $$$$D_{velvet}(v,h,\alpha) = c_{norm}(1 + 4 exp\left(\frac{-{cot}^2\theta_{h}}{\alpha^2}\right))$$$$ This NDF is a variant of the NDF the same authors describe in [#Ashikhmin00], notably modified to include an offset (set to 1 here) and an amplitude (4). In [#Neubelt13], Neubelt and Pettineo propose a normalized version of this NDF: $$$$D_{velvet}(v,h,\alpha) = \frac{1}{\pi(1 + 4\alpha^2)} (1 + 4 \frac{exp\left(\frac{-{cot}^2\theta_{h}}{\alpha^2}\right)}{{sin}^4\theta_{h}})$$$$ For the full specular BRDF, we also follow [#Neubelt13] and replace the traditional denominator with a smoother variant: $$$$\label{clothSpecularBRDF} f_{r}(v,h,\alpha) = \frac{D_{velvet}(v,h,\alpha)}{4(\NoL + \NoV - (\NoL)(\NoV))}$$$$ The implementation of the velvet NDF is presented in listing [clothBRDF], optimized to properly fit in half float formats and to avoid computing a costly cotangent, relying instead on trigonometric identities. Note that we removed the Fresnel component from this BRDF. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ float D_Ashikhmin(float roughness, float NoH) { // Ashikhmin 2007, "Distribution-based BRDFs" float a2 = roughness * roughness; float cos2h = NoH * NoH; float sin2h = max(1.0 - cos2h, 0.0078125); // 2^(-14/2), so sin2h^2 > 0 in fp16 float sin4h = sin2h * sin2h; float cot2 = -cos2h / (a2 * sin2h); return 1.0 / (PI * (4.0 * a2 + 1.0) * sin4h) * (4.0 * exp(cot2) + sin4h); } ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ [Listing [clothBRDF]: Implementation of Ashikhmin's velvet NDF in GLSL] In [#Estevez17] Estevez and Kulla propose a different NDF (called the "Charlie" sheen) that is based on an exponentiated sinusoidal instead of an inverted Gaussian. This NDF is appealing for several reasons: its parameterization feels more natural and intuitive, it provides a softer appearance and, as shown in equation $\ref{charlieNDF}$, its implementation is simpler: $$$$\label{charlieNDF} D(m) = \frac{(2 + \frac{1}{\alpha}) sin(\theta)^{\frac{1}{\alpha}}}{2 \pi}$$$$ [#Estevez17] also presents a new shadowing term that we omit here because of its cost. We instead rely on the visibility term from [#Neubelt13] (shown in equation $\ref{clothSpecularBRDF}$ above). The implementation of this NDF is presented in listing [clothCharlieBRDF], optimized to properly fit in half float formats. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ float D_Charlie(float roughness, float NoH) { // Estevez and Kulla 2017, "Production Friendly Microfacet Sheen BRDF" float invAlpha = 1.0 / roughness; float cos2h = NoH * NoH; float sin2h = max(1.0 - cos2h, 0.0078125); // 2^(-14/2), so sin2h^2 > 0 in fp16 return (2.0 + invAlpha) * pow(sin2h, invAlpha * 0.5) / (2.0 * PI); } ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ [Listing [clothCharlieBRDF]: Implementation of the "Charlie" NDF in GLSL] #### Sheen color To offer better control over the appearance of cloth and to give users the ability to recreate two-tone specular materials, we introduce the ability to directly modify the specular reflectance. Figure [materialClothSheen] shows an example of using the parameter we call "sheen color". ![Figure [materialClothSheen]: Blue fabric without (left) and with (right) sheen](images/screenshot_cloth_sheen.png) ### Cloth diffuse BRDF Our cloth material model still relies on a Lambertian diffuse BRDF. It is however slightly modified to be energy conservative (akin to the energy conservation of our clear coat material model) and offers an optional subsurface scattering term. This extra term is not physically based and can be used to simulate the scattering, partial absorption and re-emission of light in certain types of fabrics. First, here is the diffuse term without the optional subsurface scattering: $$$$f_{d}(v,h) = \frac{c_{diff}}{\pi}(1 - F(v,h))$$$$ Where $F(v,h)$ is the Fresnel term of the cloth specular BRDF in equation $\ref{clothSpecularBRDF}$. In practice we've opted to leave out the $1 - F(v, h)$ term in the diffuse component. The effect is a bit subtle and we deemed it wasn't worth the added cost. Subsurface scattering is implemented using the wrapped diffuse lighting technique, in its energy conservative form: $$$$f_{d}(v,h) = \frac{c_{diff}}{\pi}(1 - F(v,h)) \left< \NoL + \frac{w}{(1 + w)^2} \right> \left< c_{subsurface} + \NoL \right>$$$$ Where $w$ is a value between 0 and 1 defining by how much the diffuse light should wrap around the terminator. To avoid introducing another parameter, we fix $w = 0.5$. Note that with wrap diffuse lighting, the diffuse term must not be multiplied by $\NoL$. The effect of this cheap subsurface scattering approximation can be seen in figure [materialClothSubsurface]. ![Figure [materialClothSubsurface]: White cloth (left column) vs white cloth with brown subsurface scattering (right)](images/screenshot_cloth_subsurface.png) The complete implementation of our cloth BRDF, including sheen color and optional subsurface scattering, can be found in listing [clothFullBRDF]. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ // specular BRDF float D = distributionCloth(roughness, NoH); float V = visibilityCloth(NoV, NoL); vec3 F = sheenColor; vec3 Fr = (D * V) * F; // diffuse BRDF float diffuse = diffuse(roughness, NoV, NoL, LoH); #if defined(MATERIAL_HAS_SUBSURFACE_COLOR) // energy conservative wrap diffuse diffuse *= saturate((dot(n, light.l) + 0.5) / 2.25); #endif vec3 Fd = diffuse * pixel.diffuseColor; #if defined(MATERIAL_HAS_SUBSURFACE_COLOR) // cheap subsurface scatter Fd *= saturate(subsurfaceColor + NoL); vec3 color = Fd + Fr * NoL; color *= (lightIntensity * lightAttenuation) * lightColor; #else vec3 color = Fd + Fr; color *= (lightIntensity * lightAttenuation * NoL) * lightColor; #endif ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ [Listing [clothFullBRDF]: Implementation of our cloth BRDF in GLSL] ### Cloth parameterization The cloth material model encompasses all the parameters previously defined for the standard material mode except for _metallic_ and _reflectance_. Two extra parameters described in table [clothParameters] are also available. Parameter | Definition ---------------------:|:--------------------- **SheenColor** | Specular tint to create two-tone specular fabrics (defaults to 0.04 to match the standard reflectance) **SubsurfaceColor** | Tint for the diffuse color after scattering and absorption through the material [Table [clothParameters]: Cloth model parameters] To create a velvet-like material, the base color can be set to black (or a dark color). Chromaticity information should instead be set on the sheen color. To create more common fabrics such as denim, cotton, etc. use the base color for chromaticity and use the default sheen color or set the sheen color to the luminance of the base color. # Lighting The correctness and coherence of the lighting environment is paramount to achieving plausible visuals. After surveying existing rendering engines (such as Unity or Unreal Engine 4) as well as the traditional real-time rendering literature, it is obvious that coherency is rarely achieved. The Unreal Engine, for instance, lets artists specify the "brightness" of a point light in lumens, a unit of luminous power. The brightness of directional lights is however expressed using an arbitrary unnamed unit. To match the brightness of a point light with a luminous power of 5,000 lumens, the artist must use a directional light of brightness 10. This kind of mismatch makes it difficult for artists to maintain the visual integrity of a scene when adding, removing or modifying lights. Using solely arbitrary units is a coherent solution but it makes reusing lighting rigs a difficult task. For instance, an outdoor scene will use a directional light of brightness 10 as the sun and all other lights will be defined relative to that value. Moving these lights to an indoor environment would make them too bright. Our goal is therefore to make all lighting correct by default, while giving artists enough freedom to achieve the desired look. We will support a number of lights, split in two categories, direct and indirect lighting: **Direct lighting**: punctual lights, photometric lights, area lights. **Indirect lighting**: image based lights (IBLs), for both local[^localProbesMobile] and distant light probes. [^localProbesMobile]: Local light probes might be too expensive to support on mobile, we will first focus our efforts on distant light probes set at infinity ## Units The following sections will discuss how to implement various types of lights and the proposed equations make use of different symbols and units summarized in table [lightUnits]. Photometric term | Notation | Unit -----------------------:|:------------------:|:----------------- Luminous power | $\Phi$ | Lumen ($lm$) Luminous intensity | $I$ | Candela ($cd$) or $\frac{lm}{sr}$ Illuminance | $E$ | Lux ($lx$) or $\frac{lm}{m^2}$ Luminance | $L$ | Nit ($nt$) or $\frac{cd}{m^2}$ Radiant power | $\Phi_e$ | Watt ($W$) Luminous efficacy | $\eta$ | Lumens per watt ($\frac{lm}{W}$) Luminous efficiency | $V$ | Percentage (%) [Table [lightUnits]: Photometric units] To get properly coherent lighting, we must use light units that respect the ratio between various light intensities found in real-world scenes. These intensities can vary greatly, from around 800 $lm$ for a household light bulb to 120,000 $lx$ for a daylight sky and sun illumination. The easiest way to achieve lighting coherency is to adopt physical light units. This will in turn enable full reusability of lighting rigs. Using physical light units also allows us to use a physically based camera. Table [lightTypesUnits] shows the light unit associated with each type of light we intend to support. Light type | Unit ------------------------:|:--------------------- Directional light | Illuminance ($lx$ or $\frac{lm}{m^2}$) Point light | Luminous power ($lm$) Spot light | Luminous power ($lm$) Photometric light | Luminous intensity ($cd$) Masked photometric light | Luminous power ($lm$) Area light | Luminous power ($lm$) Image based light | Luminance ($\frac{cd}{m^2}$) [Table [lightTypesUnits]: Intensity unity for each light type] **Notes about the radiant power unit** Even though commercially available light bulbs often display their brightness in lumens on the packaging, it is common to refer to the brightness of a light bulb by using its required energy in watts. The number of watts only indicates how much energy a bulb uses, not how bright it is. It is even more important to understand this difference now that more energy efficient bulbs are readily available (halogens, LEDs, etc.). However, since artists might be accustomed to gauging a light's brightness by its power, we should allow users to use the power unit to define the brightness of a light. The conversion is presented in equation $\ref{radiantPowerToLuminousPower}$. $$$$\label{radiantPowerToLuminousPower} \Phi = \Phi_e \eta$$$$ In equation $\ref{radiantPowerToLuminousPower}$, $\eta$ is the luminous efficacy of the light, expressed in lumens per watt. Knowing that the [maximum possible luminous efficacy](http://en.wikipedia.org/wiki/Luminous_efficacy) is 683 $\frac{lm}{W}$ we can also use luminous efficiency $V$ (also called luminous coefficient), as shown in equation $\ref{radiantPowerLuminousEfficiency}$. $$$$\label{radiantPowerLuminousEfficiency} \Phi = \Phi_e 683 \times V$$$$ Table [lightTypesEfficacy] can be used as a reference to convert watts to lumens using either the luminous efficacy or the luminous efficiency of various types of lights. More specific values are available on Wikipedia's [luminous efficacy](http://en.wikipedia.org/wiki/Luminous_efficacy) page. Light type | Efficacy $\eta$ | Efficiency $V$ -----------------------:|:------------------:|:----------------- Incandescent | 14-35 | 2-5% LED | 28-100 | 4-15% Fluorescent | 60-100 | 9-15% [Table [lightTypesEfficacy]: Efficacy and efficiency of various light types] ### Light units validation One of the big advantages of using physical light units is the ability to physically validate our equations. We can use specialized devices to measure three light units. #### Illuminance The illuminance reaching a surface can be measured using an incident light meter. For our tests, we use a [Sekonic L-478D](http://www.sekonic.com/products/l-478d/overview.aspx), shown in figure [sekonic]. The incident light meter uses a white diffuse dome to capture the illuminance reaching a surface. It is important to orient the dome properly depending on the desired measurement. For instance, orienting the dome perpendicular to the sun on a bright clear day will give very different results than orienting the dome horizontally. ![Figure [sekonic]: Sekonic L-478D incident light meter](images/photo_light_meter.jpg) #### Luminance The luminance at a surface, or the product of the incident light and the surface, can be measured using a luminance meter, also often called a spot meter. While incident light meters use a diffuse hemisphere to capture light from all directions, a spot meter uses a shield to measure incident light from a single direction. For our tests, we use a [Sekonic 5 degree Viewfinder](http://www.sekonic.com/products/l-478dr/accessories/np-finder-5-degree-for-l-478.aspx) that can replace the diffuser on the L-478D to measure luminance in a 5 degree cone. ![Sekonic L-478D working as a luminance meter using a special viewfinder](images/photo_incident_light_meter.jpg) #### Luminous intensity The luminous intensity of a light source cannot be measured directly but can be derived from the measured illuminance if we know the distance between the measuring device and the light source. Equation $\ref{derivedLuminousIntensity}$ is a simple application of the inverse square law discussed in section [Punctual lights]. $$$$\label{derivedLuminousIntensity} I = E \cdot d^2$$$$ ## Direct lighting We have defined the light units for all the light types supported by the renderer in the section above but we have not defined the light unit for the result of the lighting equations. Choosing physical light units means that we will compute luminance values in our shaders, and therefore that all our light evaluation functions will compute the luminance $L_{out}$ (or outgoing radiance) at any given point. The luminance depends on the illuminance $E$ and the BSDF $f(v,l)$ : $$$$\label{luminanceEquation} L_{out} = f(v,l)E$$$$ ### Directional lights The main purpose of directional lights is to recreate important light sources for outdoor environment, i.e. the sun and/or the moon. While directional lights do not truly exist in the physical world, any light source sufficiently far from the light receptor can be assumed to be directional (i.e. all the incident light rays are parallel, as shown in figure [directionalLight]). ![Figure [directionalLight]: Interaction between a directional light and a surface. The light source is a virtual construct that can only be represented by a direction](images/diagram_directional_light.png) This approximation proves to work incredibly well for the diffuse response of a surface but the specular response is incorrect. The Frostbite engine solves this problem by treating the "sun" directional light as a disc area light. However, our tests have shown that the quality increase does not justify the added computational costs. We earlier stated that we chose an illuminance light unit ($lx$) for directional lights. This is in part due to the fact that we can easily find illuminance values for the sky and the sun (online or with a light meter) but also to simplify the luminance equation described in $\ref{luminanceEquation}$. $$$$\label{directionalLuminanceEquation} L_{out} = f(v,l) E_{\bot} \left< \NoL \right>$$$$ In the simplified luminance equation $\ref{directionalLuminanceEquation}$, $E_{\bot}$ is the illuminance of the light source for a surface perpendicular to said light source. If the directional light source simulates the sun, $E_{\bot}$ is the illuminance of the sun for a surface perpendicular to the sun direction. Table [sunSkyIlluminance] provides useful reference values for the sun and sky illumination, measured[^illuminanceMeasures] on a clear day in March, in California. Light | 10am | 12pm | 5:30pm --------------------------:|---------:|---------:|---------: $Sky_{\bot} + Sun_{\bot}$ | 120,000 | 130,000 | 90,000 $Sky_{\bot}$ | 20,000 | 25,000 | 9,000 $Sun_{\bot}$ | 100,000 | 105,000 | 81,000 [Table [sunSkyIlluminance]: Illuminance values in $lx$ (a full moon has an illuminance of 1 $lx$)] Dynamic directional lights are particularly cheap to evaluate at runtime, as shown in listing [glslDirectionalLight]. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ vec3 l = normalize(-lightDirection); float NoL = clamp(dot(n, l), 0.0, 1.0); // lightIntensity is the illuminance // at perpendicular incidence in lux float illuminance = lightIntensity * NoL; vec3 luminance = BSDF(v, l) * illuminance; ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ [Listing [glslDirectionalLight]: Implementation of directional lights in GLSL] Figure [directionalLightTest] shows the effect of lighting a simple scene with a directional light setup to approximate a midday Sun (illuminance set to 110,000 $lx$). For illustration purposes, only direct lighting is shown. ![Figure [directionalLightTest]: Series of dielectric materials of varying roughness under a directional light](images/screenshot_directional_light.png) [^illuminanceMeasures]: Measurements taken with an incident light meter (Sekonic L-478D) ### Punctual lights Our engine will support two types of punctual lights, commonly found in most if not all rendering engines: point lights and spot lights. These types of lights are traditionally physically inaccurate for two reasons: 1. They are truly punctual and infinitesimally small. 2. They do not follow the [inverse square law](http://en.wikipedia.org/wiki/Inverse-square_law). The first issue can be addressed with area lights but, given the cheaper nature of punctual lights it is deemed practical to use infinitesimally small punctual lights whenever possible. The second issue is easy to fix. For a given punctual light, the perceived intensity decreases proportionally to the square of the distance from the viewer (more precisely, the light receptor). For punctual lights following the inverse square law, the term $E$ of equation $\ref{luminanceEquation}$ is expressed in equation $\ref{punctualLightEquation}$, where $d$ is the distance from a point at the surface to the light. $$$$\label{punctualLightEquation} E = L_{in} \left< \NoL \right> = \frac{I}{d^2} \left< \NoL \right>$$$$ The difference between point and spot lights lies in how $E$ is computed, and in particular how the luminous intensity $I$ is computed from the luminous power $\Phi$. #### Point lights A point light is defined only by a position in space, as shown in figure [pointLight]. ![Figure [pointLight]: Interaction between a point light and a surface. The attenuation only depends on the distance to the light](images/diagram_point_light.png) The luminous power of a point light is calculated by integrating the luminous intensity over the light's solid angle, as show in equation $\ref{pointLightLuminousPower}$. The luminous intensity can then be easily derived from the luminous power. $$$$\label{pointLightLuminousPower} \Phi = \int_{\Omega} I dl = \int_{0}^{2\pi} \int_{0}^{\pi} I d\theta d\phi = 4 \pi I \\ I = \frac{\Phi}{4 \pi}$$$$ By simple substitution of $I$ in $\ref{punctualLightEquation}$ and $E$ in $\ref{luminanceEquation}$ we can formulate the luminance equation of a point light as a function of the luminous power (see $\ref{pointLightLuminanceEquation}$). $$$$\label{pointLightLuminanceEquation} L_{out} = f(v,l) \frac{\Phi}{4 \pi d^2} \left< \NoL \right>$$$$ Figure [pointLightTest] shows the effect of lighting a simple scene with a point light subject to distance attenuation. Light falloff is exaggerated for illustration purposes. ![Figure [pointLightTest]: Inverse square law applied to point lights evaluation](images/screenshot_point_light.png) #### Spot lights A spot light is defined by a position in space, a direction vector and two cone angles, $\theta_{inner}$ and $\theta_{outer}$ (see figure [spotLight]). These two angles are used to define the angular falloff attenuation of the spot light. The light evaluation function of a spot light must therefore take into account both the inverse square law and these two angles to properly evaluate the luminance attenuation. ![Figure [spotLight]: Interaction between a spot light and a surface. The attenuation depends on the distance to the light and the angle between the surface the spot light's direction vector](images/diagram_spot_light.png) Equation $\ref{spotLightLuminousPower}$ describes how the luminous power of a spot light can be calculated in a similar fashion to point lights, using $\theta_{outer}$ the outer angle of the spot light's cone in the range [0..$\pi$]. $$$$\label{spotLightLuminousPower} \Phi = \int_{\Omega} I dl = \int_{0}^{2\pi} \int_{0}^{\theta_{outer}} I d\theta d\phi = 2 \pi (1 - cos\frac{\theta_{outer}}{2})I \\ I = \frac{\Phi}{2 \pi (1 - cos\frac{\theta_{outer}}{2})}$$$$ While this formulation is physically correct, it makes spot lights a little difficult to use: changing the outer angle of the cone changes the illumination levels. Figure [spotLightTestFocused] shows the same scene lit by a spot light, with an outer angle of 55 degrees and an outer angle of 15 degrees. Observes how the illumination level increases as the cone aperture decreases. ![Figure [spotLightTestFocused]: Comparison of spot light outer angles, 55 degrees (left) and 15 degrees (right)](images/screenshot_spot_light_focused.png) The coupling of illumination and the outer cone means that an artist cannot tweak the influence cone of a spot light without also changing the perceived illumination. It therefore makes sense to provide artists with a parameter to disable this coupling. Equations $\ref{spotLightLuminousPowerB}$ shows how to formulate the luminous power for that purpose. $$$$\label{spotLightLuminousPowerB} \Phi = \pi I \\ I = \frac{\Phi}{\pi} \\$$$$ With this new formulation to compute the luminous intensity, the test scene in figure [spotLightTest] exhibits similar illumination levels with both cone apertures. ![Figure [spotLightTest]: Comparison of spot light outer angles, 55 degrees (left) and 15 degrees (right)](images/screenshot_spot_light.png) This new formulation can also be considered physically based if the spot's reflector is replaced with a matte, diffuse mask that absorbs light perfectly. The spot light evaluation function can be expressed in two ways: - **With a light absorber** $$$$\label{spotAbsorber} L_{out} = f(v,l) \frac{\Phi}{\pi d^2} \left< \NoL \right> \lambda(l)$$$$ - **With a light reflector** $$$$\label{spotReflector} L_{out} = f(v,l) \frac{\Phi}{2 \pi (1 - cos\frac{\theta_{outer}}{2}) d^2} \left< \NoL \right> \lambda(l)$$$$ The term $\lambda(l)$ in equations $\ref{spotAbsorber}$ and $\ref{spotReflector}$ is the spot's angle attenuation factor described in equation $\ref{spotAngleAtt}$ below. $$$$\label{spotAngleAtt} \lambda(l) = \frac{l \times spotDirection - cos\theta_{outer}}{cos\theta_{inner} - cos\theta_{outer}}$$$$ #### Attenuation function A proper evaluation of the inverse square law attenuation factor is mandatory for physically based punctual lights. The simple mathematical formulation is unfortunately impractical for implementation purposes: 1. The division by the squared distance can lead to divides by 0 when objects intersect or "touch" light sources. 2. The influence sphere of each light is infinite ($\frac{I}{d^2}$ is asymptotic, it never reaches 0) which means that to correctly shade a pixel we need to evaluate every light in the world. The first issue can be solved easily by setting the assumption that punctual lights are not truly punctual but instead small area lights. To do this we can simply treat punctual lights as spheres of 1 cm radius, as show in equation $\ref{finitePunctualLight}$. $$$$\label{finitePunctualLight} E = \frac{I}{max(d^2, {0.01}^2)}$$$$ We can solve the second issue by introducing an influence radius for each light. There are several advantages to this solution. Tools can quickly show artists what parts of the world will be influenced by every light (the tool just needs to draw a sphere centered on each light). The rendering engine can cull lights more aggressively using this extra piece of information and artists/developers can assist the engine by manually tweaking the influence radius of a light. Mathematically, the illuminance of a light should smoothly reach zero at the limit defined by the influence radius. [#Karis13] proposes to window the inverse square function in such a way that the majority of the light's influence remains unaffected. The proposed windowing is described in equation $\ref{attenuationWindowing}$, where $r$ is the light's radius of influence. $$$$\label{attenuationWindowing} E = \frac{I}{max(d^2, {0.01}^2)} \left< 1 - \frac{d^4}{r^2} \right>$$$$ Listing [glslPunctualLight] demonstrates how to implement physically based punctual lights in GLSL. Note that the light intensity used in this piece of code is the luminous intensity $I$ in $cd$, converted from the luminous power CPU-side. This snippet is not optimized and some of the computations can be offloaded to the CPU (for instance the square of the light's inverse falloff radius, or the spot scale and angle). ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ float getSquareFalloffAttenuation(vec3 posToLight, float lightInvRadius) { float distanceSquare = dot(posToLight, posToLight); float factor = distanceSquare * lightInvRadius * lightInvRadius; float smoothFactor = max(1.0 - factor * factor, 0.0); return (smoothFactor * smoothFactor) / max(distanceSquare, 1e-4); } float getSpotAngleAttenuation(vec3 l, vec3 lightDir, float innerAngle, float outerAngle) { // the scale and offset computations can be done CPU-side float cosOuter = cos(outerAngle); float spotScale = 1.0 / max(cos(innerAngle) - cosOuter, 1e-4) float spotOffset = -cosOuter * spotScale float cd = dot(normalize(-lightDir), l); float attenuation = clamp(cd * spotScale + spotOffset, 0.0, 1.0); return attenuation * attenuation; } vec3 evaluatePunctualLight() { vec3 l = normalize(posToLight); float NoL = clamp(dot(n, l), 0.0, 1.0); vec3 posToLight = lightPosition - worldPosition; float attenuation; attenuation = getSquareFalloffAttenuation(posToLight, lightInvRadius); attenuation *= getSpotAngleAttenuation(l, lightDir, innerAngle, outerAngle); vec3 luminance = (BSDF(v, l) * lightIntensity * attenuation * NoL) * lightColor; return luminance; } ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ [Listing [glslPunctualLight]: Implementation of punctual lights in GLSL] ### Photometric lights Punctual lights are an extremely practical and efficient way to light a scene but do not give artists enough control over the light distribution. The field of architectural lighting design concerns itself with designing lighting systems to serve humans needs by taking into account: - The amount of light provided - The color of the light - The distribution of light within the space The lighting system we have described so far can easily address the first two points but we need a way to define the distribution of light within the space. Light distribution is especially important for indoor scenes or for some types of outdoor scenes or even road lighting. Figure [lightDistributionTest] shows scenes where the light distribution is controlled by the artist. This type of distribution control is widely used when putting objects on display (museums, stores or galleries for instance). ![Figure [lightDistributionTest]: Controlling the distribution of a point light](images/screenshot_photometric_lights.png) Photometric lights use a photometric profile to describe their intensity distribution. There are two commonly used formats, IES (Illuminating Engineering Society) and EULUMDAT (European Lumen Data format) but we will focus on the former. IES profiles are supported by many tools and engines, such as Unreal Engine 4, Frostbite, Renderman, Maya and Killzone. In addition, IES light profiles are commonly made available by bulbs and luminaires manufacturers (Philips offers [an extensive array of IES files](http://www.usa.lighting.philips.com/connect/tools_literature/photometric_data_1.wpd) for download for instance). Photometric profiles are particularly useful when they measure a luminaire or light fixture, in which the light source is partially covered. The luminaire will block the light emitted in certain directions, thus shaping the light distribution. ![Example of a real world luminaires that can be described by photometric profiles](images/photo_photometric_lights.jpg) An IES profile stores luminous intensity for various angles on a sphere around the measured light source. This spherical coordinate system is usually referred to as the photometric web, which can be visualized using specialized tools such as [IESviewer](http://www.photometricviewer.com/). Figure [xarrow] below shows the photometric web of the XArrow IES profile [provided by Pixar](http://renderman.pixar.com/view/DP25764) for use with Renderman. This picture also shows a rendering in 3D space of the XArrow IES profile by our tool lightgen. ![Figure [xarrow]: The XArrow IES profile rendered as a photometric web and as a point light in 3D space](images/screenshot_xarrow.png) The IES format is poorly documented and it is not uncommon to find syntax variations between files found on the Internet. The best resource to understand IES profile is Ian Ashdown's "Parsing the IESNA LM-63 photometric data file" document [#Ashdown98]. Succinctly, an IES profiles stores luminous intensities in candela at various angles around the light source. For each measured horizontal angle, a series of luminous intensities at different vertical angles is provided. It is however fairly common for measured light sources to be horizontally symmetrical. The XArrow profile shown above is a good example: intensities vary with vertical angles (vertical axis) but are symmetrical on the horizontal axis. The range of vertical angles in an IES profile is 0 to 180 degrees and the range of horizontal angles is 0 to 360 degrees. Figure [lightenSamples] shows the series of IES profiles provided by Pixar for Renderman, rendered using our lightgen tool. ![Figure [lightenSamples]: Series of IES light profiles rendered with lightgen](images/screenshot_lightgen_samples.png) IES profiles can be applied directly to any punctual light, point or spot. To do so, we must first process the IES profile and generate a photometric profile as a texture. For performance considerations, the photometric profile we generate is a 1D texture that represents the average luminous intensity for all horizontal angles at a specific vertical angle (i.e., each pixel represents a vertical angle). To truly represent a photometric light, we should use a 2D texture but since most lights are fully, or mostly, symmetrical on the horizontal plane, we can accept this approximation. The values stored in the texture are normalized by the inverse maximum intensity defined in the IES profile. This allows us to easily store the texture in any float format or, at the cost of a bit of precision, in a luminance 8-bit texture (grayscale PNG for instance). Storing normalized values also allows us to treat photometric profiles as a mask: Photometric profile as a mask : The luminous intensity is defined by the artist by setting the luminous power of the light, as with any other punctual light. The artist defined intensity is divided by the intensity of the light computed from the IES profile. IES profiles contain a luminous intensity but it is only valid for a bare light bulb whereas the measured intensity values take into account the light fixture. To measure the intensity of the luminaire, instead of the bulb, we perform a Monte-Carlo integration of the unit sphere using the intensities from the profile[^xarrowIntensity]. Photometric profile : The luminous intensity comes from the profile itself. All the values sampled from the 1D texture are simply multiplied by the maximum intensity. We also provide a multiplier for convenience. The photometric profile can be applied at rendering time as a simple attenuation. The luminance equation $\ref{photometricLightEvaluation}$ describes the photometric point light evaluation function. $$$$\label{photometricLightEvaluation} L_{out} = f(v,l) \frac{I}{d^2} \left< \NoL \right> \Psi(l)$$$$ The term $\Psi(l)$ is the photometric attenuation function. It depends on the light vector, but also on the direction of the light. Spot lights already possess a direction vector but we need to introduce one for photometric point lights as well. The photometric attenuation function can be easily implemented in GLSL by adding a new attenuation factor to the implementation of punctual lights (listing [glslPunctualLight]). The modified implementation is show in listing [glslPhotometricPunctualLight]. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ float getPhotometricAttenuation(vec3 posToLight, vec3 lightDir) { float cosTheta = dot(-posToLight, lightDir); float angle = acos(cosTheta) * (1.0 / PI); return texture2DLodEXT(lightProfileMap, vec2(angle, 0.0), 0.0).r; } vec3 evaluatePunctualLight() { vec3 l = normalize(posToLight); float NoL = clamp(dot(n, l), 0.0, 1.0); vec3 posToLight = lightPosition - worldPosition; float attenuation; attenuation = getSquareFalloffAttenuation(posToLight, lightInvRadius); attenuation *= getSpotAngleAttenuation(l, lightDirection, innerAngle, outerAngle); attenuation *= getPhotometricAttenuation(l, lightDirection); float luminance = (BSDF(v, l) * lightIntensity * attenuation * NoL) * lightColor; return luminance; } ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ [Listing [glslPhotometricPunctualLight]: Implementation of attenuation from photometric profiles in GLSL] The light intensity is computed CPU-side (listing [photometricLightIntensity]) and depends on whether the photometric profile is used as a mask. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ float multiplier; // Photometric profile used as a mask if (photometricLight.isMasked()) { // The desired intensity is set by the artist // The integrated intensity comes from a Monte-Carlo // integration over the unit sphere around the luminaire multiplier = photometricLight.getDesiredIntensity() / photometricLight.getIntegratedIntensity(); } else { // Multiplier provided for convenience, set to 1.0 by default multiplier = photometricLight.getMultiplier(); } // The max intensity in cd comes from the IES profile float lightIntensity = photometricLight.getMaxIntensity() * multiplier; ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ [Listing [photometricLightIntensity]: Computing the intensity of a photometric light on the CPU] [^xarrowIntensity]: The XArrow profile declares a luminous intensity of 1,750 lm but a Monte-Carlo integration shows an intensity of only 350 lm. ### Area lights [TODO] ### Lights parameterization Similarly to the parameterization of the standard material model, our goal is to make lights parameterization intuitive and easy to use for artists and developers alike. In that spirit, we decided to separate the light color (or hue) from the light intensity. A light color will therefore be defined as a linear RGB color (or sRGB in the tools UI for convenience). The full list of light parameters is presented in table [lightParameters]. Parameter | Definition --------------------------:|:--------------------- **Type** | Directional, point, spot or area **Direction** | Used for directional lights, spot lights, photometric point lights, and linear and tubular area lights (orientation) **Color** | The color of emitted light, as a linear RGB color. Can be specified as an sRGB color or a color temperature in the tools **Intensity** | The light's brightness. The unit depends on the type of light **Falloff radius** | Maximum distance of influence **Inner angle** | Angle of the inner cone for spot lights, in degrees **Outer angle** | Angle of the outer cone for spot lights, in degrees **Length** | Length of the area light, used to create linear or tubular lights **Radius** | Radius of the area light, used to create spherical or tubular lights **Photometric profile** | Texture representing a photometric light profile, works only for punctual lights **Masked profile** | Boolean indicating whether the IES profile is used as a mask or not. When used as a mask, the light's brightness will be multiplied by the ratio between the user specified intensity and the integrated IES profile intensity. When not used as a mask, the user specified intensity is ignored but the IES multiplier is used instead **Photometric multiplier** | Brightness multiplier for photometric lights (if IES as mask is turned off) [Table [lightParameters]: Light types parameters] **Note**: to simplify the implementation, all luminous powers will converted to luminous intensities ($cd$) before being sent to the shader. The conversion is light dependent and is explained in the previous sections. **Note**: the light type can be inferred from other parameters (e.g. a point light has a length, radius, inner angle and outer angle of 0). #### Color temperature However, real-world artificial lights are often defined by their color temperature, measured in Kelvin (K). The color temperature of a light source is the temperature of an ideal black-body radiator that radiates light of comparable hue to that of the light source. For convenience, the tools should allow the artist to specify the hue of a light source as a color temperature (a meaningful range is 1,000 K to 12,500 K). To compute RGB values from a temperature, we can use the Planckian locus, shown in figure [planckianLocus]. This locus is the path that the color of an incandescent black body takes in a chromaticity space as the body's temperature changes. ![Figure [planckianLocus]: The Planckian locus visualized on a CIE 1931 chromaticity diagram (source: Wikipedia)](images/diagram_planckian_locus.png) The easiest way to compute RGB values from this locus is to use the formula described in [#Krystek85]. Krystek's algorithm (equation $\ref{krystek}$) works in the CIE 1960 (UCS) space, using the following formula where $T$ is the desired temperature, and $u$ and $v$ the coordinates in UCS. $$$$\label{krystek} u(T) = \frac{0.860117757 + 1.54118254 \times 10^{-4}T + 1.28641212 \times 10^{-7}T^2}{1 + 8.42420235 \times 10^{-4}T + 7.08145163 \times 10^{-7}T^2} \\ v(T) = \frac{0.317398726 + 4.22806245 \times 10^{-5}T + 4.20481691 \times 10^{-8}T^2}{1 - 2.89741816 \times 10^{-5}T + 1.61456053 \times 10^{-7}T^2}$$$$ This approximation is accurate to roughly $9 \times 10^{-5}$ in the range 1,000K to 15,000K. From the CIE 1960 space we can compute the coordinates in xyY space (CIES 1931), using the formula from equation $\ref{cieToxyY}$. $$$$\label{cieToxyY} x = \frac{3u}{2u - 8v + 4} \\ y = \frac{2v}{2u - 8v + 4}$$$$ The formulas above are valid for black body color temperatures, and therefore correlated color temperatures of standard illuminants. If we wish to compute the precise chromaticity coordinates of standard CIE illuminants in the D series we can use equation $\ref{seriesDtoxyY}$. $$$$\label{seriesDtoxyY} x = \begin{cases} 0.244063 + 0.09911 \frac{10^3}{T} + 2.9678 \frac{10^6}{T^2} - 4.6070 \frac{10^9}{T^3} & 4,000K \le T \le 7,000K \\ 0.237040 + 0.24748 \frac{10^3}{T} + 1.9018 \frac{10^6}{T^2} - 2.0064 \frac{10^9}{T^3} & 7,000K \le T \le 25,000K \end{cases} \\ y = -3x^2 + 2.87 x - 0.275$$$$ From the xyY space, we can then convert to the CIE XYZ space (equation $\ref{xyYtoXYZ}$). $$$$\label{xyYtoXYZ} X = \frac{xY}{y} \\ Z = \frac{(1 - x - y)Y}{y}$$$$ For our needs, we will fix $Y = 1$. This allows us to convert from the XYZ space to linear RGB with a simple 3x3 matrix, as shown in equation $\ref{XYZtoRGB}$. $$$$\label{XYZtoRGB} \left[ \begin{matrix} R \\ G \\ B \end{matrix} \right] = M^{-1} \left[ \begin{matrix} X \\ Y \\ Z \end{matrix} \right]$$$$ The transformation matrix M is calculated from the target RGB color space primaries. Equation $\ref{XYZtoRGBValues}$ shows the conversion using the inverse matrix for the sRGB color space. $$$$\label{XYZtoRGBValues} \left[ \begin{matrix} R \\ G \\ B \end{matrix} \right] = \left[ \begin{matrix} 3.2404542 & -1.5371385 & -0.4985314 \\ -0.9692660 & 1.8760108 & 0.0415560 \\ 0.0556434 & -0.2040259 & 1.0572252 \end{matrix} \right] \left[ \begin{matrix} X \\ Y \\ Z \end{matrix} \right]$$$$ The result of these operations is a linear RGB triplet in the sRGB color space. Since we care about the chromaticity of the results, we must apply a normalization step to avoid clamping values greater than 1.0 and distort resulting colors: $$$$\label{normalizedRGB} \hat{C}_{linear} = \frac{C_{linear}}{max(C_{linear})}$$$$ We must finally apply the sRGB opto-electronic conversion function (OECF, shown in equation $\ref{OECFsRGB}$) to obtain a displayable value (the value should remain linear if passed to the renderer for shading). $$$$\label{OECFsRGB} C_{sRGB} = \begin{cases} 12.92 \times \hat{C}_{linear} & \hat{C}_{linear} \le 0.0031308 \\ 1.055 \times \hat{C}_{linear}^{\frac{1}{2.4}} - 0.055 & \hat{C}_{linear} \gt 0.0031308 \end{cases}$$$$ For convenience, figure [colorTemperatureScaleCCT] shows the range of correlated color temperatures from 1,000K to 12,500K. All the colors used below assume CIE $D_{65}$ as the white point (as is the case in the sRGB color space). ![Figure [colorTemperatureScaleCCT]: Scale of correlated color temperatures](images/diagram_color_temperature_cct.png) Similarly, figure [colorTemperatureScaleCIE] shows the range of CIE standard illuminants series D from 1,000K to 12,500K. ![Figure [colorTemperatureScaleCIE]: Scale of CIE standard illuminants series D](images/diagram_color_temperature_cie.png) For reference, figure [colorTemperatureScaleCCTClamped] shows the range of correlated color temperatures without the normalization step presented in equation $\ref{normalizedRGB}$. ![Figure [colorTemperatureScaleCCTClamped]: Unnormalized scale of correlated color temperatures](images/diagram_color_temperature_cct_clamped.png) Table [colorTemperatureSamples] presents the correlated color temperature of various common light sources as sRGB color swatches. These colors are relative to the $D_{65}$ white point, so their perceived hue might vary based on your display's white point. See [What colour is the Sun?](http://jila.colorado.edu/~ajsh/colour/Tspectrum.html) for more information. Temperature (K) | Light source | Color --------------------:|:-----------------------------|------------------------------------------------------- 1,700-1,800 | Match flame |

1,850-1,930 | Candle flame |

2,000-3,000 | Sun at sunrise/sunset |

2,500-2,900 | Household tungsten lightbulb |

3,000 | Tungsten lamp 1K |

3,200-3,500 | Quartz lights |

3,200-3,700 | Fluorescent lights |

3,275 | Tungsten lamp 2K |

3,380 | Tungsten lamp 5K, 10K |

5,000-5,400 | Sun at noon |

5,500-6,500 | Daylight (sun + sky) |

5,500-6,500 | Sun through clouds/haze |

6,000-7,500 | Overcast sky |

6,500 | RGB monitor white point |

7,000-8,000 | Shaded areas outdoors |

8,000-10,000 | Partly cloudy sky |

[Table [colorTemperatureSamples]: Normalized correlated color temperatures for common light sources] ### Pre-exposed lights Physically based rendering and physical light units pose an interesting challenge: how to store and handle the large range of values produced by the lighting code? Assuming computations performed at full precision in the shaders, we still want to be able to store the linear output of the lighting pass in a reasonably sized buffer (RGB16F or equivalent). The most obvious and easiest way to achieve this is to simply apply the camera exposure (see the Physically based camera section for more information) before writing out the result of the lighting pass. This simple step is shown in listing [preexposedLighting]: ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ fragColor = luminance * camera.exposure; ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ [Listing [preexposedLighting]: The output of the lighting pass is pre-exposed to fit in half-float buffers] This solution solves the storage problem but requires intermediate computations to be performed with single precision floats. We would instead prefer to perform all (or at least most) of the lighting work using half precision floats instead. Doing so can greatly improve performance and power usage, particularly on mobile devices. Half precision floats are however ill-suited for this kind of work as common illuminance and luminance values (for the sun for instance) can exceed their range. The solution is to simply pre-expose the lights themselves instead of the result of the lighting pass. This can be done efficiently on the CPU if updating a light's constant buffer is cheap. This can also be done on the GPU, as shown in listing [preexposedLights]. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ // The inputs must be highp/single precision, // both for range (intensity) and precision (exposure) // The output is mediump/half precision float computePreExposedIntensity(highp float intensity, highp float exposure) { return intensity * exposure; } Light getPointLight(uint index) { Light light; uint lightIndex = // fetch light index; // the intensity must be highp/single precision highp vec4 colorIntensity = lightsUniforms.lights[lightIndex][1]; // pre-expose the light light.colorIntensity.w = computePreExposedIntensity( colorIntensity.w, frameUniforms.exposure); return light; } ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ [Listing [preexposedLights]: Pre-exposing lights allows the entire shading pipeline to use half precision floats] In practice we pre-expose the following lights: - Punctual lights (point and spot): on the GPU - Directional light: on the CPU - IBLs: on the CPU - Material emissive: on the GPU ## Image based lights In real life, light comes from every direction either directly from light sources or indirectly after bouncing off objects in the environment, being partially absorbed in the process. In a way the whole environment around an object can be seen as a light source. Images, in particular cubemaps, are a great way to encode such an “environment light”. This is called Image Based Lighting (IBL) or sometimes Indirect Lighting. ![Figure [iblBall]: The object shown here is lit only by image-encoded environment lights. Notice the subtle lighting effects that can be applied using this technique.](images/screenshot_ball_ibl.png) There are limitations with image-based lighting. Obviously the environment image must be acquired somehow and as we'll see below it needs to be pre-processed before it can be used for lighting. Typically, the environment image is acquired offline in the real world, or generated by the engine either offline or at run time; either way, local or distant probes are used. These probes can be used to acquire the distant or local environment. In this document, we're focusing on distant environment probes, where the light is assumed to come from infinitely far away (which means every point on the object's surface uses the same environment map). The whole environment contributes light to a given point on the object's surface; this is called _irradiance_ ($E$). The resulting light bouncing off of the object is called radiance ($L_{out}$). Incident lighting must be applied consistently to the diffuse and specular parts of the BRDF. The radiance $L_{out}$ resulting from the interaction between an image based light's (IBL) irradiance and a material model (BRDF) $f(\Theta)$[^ibl1] is computed as follows: $$$$L_{out}(n, v, \Theta) = \int_\Omega f(l, v, \Theta) L_{\bot}(l) \left< \NoL \right> dl$$$$ Note that here we're looking at the behavior of the surface at **macro** level (not to be confused with the micro level equation), which is why it only depends on $\vec n$ and $\vec v$. Essentially, we're applying the BRDF to “point-lights” coming from all directions and encoded in the IBL. ### IBL Types ### There are four common types of IBLs used in modern rendering engines: - **Distant light probes**, used to capture lighting information at "infinity", where parallax can be ignored. Distant probes typically contain the sky, distant landscape features or buildings, etc. They are either captured by the engine or acquired from a camera as high dynamic range images (HDRI). - **Local light probes**, used to capture a certain area of the world from a specific point of view. The capture is projected on a cube or sphere depending on the surrounding geometry. Local probes are more accurate than distance probes and are particularly useful to add local reflections to materials. - **Planar reflections**, used to capture reflections by rendering the scene mirrored by a plane. This technique works only for flat surfaces such as building floors, roads and water. - **Screen space reflection**, used to capture reflections based on the rendered scene (using the previous frame for instance) by ray-marching in the depth buffer. SSR gives great result but can be very expensive. In addition we must distinguish between static and dynamic IBLs. Implementing a fully dynamic day/night cycle requires for instance to recompute the distant light probes dynamically[^iblTypes1]. Both planar and screen space reflections are inherently dynamic. ### IBL Unit ### As discussed previously in the direct lighting section, all our lights must use physical units. As such our IBLs will use the luminance unit $\frac{cd}{m^2}$, which is also the output unit of all our direct lighting equations. Using the luminance unit is straightforward for light probes captures by the engine (dynamically or statically offline). High dynamic range images are a bit more delicate to handle however. Cameras do not record measured luminance but a device-dependent value that is only _related_ to the original scene luminance. As such, we must provide artists with a multiplier that allows them to recover, or at the very least closely approximate, the original absolute luminance. To properly reconstruct the luminance of an HDRI for IBL, artists must do more than simply take photos of the environment and record extra information: - **Color calibration**: using a gray card or a [MacBeth ColorChecker](http://en.wikipedia.org/wiki/ColorChecker) - **Camera settings**: aperture, shutter and ISO - **Luminance samples**: using a spot/luminance meter [TODO] Measure and list common luminance values (clear sky, interior, etc.) ### Processing light probes ### We saw previously that the radiance of an IBL is computed by integrating over the surface's hemisphere. Since this would obviously be too expensive to do in real-time, we must first pre-process our light probes to convert them into a format better suited for real-time interactions. The sections below will discuss the techniques used to accelerate the evaluation of light probes: - **Specular reflectance**: pre-filtered importance sampling and split-sum approximation - **Diffuse reflectance**: irradiance map and spherical harmonics ### Distant light probes ### #### Diffuse BRDF integration #### Using the Lambertian BRDF[^iblDiffuse1], we get the radiance: \begin{align*} f_d(\sigma) &= \frac{\sigma}{\pi} \\ L_d(n, \sigma) &= \int_{\Omega} f_d(\sigma) L_{\bot}(l) \left< \NoL \right> dl \\ &= \frac{\sigma}{\pi} \int_{\Omega} L_{\bot}(l) \left< \NoL \right> dl \\ &= \frac{\sigma}{\pi} E_d(n) \quad \text{with the irradiance} \; E_d(n) = \int_{\Omega} L_{\bot}(l) \left< \NoL \right> dl \end{align*} Or in the discrete domain: $$E_d(n) \equiv \sum_{\forall \, i \in image} L_{\bot}(s_i) \left< n \cdot s_i \right> \Omega_s$$ $\Omega_s$ is the solid-angle[^iblDiffuse2] associated to sample $i$. The irradiance integral $\Ed$ can be trivially, albeit slowly[^iblDiffuse3], precomputed and stored into a cubemap for efficient access at runtime. Typically, _image_ is a cubemap or an equirectangular image. The term $\frac{\sigma}{\pi}$ is independent of the IBL and is added at runtime to obtain the _radiance_. ![Figure [iblOriginal]: Image-based environment](images/ibl/ibl_river_roughness_m0.png style="max-width:100%;") ![Figure [iblIrradiance]: Image-based irradiance map using the Lambertian BRDF](images/ibl/ibl_irradiance.png style="max-width:100%;") [^ibl1]: $\Theta$ represents the parameters of the material model $f$, i.e.: _roughness_, albedo and so on... [^iblTypes1]: This can be done through blending of static probes or by spreading the workload over time [^iblDiffuse1]: The Lambertian BRDF doesn't depend on $\vec l$, $\vec v$ or $\theta$, so $L_d(n,v,\theta) \equiv L_d(n,\sigma)$ [^iblDiffuse2]: $\Omega_s$ can be approximated by $\frac{2\pi}{6 \cdot width \cdot height}$ for a cubemap [^iblDiffuse3]: $O(12\,n^2\,m^2)$, with $n$ and $m$ respectively the dimensions of the environment and the precomputed cubemap However, the irradiance can also be approximated very closely by a decomposition into Spherical Harmonics (SH, described in more details in the Spherical Harmonics section) and calculated at runtime cheaply. It is usually best to avoid texture fetches on mobile and free-up a texture unit. Even if it is stored into a cubemap, it is orders of magnitude faster to pre-compute the integral using SH decomposition followed by a rendering. SH decomposition is similar in concept to a Fourier transform, it expresses the signal over an orthonormal base in the frequency domain. The properties that interests us most are: - Very few coefficients are needed to encode $\cosTheta$ - Convolutions by a kernel that _has a circular symmetry_ are very inexpensive and become products in SH space In practice only 4 or 9 coefficients (i.e.: 2 or 3 bands) are enough for $\cosTheta$ meaning we don't need more either for $\Lt$. ![Figure [iblSH3]: 3 bands (9 coefficients)](images/ibl/ibl_irradiance_sh3.png style="max-width:100%;") ![Figure [iblSH2]: 2 bands (4 coefficients)](images/ibl/ibl_irradiance_sh2.png style="max-width:100%;") In practice we pre-convolve $\Lt$ with $\cosTheta$ and pre-scale these coefficients by the basis scaling factors $K_l^m$ so that the reconstruction code is as simple as possible in the shader: ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ vec3 irradianceSH(vec3 n) { // uniform vec3 sphericalHarmonics[9] // We can use only the first 2 bands for better performance return sphericalHarmonics[0] + sphericalHarmonics[1] * (n.y) + sphericalHarmonics[2] * (n.z) + sphericalHarmonics[3] * (n.x) + sphericalHarmonics[4] * (n.y * n.x) + sphericalHarmonics[5] * (n.y * n.z) + sphericalHarmonics[6] * (3.0 * n.z * n.z - 1.0) + sphericalHarmonics[7] * (n.z * n.x) + sphericalHarmonics[8] * (n.x * n.x - n.y * n.y); } ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ [Listing [irradianceSH]: GLSL code to reconstruct the irradiance from the pre-scaled SH] Note that with 2 bands, the computation above becomes a single $4 \times 4$ matrix-by-vector multiply. Additionally, because of the pre-scaling by $K_l^m$, the SH coefficients can be thought of as colors, in particular sphericalHarmonics[0] is directly the average irradiance. #### Specular BRDF integration #### As we've seen above, the radiance $\Lout$ resulting from the interaction between an IBL's irradiance and a BRDF is: $$$$\label{specularBRDFIntegration} \Lout(n, v, \Theta) = \int_\Omega f(l, v, \Theta) \Lt(l) \left< \NoL \right> \partial l$$$$ We recognize the convolution of $\Lt$ by $f(l, v, \Theta) \left< \NoL \right>$, i.e.: the environment is *filtered* using the BRDF as a kernel. Indeed at higher roughness, specular reflections look more *blurry*. Plugging the expression of $f$ in equation $\ref{specularBRDFIntegration}$, we obtain: $$$$\Lout(n,v,\Theta) = \int_\Omega D(l, v, \alpha) F(l, v, f_0, f_{90}) V(l, v, \alpha) \left< \NoL \right> \Lt(l) \partial l$$$$ This expression depends on $v$, $\alpha$, $f_0$ and $f_{90}$ inside the integral, which makes its evaluation extremely costly and unsuitable for real-time on mobile (even using pre-filtered importance sampling). ##### Simplifying the BRDF integration ##### Since there is no closed-form solution or an easy way to compute the $\Lout$ integral, we use a simplified equation instead: $\hat{I}$, whereby we assume that $v = n$, that is the view direction $v$ is always equal to the surface normal $n$. Clearly, this assumption will break all view-dependant effects of the convolution, such as the increased blur in reflections closer to the viewer (a.k.a. stretchy reflections). Such a simplification would also have a severe impact on constant environments, such as the white furnace, because it would affect the magnitude of the constant (i.e. DC) term of the result. We can at least correct for that by using a scale factor, $K$, in our simplified integral, which will make sure the average irradiance stay correct when chosen properly. - $I$ is our original integral, i.e.: $I(g) = \int_\Omega g(l) \left< \NoL \right> \partial l$ - $\hat{I}$ is the simplified integral where $v = n$ - $K$ is a scale factor that ensures the average irradiance is unchanged by $\hat{I}$ - $\tilde{I}$ is our final approximation of $I$, $\tilde{I} = \hat{I} \times K$ Because $I$ is an integral multiplications can be distributed over it. i.e.: $I(g()f()) = I(g())I(f())$. Armed with that, $$$$I( f(\Theta) \Lt ) \approx \tilde{I}( f(\Theta) \Lt ) \\ \tilde{I}( f(\Theta) \Lt ) = K \times \hat{I}( f(\Theta) \Lt ) \\ K = \frac{I(f(\Theta))}{\hat{I}(f(\Theta))}$$$$ From the equation above we can see that $\tilde{I}$ is equivalent to $I$ when $\Lt$ is a constant, and yields the correct result: \begin{align*} \tilde{I}(f(\Theta)\Lt^{constant}) &= \Lt^{constant} \hat{I}(f(\Theta)) \frac{I(f(\Theta))}{\hat{I}(f(\Theta))} \\ &= \Lt^{constant} I(f(\Theta)) \\ &= I(f(\Theta)\Lt^{constant}) \end{align*} Similarly, we can also demonstrate that the result is correct when $v = n$, since in that case $I = \hat{I}$: \begin{align*} \tilde{I}(f(\Theta)\Lt) &= I(f(\Theta)\Lt) \frac{I(f(\Theta))}{I(f(\Theta))} \\ &= I(f(\Theta)\Lt) \end{align*} Finally, we can show that the scale factor $K$ satisfies our average irradiance ($\bar{\Lt}$) requirement by plugging $\Lt = \bar{\Lt} + (\Lt - \bar{\Lt}) = \bar{\Lt} + \Delta\Lt$ into $\tilde{I}$: \begin{align*} \tilde{I}(f(\Theta)\Lt) &= \tilde{I}\left[f\left(\Theta\right) \left(\bar{\Lt} + \Delta\Lt\right)\right] \\ &= K \times \hat{I}\left[f\left(\Theta\right) \left(\bar{\Lt} + \Delta\Lt\right)\right] \\ &= K \times \left[\hat{I}\left(f\left(\Theta\right)\bar{\Lt}\right) + \hat{I}\left(f\left(\Theta\right)\Delta\Lt\right)\right] \\ &= K \times \hat{I}\left(f\left(\Theta\right)\bar{\Lt}\right) + K \times \hat{I}\left(f\left(\Theta\right) \Delta\Lt\right) \\ &= \tilde{I}\left(f\left(\Theta\right)\bar{\Lt}\right) + \tilde{I}\left(f\left(\Theta\right) \Delta\Lt\right) \\ &= I\left(f\left(\Theta\right)\bar{\Lt}\right) + \tilde{I}\left(f\left(\Theta\right) \Delta\Lt\right) \end{align*} The above result shows that the average irradiance is computed correctly, i.e.: $I(f(\Theta)\bar{\Lt})$. A way to think about this approximation is that it splits the radiance $\Lt$ in two parts, the average $\bar{\Lt}$ and the delta from the average $\Delta\Lt$ and computes the correct integration of the average part then adds the simplified integration of the delta part: $$$$approximation(\Lt) = correct(\bar{\Lt}) + simplified(\Lt - \bar{\Lt})$$$$ Now, let's look at each term: $$$$\label{iblPartialEquations} \hat{I}(f(n, \alpha) \Lt) = \int_\Omega f(l, n, \alpha) \Lt(l) \left< \NoL \right> \partial l \\ \hat{I}(f(n, \alpha)) = \int_\Omega f(l, n, \alpha) \left< \NoL \right> \partial l \\ I(f(n, v, \alpha)) = \int_\Omega f(l, n, v, \alpha) \left< \NoL \right> \partial l$$$$ All three of these equations can be easily pre-calculated and stored in look-up tables, as explained below. ##### Discrete Domain ##### In the discrete domain the equations in \ref{iblPartialEquations} become: $$$$\hat{I}(f(n, \alpha) \Lt) \equiv \frac{1}{N}\sum_{\forall \, i \in image} f(l_i, n, \alpha) \Lt(l_i) \left<\NoL\right> \\ \hat{I}(f(n, \alpha)) \equiv \frac{1}{N}\sum_{\forall \, i \in image} f(l_i, n, \alpha) \left<\NoL\right> \\ I(f(n, v, \alpha)) \equiv \frac{1}{N}\sum_{\forall \, i \in image} f(l_i, n, v, \alpha) \left<\NoL\right>$$$$ However, in practice we're using _importance sampling_ which needs to take the $pdf$ of the distribution into account and adds a term $\frac{\left<\VoH\right>}{D(h_i, \alpha)\left<\NoH\right>}$. See Importance Sampling For The IBL section: $$$$\label{iblImportanceSampling} \hat{I}(f(n, \alpha) \Lt) \equiv \frac{4}{N}\sum_i^N f(l_i, n, \alpha) \frac{\left<\VoH\right>}{D(h_i, \alpha)\left<\NoH\right>} \Lt(l_i) \left<\NoL\right> \\ \hat{I}(f(n, \alpha)) \equiv \frac{4}{N}\sum_i^N f(l_i, n, \alpha) \frac{\left<\VoH\right>}{D(h_i, \alpha)\left<\NoH\right>} \left<\NoL\right> \\ I(f(n, v, \alpha)) \equiv \frac{4}{N}\sum_i^N f(l_i, n, v, \alpha) \frac{\left<\VoH\right>}{D(h_i, \alpha)\left<\NoH\right>} \left<\NoL\right>$$$$ Recalling that for $\hat{I}$, we assume that $v = n$, equations \ref{iblImportanceSampling}, simplifies to: $$$$\hat{I}(f(n, \alpha) \Lt) \equiv \frac{4}{N}\sum_i^N \frac{f(l_i, n, \alpha)}{D(h_i, \alpha)} \Lt(l_i) \left<\NoL\right> \\ \hat{I}(f(n, \alpha)) \equiv \frac{4}{N}\sum_i^N \frac{f(l_i, n, \alpha)}{D(h_i, \alpha)} \left<\NoL\right> \\ I(f(n, v, \alpha)) \equiv \frac{4}{N}\sum_i^N \frac{f(l_i, n, v, \alpha)}{D(h_i, \alpha)} \frac{\left<\VoH\right>}{\left<\NoH\right>} \left<\NoL\right>$$$$ Then, the first two equations can be merged together such that $LD(n, \alpha) = \frac{\hat{I}(f(n, \alpha) \Lt)}{\hat{I}(f(n, \alpha))}$ $$$$\label{iblLD} LD(n, \alpha) \equiv \frac{\sum_i^N \frac{f(l_i, n, \alpha)}{D(h_i, \alpha)} \Lt(l_i) \left<\NoL\right>}{\sum_i^N \frac{f(l_i, n, \alpha)}{D(h_i, \alpha)}\left<\NoL\right>}$$$$ $$$$\label{iblDFV} I(f(n, v, \alpha)) \equiv \frac{4}{N}\sum_i^N \frac{f(l_i, n, v, \alpha)}{D(h_i, \alpha)} \frac{\left<\VoH\right>}{\left<\NoH\right>} \left<\NoL\right>$$$$ Note that at this point, we could almost compute both remaining equations off-line. The only difficulty is that we don't know $f_0$ nor $f_{90}$ when we precompute those integrals. We will see below that we can incorporate these terms at runtime for equation \ref{iblDFV}, alas, this is not possible for equation \ref{iblLD} and we have to assume $f_0 = f_{90} = 1$ (i.e.: the fresnel term always evaluates to 1). We also have to deal with the visibility term of the brdf, in practice keeping it yields to slightly worst results compared to the ground truth, so we also set $V = 1$. Let's substitute $f$ in equations \ref{iblLD} and \ref{iblDFV}: $$$$f(l_i, n, \alpha) = D(h_i, \alpha)F(f_0, f_{90}, \left<\VoH\right>)V(l_i, v, \alpha)$$$$ The first simplification is that the term $D(h_i, \alpha)$ in the brdf cancels out with the denominator (which came from the $pdf$ due to importance sampling) and F and V disappear since we assume their value is 1. $$$$LD(n, \alpha) \equiv \frac{\sum_i^N V(l_i, v, \alpha)\left<\NoL\right>\Lt(l_i) }{\sum_i^N \left<\NoL\right>}$$$$ $$$$\label{iblFV} I(f(n, v, \alpha)) \equiv \frac{4}{N}\sum_i^N \color{green}{F(f_0, f_{90}, \left<\VoH\right>)} V(l_i, v, \alpha)\frac{\left<\VoH\right>}{\left<\NoH\right>} \left<\NoL\right>$$$$ Now, let's substitute the fresnel term into equation \ref{iblFV}: $$$$F(f_0, f_{90}, \left<\VoH\right>) = f_0 (1 - F_c(\left<\VoH\right>)) + f_{90} F_c(\left<\VoH\right>) \\ F_c(\left<\VoH\right>) = (1 - \left<\VoH\right>)^5$$$$ $$$$I(f(n, v, \alpha)) \equiv \frac{4}{N}\sum_i^N \left[\color{green}{f_0 (1 - F_c(\left<\VoH\right>)) + f_{90} F_c(\left<\VoH\right>)}\right] V(l_i, v, \alpha)\frac{\left<\VoH\right>}{\left<\NoH\right>} \left<\NoL\right> \\$$$$ \begin{align*} I(f(n, v, \alpha)) \equiv & \color{green}{f_0 } \frac{4}{N}\sum_i^N \color{green}{(1 - F_c(\left<\VoH\right>))} V(l_i, v, \alpha)\frac{\left<\VoH\right>}{\left<\NoH\right>} \left<\NoL\right> \\ + & \color{green}{f_{90}} \frac{4}{N}\sum_i^N \color{green}{ F_c(\left<\VoH\right>) } V(l_i, v, \alpha)\frac{\left<\VoH\right>}{\left<\NoH\right>} \left<\NoL\right> \end{align*} And finally, we extract the equations that can be calculated off-line (i.e.: the part that doesn't depend on the runtime parameters $f_0$ and $f_{90}$): $$$$\label{iblAllEquations} DFG_1(\alpha, \left<\NoV\right>) = \frac{4}{N}\sum_i^N \color{green}{(1 - F_c(\left<\VoH\right>))} V(l_i, v, \alpha)\frac{\left<\VoH\right>}{\left<\NoH\right>} \left<\NoL\right> \\ DFG_2(\alpha, \left<\NoV\right>) = \frac{4}{N}\sum_i^N \color{green}{ F_c(\left<\VoH\right>) } V(l_i, v, \alpha)\frac{\left<\VoH\right>}{\left<\NoH\right>} \left<\NoL\right> \\ I(f(n, v, \alpha)) \equiv \color{green}{f_0} \color{red}{DFG_1(\alpha, \left<\NoV\right>)} + \color{green}{f_{90}} \color{red}{DFG_2(\alpha, \left<\NoV\right>)}$$$$ Notice that $DFG_1$ and $DFG_2$ only depend on $\NoV$, that is the angle between the normal $n$ and the view direction $v$. This is true because the integral is symmetrical with respect to $n$. When integrating, we can choose any $v$ we please as long as it satisfies $\NoV$ (e.g.: when calculating $\VoH$). Putting everything back together: \begin{align*} \Lout(n,v,\alpha,f_0,f_{90}) &\simeq \big[ f_0 \color{red}{DFG_1(\NoV, \alpha)} + f_{90} \color{red}{DFG_2(\NoV, \alpha)} \big] \times LD(n, \alpha) \\ DFG_1(\alpha, \left<\NoV\right>) &= \frac{4}{N}\sum_i^N \color{green}{(1 - F_c(\left<\VoH\right>))} V(l_i, v, \alpha)\frac{\left<\VoH\right>}{\left<\NoH\right>} \left<\NoL\right> \\ DFG_2(\alpha, \left<\NoV\right>) &= \frac{4}{N}\sum_i^N \color{green}{ F_c(\left<\VoH\right>) } V(l_i, v, \alpha)\frac{\left<\VoH\right>}{\left<\NoH\right>} \left<\NoL\right> \\ LD(n, \alpha) &= \frac{\sum_i^N V(l_i, n, \alpha)\left<\NoL\right>\Lt(l_i) }{\sum_i^N \left<\NoL\right>} \end{align*} #### The $DFG_1$ and $DFG_2$ term visualized #### Both $DFG_1$ and $DFG_2$ can either be pre-calculated in a regular 2D texture indexed by $(\NoV, \alpha)$ and sampled bilinearly, or computed at runtime using an analytic approximation of the surfaces. See sample code in the annex. The pre-calculated textures are shown in table [textureDFG]. A C++ implementation of the pre-computation can be found in section [Precomputing L for image-based lighting]. $DFG_1$ | $DFG_2$ | ${ DFG_1, DFG_2, 0 }$ -------------------------|--------------------------|---------------------- ![](images/ibl/dfg1.png) | ![](images/ibl/dfg2.png) | ![](images/ibl/dfg.png) [Table [textureDFG]: Y axis: $\alpha$. X axis: $cos \theta$] $DFG_1$ and $DFG_2$ are conveniently within the $[0, 1]$ range, however 8-bits textures don't have enough precision and will cause problems. Unfortunately, on mobile, 16-bits or float textures are not ubiquitous and there are a limited number of samplers. Despite the attractive simplicity of the shader code using a texture, it might be better to use an analytic approximation. Note however that since we only need to store two terms, OpenGL ES 3.0's RG16F texture format is a good candidate. Such analytic approximation is described in [#Karis14], itself based on [#Lazarov13]. [#Narkowicz14] is another interesting approximation. Note that these two approximations are not compatible with the energy compensation term presented in section [Pre-integration for multiscattering]. Table [textureApproxDFG] presents a visual representation of these approximations. $DFG_1$ | $DFG_2$ | ${ DFG_1, DFG_2, 0 }$ --------------------------------|---------------------------------|---------------------- ![](images/ibl/dfg1_approx.png) | ![](images/ibl/dfg2_approx.png) | ![](images/ibl/dfg_approx.png) [Table [textureApproxDFG]: Y axis: $\alpha$. X axis: $cos \theta$] #### The $LD$ term visualized #### $LD$ is the convolution of the environment by a function that only depends on the $\alpha$ parameter (itself related to the roughness, see section [Roughness remapping and clamping]). $LD$ can conveniently be stored in a mip-mapped cubemap where increasing LODs receive the environment pre-filtered with increasing roughness. This works well because this convolution is a powerful low-pass filter. To make good use of each mipmap level, it is necessary to remap $\alpha$; we find that using a power remapping with $\gamma = 2$ works well and is convenient. \begin{align*} \alpha &= perceptualRoughness^2 \\ lod_{\alpha} &= \alpha^{\frac{1}{2}} = perceptualRoughness \\ \end{align*} See an example below: ![$\alpha=0.0$](images/ibl/ibl_river_roughness_m0.png style="max-width:100%;") ![$\alpha=0.2$](images/ibl/ibl_river_roughness_m1.png style="max-width:100%;") ![$\alpha=0.4$](images/ibl/ibl_river_roughness_m2.png style="max-width:100%;") ![$0.6$](images/ibl/ibl_river_roughness_m3.png style="max-width:100%;") ![$0.8$](images/ibl/ibl_river_roughness_m4.png style="max-width:100%;") #### Indirect specular and indirect diffuse components visualized #### Figure [iblVisualized] shows how indirect lighting interacts with dielectrics and conductors. Direct lighting was removed for illustration purposes. ![Figure [iblVisualized]: Indirect diffuse and specular decomposition](images/ibl/ibl_visualization.jpg) #### IBL evaluation implementation #### Listing [iblEvaluation] presents a GLSL implementation to evaluate the IBL, using the various textures described in the previous sections. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ vec3 ibl(vec3 n, vec3 v, vec3 diffuseColor, vec3 f0, vec3 f90, float perceptualRoughness) { vec3 r = reflect(n); vec3 Ld = textureCube(irradianceEnvMap, r) * diffuseColor; float lod = computeLODFromRoughness(perceptualRoughness); vec3 Lld = textureCube(prefilteredEnvMap, r, lod); vec2 Ldfg = textureLod(dfgLut, vec2(dot(n, v), perceptualRoughness), 0.0).xy; vec3 Lr = (f0 * Ldfg.x + f90 * Ldfg.y) * Lld; return Ld + Lr; } ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ [Listing [iblEvaluation]: GLSL implementation of image based lighting evaluation] We can however save a couple of texture lookups by using Spherical Harmonics instead of an irradiance cubemap and the analytical approximation of the $DFG$ LUT, as shown in listing [optimizedIblEvaluation]. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ vec3 irradianceSH(vec3 n) { // uniform vec3 sphericalHarmonics[9] // We can use only the first 2 bands for better performance return sphericalHarmonics[0] + sphericalHarmonics[1] * (n.y) + sphericalHarmonics[2] * (n.z) + sphericalHarmonics[3] * (n.x) + sphericalHarmonics[4] * (n.y * n.x) + sphericalHarmonics[5] * (n.y * n.z) + sphericalHarmonics[6] * (3.0 * n.z * n.z - 1.0) + sphericalHarmonics[7] * (n.z * n.x) + sphericalHarmonics[8] * (n.x * n.x - n.y * n.y); } // NOTE: this is the DFG LUT implementation of the function above vec2 prefilteredDFG_LUT(float coord, float NoV) { // coord = sqrt(roughness), which is the mapping used by the // IBL prefiltering code when computing the mipmaps return textureLod(dfgLut, vec2(NoV, coord), 0.0).rg; } vec3 evaluateSpecularIBL(vec3 r, float perceptualRoughness) { // This assumes a 256x256 cubemap, with 9 mip levels float lod = 8.0 * perceptualRoughness; // decodeEnvironmentMap() either decodes RGBM or is a no-op if the // cubemap is stored in a float texture return decodeEnvironmentMap(textureCubeLodEXT(environmentMap, r, lod)); } vec3 evaluateIBL(vec3 n, vec3 v, vec3 diffuseColor, vec3 f0, vec3 f90, float perceptualRoughness) { float NoV = max(dot(n, v), 0.0); vec3 r = reflect(-v, n); // Specular indirect vec3 indirectSpecular = evaluateSpecularIBL(r, perceptualRoughness); vec2 env = prefilteredDFG_LUT(perceptualRoughness, NoV); vec3 specularColor = f0 * env.x + f90 * env.y; // Diffuse indirect // We multiply by the Lambertian BRDF to compute radiance from irradiance // With the Disney BRDF we would have to remove the Fresnel term that // depends on NoL (it would be rolled into the SH). The Lambertian BRDF // can be baked directly in the SH to save a multiplication here vec3 indirectDiffuse = max(irradianceSH(n), 0.0) * Fd_Lambert(); // Indirect contribution return diffuseColor * indirectDiffuse + indirectSpecular * specularColor; } ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ [Listing [optimizedIblEvaluation]: GLSL implementation of image based lighting evaluation] #### Pre-integration for multiscattering #### In section [Energy loss in specular reflectance] we discussed how to use a second scaled specular lobe to compensate for the energy loss due to only accounting for a single scattering event in our BRDF. This energy compensation lobe is scaled by a term that depends on $r$ defined in the following way: $$$$r = \int_{\Omega} D(l,v) V(l,v) \left< \NoL \right> \partial l$$$$ Or, evaluated with importance sampling (See Importance Sampling For The IBL section): $$$$r \equiv \frac{4}{N}\sum_i^N V(l_i, v, \alpha)\frac{\left<\VoH\right>}{\left<\NoH\right>} \left<\NoL\right>$$$$ This equality is very similar to the terms $DFG_1$ and $DFG_2$ seen in equation $\ref{iblAllEquations}$. In fact, it's the same, except without the Fresnel term. By making the further assumption that $f_{90} = 1$, we can rewrite $DFG_1$ and $DFG_2$ and the $\Lout$ reconstruction: \begin{align*} \Lout(n,v,\alpha,f_0) &\simeq \big[ (1 - f_0) \color{red}{DFG_1^{multiscatter}(\NoV, \alpha)} + f_0 \color{red}{DFG_2^{multiscatter}(\NoV, \alpha)} \big] \times LD(n, \alpha) \\ DFG_1^{multiscatter}(\alpha, \left<\NoV\right>) &= \frac{4}{N}\sum_i^N \color{green}{F_c(\left<\VoH\right>)} V(l_i, v, \alpha)\frac{\left<\VoH\right>}{\left<\NoH\right>} \left<\NoL\right> \\ DFG_2^{multiscatter}(\alpha, \left<\NoV\right>) &= \frac{4}{N}\sum_i^N V(l_i, v, \alpha)\frac{\left<\VoH\right>}{\left<\NoH\right>} \left<\NoL\right> \\ LD(n, \alpha) &= \frac{\sum_i^N V(l_i, n, \alpha)\left<\NoL\right>\Lt(l_i) }{\sum_i^N V(l_i, n, \alpha)\left<\NoL\right>} \end{align*} These two new $DFG$ terms simply need to replace the ones used in the implementation shown in section [Precomputing L for image-based lighting]: ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ float Fc = pow(1 - VoH, 5.0f); r.x += Gv * Fc; r.y += Gv; ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ [Listing [multiscatterIBLPreintegration]: C++ implementation of the $L_{DFG}$ term for multiscattering] To perform the reconstruction we need to slightly modify listing [multiscatterIBLEvaluation]: ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ vec2 dfg = textureLod(dfgLut, vec2(dot(n, v), perceptualRoughness), 0.0).xy; // (1 - f0) * dfg.x + f0 * dfg.y vec3 specularColor = mix(dfg.xxx, dfg.yyy, f0); ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ [Listing [multiscatterIBLEvaluation]: GLSL implementation of image based lighting evaluation, with multiscattering LUT] #### Summary #### In order to calculate the specular contribution of distant image-based lights, we had to make a few approximations and compromises: - $v = n$, by far the assumption contributing to the largest error when integrating the non-constant part of the IBL. This results in the complete loss of roughness anisotropy with respect to the view point. - Roughness contribution for the non-constant part of the IBL is quantized and trilinear filtering is used to interpolate between these levels. This is most visible at low roughnes (e.g.: around 0.0625 for a 9 LODs cubemap). - Because mipmap levels are used to store the pre-integrated environment, they can't be used for texture minification, as they ought to. This can causes aliasing or moiré artifacts in high frequency regions or the environment at low roughness and/or distant or small objects. This can also impact performance due to the resulting poor cache access pattern. - No Fresnel for the non-constant part of the IBL. - Visibility = 1 for the non-constant part of the IBL. - Schlick's Fresnel - $f_{90} = 1$ in the multiscattering case. ![Figure [iblPrefilterVsImportanceSampling]: Comparison between importance-sampled reference (top) and prefiltered IBL (middle).](images/ibl/ibl_prefilter_vs_reference.png) ![Figure [iblStretchyReflectionLoss]: Error in reflections due to assuming $v = n$ (bottom) -- loss of "stretchy reflections".](images/ibl/ibl_stretchy_reflections_error.png) ![Figure [iblRoughnessInLods0]: Error due to storing the roughness in cubemaps LODs at roughness = 0.0625 (i.e.: sampling exactly between levels). Notice how instead of bluring we see a "cross-fade" between two blurs.](images/ibl/ibl_trilinear_0.png) ![Figure [iblRoughnessInLods1]: Error due to storing the roughness in cubemaps LODs at roughness = 0.125 (i.e.: sampling exactly level 1). When the roughness closely matches a LOD, the error due to trilinear filtering in the cubemap is reduced. Notice the errors due to $v = n$ at grazing angles.](images/ibl/ibl_trilinear_1.png) ![Figure [iblMoirePattern]: Moiré pattern due to texture minification on a metallic sphere at $\alpha = 0$ using an environment made of colored vertical stripes (skybox hidden).](images/ibl/ibl_no_mipmaping.png) ### Clear coat ### When sampling the IBL, the clear coat layer is calculated as a second specular lobe. This specular lobe is oriented along the view direction since we cannot reasonably integrate over the hemisphere. Listing [clearCoatIBL] demonstrates this approximation in practice. It also shows the energy conservation step. It is important to note that this second specular lobe is computed exactly the same way as the main specular lobe, using the same DFG approximation. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ // clearCoat_NoV == shading_NoV if the clear coat layer doesn't have its own normal map float Fc = F_Schlick(0.04, 1.0, clearCoat_NoV) * clearCoat; // base layer attenuation for energy compensation iblDiffuse *= 1.0 - Fc; iblSpecular *= sq(1.0 - Fc); iblSpecular += specularIBL(r, clearCoatPerceptualRoughness) * Fc; ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ [Listing [clearCoatIBL]: GLSL implementation of the clear coat specular lobe for image-based lighting] ### Anisotropy ### [#McAuley15] describes a technique called “bent reflection vector”, based [#Revie12]. The bent reflection vector is a rough approximation of anisotropic lighting but the alternative is to use importance sampling. This approximation is sufficiently cheap to compute and provides good results, as shown in figure [anisotropicIBL1] and figure [anisotropicIBL2]. ![Figure [anisotropicIBL1]: Anisotropic indirect specular reflections using bent normals (left: roughness 0.3, right: roughness: 0.0; both: anisotropy 1.0)](images/screenshot_anisotropic_ibl1.jpg) ![Figure [anisotropicIBL2]: Anisotropic reflections with varying roughness, metallicness, etc.](images/screenshot_anisotropic_ibl2.jpg) The implementation of this technique is straightforward, as demonstrated in listing [bentReflectionVector]. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ vec3 anisotropicTangent = cross(bitangent, v); vec3 anisotropicNormal = cross(anisotropicTangent, bitangent); vec3 bentNormal = normalize(mix(n, anisotropicNormal, anisotropy)); vec3 r = reflect(-v, bentNormal); ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ [Listing [bentReflectionVector]: GLSL implementation of the bent reflection vector] This technique can be made more useful by accepting negative anisotropy values, as shown in listing [bentReflectionVectorDirection]. When the anisotropy is negative, the highlights are not in the direction of the tangent, but in the direction of the bitangent instead. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ vec3 anisotropicDirection = anisotropy >= 0.0 ? bitangent : tangent; vec3 anisotropicTangent = cross(anisotropicDirection, v); vec3 anisotropicNormal = cross(anisotropicTangent, anisotropicDirection); vec3 bentNormal = normalize(mix(n, anisotropicNormal, anisotropy)); vec3 r = reflect(-v, bentNormal); ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ [Listing [bentReflectionVectorDirection]: GLSL implementation of the bent reflection vector] Figure [anisotropicDirection] demonstrates this modified implementation in practice. ![Figure [anisotropicDirection]: Control of the anisotropy direction using positive (left) and negative (right) values](images/screenshot_anisotropy_direction.png) ### Subsurface ### [TODO] Explain subsurface and IBL ### Cloth ### The IBL implementation for the cloth material model is more complicated than for the other material models. The main difference stems from the use of a different NDF ("Charlie" vs height-correlated Smith GGX). As described in this section, we use the split-sum approximation to compute the DFG term of the BRDF when computing an IBL. This DFG term is designed for a different BRDF and cannot be used for the cloth BRDF. Since we designed our cloth BRDF to not need a Fresnel term, we can generate a single DG term in the 3rd channel of the DFG LUT. The result is shown in figure [dfgClothLUT]. The DG term is generated using uniform sampling as recommended in [#Estevez17]. With uniform sampling the $pdf$ is simply $\frac{1}{2\pi}$ and we must still use the Jacobian $\frac{1}{4\left< \VoH \right>}$. ![Figure [dfgClothLUT]: DFG LUT with a 3rd channel encoding the DG term of the cloth BRDF](images/ibl/dfg_cloth.png) The remainder of the image-based lighting implementation follows the same steps as the implementation of regular lights, including the optional subsurface scattering term and its wrap diffuse component. Just as with the clear coat IBL implementation, we cannot integrate over the hemisphere and use the view direction as the dominant light direction to compute the wrap diffuse component. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ float diffuse = Fd_Lambert() * ambientOcclusion; #if defined(SHADING_MODEL_CLOTH) #if defined(MATERIAL_HAS_SUBSURFACE_COLOR) diffuse *= saturate((NoV + 0.5) / 2.25); #endif #endif vec3 indirectDiffuse = irradianceIBL(n) * diffuse; #if defined(SHADING_MODEL_CLOTH) && defined(MATERIAL_HAS_SUBSURFACE_COLOR) indirectDiffuse *= saturate(subsurfaceColor + NoV); #endif vec3 ibl = diffuseColor * indirectDiffuse + indirectSpecular * specularColor; ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ [Listing [clothApprox]: GLSL implementation of the DFG approximation for the cloth NDF] It is important to note that this only addresses part of the IBL problem. The pre-filtered specular environment maps described earlier are convolved with the standard shading model's BRDF, which differs from the cloth BRDF. To get accurate result we should in theory provide one set of IBLs per BRDF used in the engine. Providing a second set of IBLs is however not practical for our use case so we decided to rely on the existing IBLs instead. ## Static lighting [TODO] Spherical-harmonics or spherical-gaussian lightmaps, irradiance volumes, PRT?… ## Transparency and translucency lighting Transparent and translucent materials are important to add realism and correctness to scenes. Filament must therefore provide lighting models for both types of materials to allow artists to properly recreate realistic scenes. Translucency can also be used effectively in a number of non-realistic settings. ### Transparency To properly light a transparent surface, we must first understand how the material's opacity is applied. Observe a window and you will see that the diffuse reflectance is transparent. On the other hand, the brighter the specular reflectance, the less opaque the window appears. This effect can be seen in figure [cameraTransparency]: the scene is properly reflected onto the glass surfaces but the specular highlight of the sun is bright enough to appear opaque. ![Figure [cameraTransparency]: Example of a complex object where lit surface transparency plays an important role](images/screenshot_camera_transparency.jpg) ![Figure [litCar]: Example of a complex object where lit surface transparency plays an important role](images/screenshot_car.jpg) To properly implement opacity, we will use the premultiplied alpha format. Given a desired opacity noted $\alpha_{opacity}$ and a diffuse color $\sigma$ (linear, unpremultiplied), we can compute the effective opacity of a fragment. \begin{align*} color &= \sigma * \alpha_{opacity} \\ opacity &= \alpha_{opacity} \end{align*} The physical interpretation is that the RGB components of the source color define how much light is emitted by the pixel, whereas the alpha component defines how much of the light behind the pixel is blocked by said pixel. We must therefore use the following blending functions: \begin{align*} Blend_{src} &= 1 \\ Blend_{dst} &= 1 - src_{\alpha} \end{align*} The GLSL implementation of these equations is presented in listing [surfaceTransparency]. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ // baseColor has already been premultiplied vec4 shadeSurface(vec4 baseColor) { float alpha = baseColor.a; vec3 diffuseColor = evaluateDiffuseLighting(); vec3 specularColor = evaluateSpecularLighting(); return vec4(diffuseColor + specularColor, alpha); } ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ [Listing [surfaceTransparency]: Implementation of lit surface transparency in GLSL] ### Translucency Translucent materials can be divided into two categories: - Surface translucency - Volume translucency Volume translucency is useful to light particle systems, for instance clouds or smoke. Surface translucency can be used to imitate materials with transmitted scattering such as wax, marble, skin, etc. [TODO] Surface translucency (BRDF+BTDF, BSSRDF) ![Figure [translucency]: Front-lit translucent object (left) and back-lit translucent object (right), using approximated BTDF and BSSRDF. Model: Lucy from the Stanford University Computer Graphics Laboratory](images/screenshot_translucency.png) ## Occlusion Occlusion is an important darkening factor used to recreate shadowing at various scales: Small scale : Micro-occlusion used to handle creases, cracks and cavities. Medium scale : Macro-occlusion used to handle occlusion by an object's own geometry or by geometry baked in normal maps (bricks, etc.). Large scale : Occlusion coming from contact between objects, or from an object's own geometry. We currently ignore micro-occlusion, which is often exposed in tools and engines under the form of a "cavity map". Sébastien Lagarde offers an interesting discussion in [#Lagarde14] on how micro-occlusion is handled in Frostbite: diffuse micro-occlusion is pre-baked in diffuse maps and specular micro-occlusion is pre-baked in reflectance textures. In our system, micro-occlusion can simply be baked in the base color map. This must be done knowing that the specular light will not be affected by micro-occlusion. Medium scale ambient occlusion is pre-baked in ambient occlusion maps, exposed as a material parameter, as seen in the material parameterization section earlier. Large scale ambient occlusion is often computed using screen-space techniques such as *SSAO* (screen-space ambient occlusion), *HBAO* (horizon based ambient occlusion), etc. Note that these techniques can also contribute to medium scale ambient occlusion when the camera is close enough to surfaces. **Note**: to prevent over darkening when using both medium and large scale occlusion, Lagarde recommends to use $min({AO}_{medium}, {AO}_{large})$. ### Diffuse occlusion Morgan McGuire formalizes ambient occlusion in the context of physically based rendering in [#McGuire10]. In his formulation, McGuire defines an ambient illumination function $L_a$, which in our case is encoded with spherical harmonics. He also defines a visibility function $V$, with $V(l)=1$ if there is an unoccluded line of sight from the surface in direction $l$, and 0 otherwise. With these two functions, the ambient term of the rendering equation can be expressed as shown in equation $\ref{diffuseAO}$. $$$$\label{diffuseAO} L(l,v) = \int_{\Omega} f(l,v) L_a(l) V(l) \left< \NoL \right> dl$$$$ This expression can be approximated by separating the visibility term from the illumination function, as shown in equation $\ref{diffuseAOApprox}$. $$$$\label{diffuseAOApprox} L(l,v) \approx \left( \pi \int_{\Omega} f(l,v) L_a(l) dl \right) \left( \frac{1}{\pi} \int_{\Omega} V(l) \left< \NoL \right> dl \right)$$$$ This approximation is only exact when the distant light $L_a$ is constant and $f$ is a Lambertian term. McGuire states however that this approximation is reasonable if both functions are relatively smooth over most of the sphere. This happens to be the case with a distant light probe (IBL). The left term of this approximation is the pre-computed diffuse component of our IBL. The right term is a scalar factor between 0 and 1 that indicates the fractional accessibility of a point. Its opposite is the diffuse ambient occlusion term, show in equation $\ref{diffuseAOTerm}$. $$$$\label{diffuseAOTerm} {AO} = 1 - \frac{1}{\pi} \int_{\Omega} V(l) \left< \NoL \right> dl$$$$ Since we use a pre-computed diffuse term, we cannot compute the exact accessibility of shaded points at runtime. To compensate for this lack of information in our precomputed term, we partially reconstruct incident lighting by applying an ambient occlusion factor specific to the surface's material at the shaded point. In practice, baked ambient occlusion is stored as a grayscale texture which can often be lower resolution than other textures (base color or normals for instance). It is important to note that the ambient occlusion property of our material model intends to recreate macro-level diffuse ambient occlusion. While this approximation is not physically correct, it constitutes an acceptable tradeoff of quality vs performance. Figure [aoComparison] shows two different materials without and with diffuse ambient occlusion. Notice how the material ambient occlusion is used to recreate the natural shadowing that occurs between the different tiles. Without ambient occlusion, both materials appear too flat. ![Figure [aoComparison]: Comparison of materials without diffuse ambient occlusion (left) and with (right)](images/screenshot_ao.jpg) Applying baked diffuse ambient occlusion in a GLSL shader is straightforward, as shown in listing [bakedDiffuseAO]. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ // diffuse indirect vec3 indirectDiffuse = max(irradianceSH(n), 0.0) * Fd_Lambert(); // ambient occlusion indirectDiffuse *= texture2D(aoMap, outUV).r; ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ [Listing [bakedDiffuseAO]: Implementation of baked diffuse ambient occlusion in GLSL] Note how the ambient occlusion term is only applied to indirect lighting. ### Specular occlusion Specular micro-occlusion can be derived from $\fNormal$, itself derived from the diffuse color. The derivation is based on the knowledge that no real-world material has a reflectance lower than 2%. Values in the 0-2% range can therefore be treated as pre-baked specular occlusion used to smoothly extinguish the Fresnel term. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ float f90 = clamp(dot(f0, 50.0 * 0.33), 0.0, 1.0); // cheap luminance approximation float f90 = clamp(50.0 * f0.g, 0.0, 1.0); ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ [Listing [specularMicroOcclusion]: Pre-baked specular occlusion in GLSL] The derivations mentioned earlier for ambient occlusion assume Lambertian surfaces and are only valid for indirect diffuse lighting. The lack of information about surface accessibility is particularly harmful to the reconstruction of indirect specular lighting. It usually manifests itself as light leaks. Sébastien Lagarde proposes an empirical approach to derive the specular occlusion term from the diffuse occlusion term in [#Lagarde14]. The result does not have any physical basis but produces visually pleasant results. The goal of his formulation is return the diffuse occlusion term unmodified for rough surfaces. For smooth surfaces, the formulation, implemented in listing [specularOcclusion], reduces the influence of occlusion at normal incidence and increases it at grazing angles. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ float computeSpecularAO(float NoV, float ao, float roughness) { return clamp(pow(NoV + ao, exp2(-16.0 * roughness - 1.0)) - 1.0 + ao, 0.0, 1.0); } // specular indirect vec3 indirectSpecular = evaluateSpecularIBL(r, perceptualRoughness); // ambient occlusion float ao = texture2D(aoMap, outUV).r; indirectSpecular *= computeSpecularAO(NoV, ao, roughness); ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ [Listing [specularOcclusion]: Implementation of Lagarde's specular occlusion factor in GLSL] Note how the specular occlusion factor is only applied to indirect lighting. #### Horizon specular occlusion When computing the specular IBL contribution for a surface that uses a normal map, it is possible to end up with a reflection vector pointing towards the surface. If this reflection vector is used for shading directly, the surface will be lit in places where it should not be lit (assuming opaque surfaces). This is another occurrence of light leaking that can easily be minimized using a simple technique described by Jeff Russell [#Russell15]. The key idea is to occlude light coming from behind the surface. This can easily be achieved since a negative dot product between the reflected vector and the surface's normal indicates a reflection vector pointing towards the surface. Our implementation shown in listing [horizonOcclusion] is similar to Russell's, albeit without the artist controlled horizon fading factor. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ // specular indirect vec3 indirectSpecular = evaluateSpecularIBL(r, perceptualRoughness); // horizon occlusion with falloff, should be computed for direct specular too float horizon = min(1.0 + dot(r, n), 1.0); indirectSpecular *= horizon * horizon; ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ [Listing [horizonOcclusion]: Implementation of horizon specular occlusion in GLSL] Horizon specular occlusion fading is cheap but can easily be omitted to improve performance as needed. ## Normal mapping There are two common use cases of normal maps: replacing high-poly meshes with low-poly meshes (using a base map) and adding surface details (using a detail map). Let's imagine that we want to render a piece of furniture covered in tufted leather. Modeling the geometry to accurately represent the tufted pattern would require too many triangles so we instead bake a high-poly mesh into a normal map. Once the base map is applied to a simplified mesh (in this case, a quad), we get the result in figure [normalMapped]. The base map used to create this effect is shown in figure [baseNormalMap]. ![Figure [normalMapped]: Low-poly mesh without normal mapping (left) and with (right)](images/screenshot_normal_mapping.jpg) ![Figure [baseNormalMap]: Normal map used as a base map](images/screenshot_normal_map.jpg) A simple problem arises if we now want to combine this base map with a second normal map. For instance, let's use the detail map shown in figure [detailNormalMap] to add cracks in the leather. ![Figure [detailNormalMap]: Normal map used as a detail map](images/screenshot_normal_map_detail.jpg) Given the nature of normal maps (XYZ components stored in tangent space), it is fairly obvious that naive approaches such as linear or overlay blending cannot work. We will use two more advanced techniques: a mathematically correct one and an approximation suitable for real-time shading. ### Reoriented normal mapping Colin Barré-Brisebois and Stephen Hill propose in [#Hill12] a mathematically sound solution called *Reoriented Normal Mapping*, which consists in rotating the basis of the detail map onto the normal from the base map. This technique relies on the shortest arc quaternion to apply the rotation, which greatly simplifies thanks to the properties of the tangent space. Following the simplifications described in [#Hill12], we can produce the GLSL implementation shown in listing [reorientedNormalMapping]. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ vec3 t = texture(baseMap, uv).xyz * vec3( 2.0, 2.0, 2.0) + vec3(-1.0, -1.0, 0.0); vec3 u = texture(detailMap, uv).xyz * vec3(-2.0, -2.0, 2.0) + vec3( 1.0, 1.0, -1.0); vec3 r = normalize(t * dot(t, u) - u * t.z); return r; ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ [Listing [reorientedNormalMapping]: Implementation of reoriented normal mapping in GLSL] Note that this implementation assumes that the normals are stored uncompressed and in the [0..1] range in the source textures. The normalization step is not strictly necessary and can be skipped if the technique is used at runtime. If so, the computation of r becomes t * dot(t, u) / t.z - u. Since this technique is slightly more expensive than the one described below, we will mostly use it offline. We therefore provide a simple offline tool to combine two normal maps. Figure [blendedNormalMaps] presents the output of the tool with the base map and the detail map shown previously. ![Figure [blendedNormalMaps]: Blended normal and detail map (left) and resulting render when combined with a diffuse map (right)](images/screenshot_normal_map_blended.jpg) ### UDN blending The technique called UDN blending, described in [#Hill12], is a variant of the partial derivative blending technique. Its main advantage is the low number of shader instructions it requires (see listing [udnBlending]). While it leads to a reduction in details over flat areas, UDN blending is interesting if blending must be performed at runtime. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ vec3 t = texture(baseMap, uv).xyz * 2.0 - 1.0; vec3 u = texture(detailMap, uv).xyz * 2.0 - 1.0; vec3 r = normalize(t.xy + u.xy, t.z); return r; ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ [Listing [udnBlending]: Implementation of UDN blending in GLSL] The results are visually close to Reoriented Normal Mapping but a careful comparison of the data shows that UDN is indeed less correct. Figure [blendedNormalMapsUDN] presents the result of the UDN blending approach using the same source data as in the previous examples. ![Figure [blendedNormalMapsUDN]: Blended normal and detail map using the UDN blending technique](images/screenshot_normal_map_blended_udn.jpg) # Volumetric effects ## Exponential height fog ![Figure [exponentialHeightFog1]: Example of directional in-scattering with exponential height fog](images/screenshot_fog1.jpg) ![Figure [exponentialHeightFog2]: Example of directional in-scattering with exponential height fog](images/screenshot_fog2.jpg) # Anti-aliasing [TODO] MSAA, geometric AA (normals and roughness), shader anti-aliasing (object-space shading?) # Imaging pipeline The lighting section of this document describes how light interacts with surfaces in the scene in a physically based manner. To achieve plausible results, we must go a step further and consider the transformations necessary to convert the scene luminance, as computed by our lighting equations, into displayable pixel values. The series of transformations we are going to use form the following imaging pipeline: ************************************************************************************* * .-------------. .--------------. .---------------. * * | Scene | | Normalized | | | * * | luminance +----->| luminance +----->| White balance | * * | | | (HDR) | | | * * '-------------' '--------------' '-------+-------' * * | * * v * * .---------------. * * | | * * | Color grading | * * | | * * '-------+-------' * * | * * v * * .---------------. * * | | * * | Tone mapping | * * | | * * '-------+-------' * * | * * v * * .---------------. .-------------. * * | | | Pixel | * * | OETF +----->| value | * * | | | (LDR) | * * '---------------' '-------------' * ************************************************************************************* **Note**: the *OETF* step is the application of the opto-electronic transfer function of the target color space. For clarity this diagram does not include post-processing steps such as vignette, bloom, etc. These effects will be discussed separately. [TODO] Color spaces (ACES, sRGB, Rec. 709, Rec. 2020, etc.), gamma/linear, etc. ## Physically based camera The first step in the image transformation process is to use a physically based camera to properly expose the scene's outgoing luminance. ### Exposure settings Because we use photometric units throughout the lighting pipeline, the light reaching the camera is an energy expressed in luminance $L$, in $cd.m^{-2}$. Light incident to the camera sensor can cover a large range of values, from $10^{-5}cd.m^{-2}$ for starlight to $10^{9}cd.m^{-2}$ for the sun. Since we obviously cannot manipulate and even less record such a large range of values, we need to remap them. This range remapping is done in a camera by exposing the sensor for a certain time. To maximize the use of the limited range of the sensor, the scene's light range is centered around the "middle gray", a value halfway between black and white. The exposition is therefore achieved by manipulating, either manually or automatically, 3 settings: - Aperture - Shutter speed - Sensitivity (also called gain) Aperture : Noted $N$ and expressed in f-stops ƒ, this setting controls how open or closed the camera system's aperture is. Since an f-stop indicate the ratio of the lens' focal length to the diameter of the entrance pupil, high-values (ƒ/16) indicate a small aperture and small values (ƒ/1.4) indicate a wide aperture. In addition to the exposition, the aperture setting controls the depth of field. Shutter speed : Noted $t$ and expressed in seconds $s$, this setting controls how long the aperture remains opened (it also controls the timing of the sensor shutter(s), whether electronic or mechanical). In addition to the exposition, the shutter speed controls motion blur. Sensitivity : Noted $S$ and expressed in ISO, this setting controls how the light reaching the sensor is quantized. Because of its unit, this setting is often referred to as simply the "ISO" or "ISO setting". In addition to the exposition, the sensitivity setting controls the amount of noise. ### Exposure value Since referring to these 3 settings in our equations would be unwieldy, we instead summarize the “exposure triangle” by an exposure value, noted EV[^reciprocity]. The EV is expressed in a base-2 logarithmic scale, with a difference of 1 EV called a stop. One positive stop (+1 EV) corresponds to a factor of two in luminance and one negative stop (-1 EV) corresponds to a factor of half in luminance. Equation $\ref{ev}$ shows the [formal definition of EV](https://en.wikipedia.org/wiki/Exposure_value). $$$$\label{ev} EV = log_2(\frac{N^2}{t})$$$$ Note that this definition is only function of the aperture and shutter speed, but not the sensitivity. An exposure value is by convention defined for ISO 100, or $EV_{100}$, and because we wish to work with this convention, we need to be able to express $EV_{100}$ as a function of the sensitivity. Since we know that EV is a base-2 logarithmic scale in which each stop increases or decreases the brightness by a factor of 2, we can formally define $EV_{S}$, the exposure value at given sensitivity (equation $\ref{evS}$). $$$$\label{evS} {EV}_S = EV_{100} + log_2(\frac{S}{100})$$$$ Calculating the $EV_{100}$ as a function of the 3 camera settings is trivial, as shown in $\ref{ev100}$. $$$$\label{ev100} {EV}_{100} = EV_{S} - log_2(\frac{S}{100}) = log_2(\frac{N^2}{t}) - log_2(\frac{S}{100})$$$$ Note that the operator (photographer, etc.) can achieve the same exposure (and therefore EV) with several combinations of aperture, shutter speed and sensitivity. This allows some artistic control in the process (depth of field vs motion blur vs grain). [^reciprocity]: We assume a digital sensor, which means we don't need to take reciprocity failure into account #### Exposure value and luminance A camera, similar to a spot meter, is able to measure the average luminance of a scene and convert it into EV to achieve automatic exposure, or at the very least offer the user exposure guidance. It is possible to define EV as a function of the scene luminance $L$, given a per-device calibration constant $K$ (equation $\ref{evK}$). $$$$\label{evK} EV = log_2(\frac{L \times S}{K})$$$$ That constant $K$ is the reflected-light meter constant, which varies between manufacturers. We could find two common values for this constant: 12.5, used by Canon, Nikon and Sekonic, and 14, used by Pentax and Minolta. Given the wide availability of Canon and Nikon cameras, as well as our own usage of Sekonic light meters, we will choose to use $K = 12.5$. Since we want to work with $EV_{100}$, we can substitute $K$ and $S$ in equation $\ref{evK}$ to obtain equation $\ref{ev100L}$. $$$$\label{ev100L} EV = log_2(L \frac{100}{12.5})$$$$ Given this relationship, it would be possible to implement automatic exposure in our engine by first measuring the average luminance of a frame. An easy way to achieve this is to simply downsample a luminance buffer down to 1 pixel and read the remaining value. This technique is unfortunately rarely stable and can easily be affected by extreme values. Many games use a different approach which consists in using a luminance histogram to remove extreme values. For validation and testing purposes, the luminance can be computed from a given EV: $$$$L = 2^{EV_{100}} \times \frac{12.5}{100} = 2^{EV_{100} - 3}$$$$ #### Exposure value and illuminance It is possible to define EV as a function of the illuminance $E$, given a per-device calibration constant $C$: $$$$\label{evC} EV = log_2(\frac{E \times S}{C})$$$$ The constant $C$ is the incident-light meter constant, which varies between manufacturers and/or types of sensors. There are two common types of sensors: flat and hemispherical. For flat sensors, a common value is 250. With hemispherical sensors, we could find two common values: 320, used by Minolta, and 340, used by Sekonic. Since we want to work with $EV_{100}$, we can substitute $S$ $\ref{evC}$ to obtain equation $\ref{ev100C}$. $$$$\label{ev100C} EV = log_2(E \frac{100}{C})$$$$ The illuminance can then be computed from a given EV. For a flat sensor with $C = 250$ we obtain equation $\ref{eFlatSensor}$. $$$$\label{eFlatSensor} E = 2^{EV_{100}} \times 2.5$$$$ For a hemispherical sensor with $C = 340$ we obtain equation $\ref{eHemisphereSensor}$ $$$$\label{eHemisphereSensor} E = 2^{EV_{100}} \times 3.4$$$$ #### Exposure compensation Even though an exposure value actually indicates combinations of camera settings, it is often used by photographers to describe light intensity. This is why cameras let photographers apply an exposure compensation to over or under-expose an image. This setting can be used for artistic control but also to achieve proper exposure (snow for instance will be exposed for as 18% middle-gray). Applying an exposure compensation $EC$ is a simple as adding an offset to the exposure value, as shown in equation $\ref{ec}$. $$$$\label{ec} EV_{100}' = EV_{100} - EC$$$$ This equation uses a negative sign because we are using $EC$ in f-stops to adjust the final exposure. Increasing the EV is akin to closing down the aperture of the lens (or reducing shutter speed or reducing sensitivity). A higher EV will produce darker images. ### Exposure To convert the scene luminance into normalized luminance, we must use the [photometric exposure](https://en.wikipedia.org/wiki/Exposure_value#Camera_settings_vs._photometric_exposure) (or luminous exposure), or amount of scene luminance that reaches the camera sensor. The photometric exposure, expressed in lux seconds and noted $H$, is given by equation $\ref{photometricExposure}$. $$$$\label{photometricExposure} H = \frac{q \cdot t}{N^2} L$$$$ Where $L$ is the luminance of the scene, $t$ the shutter speed, $N$ the aperture and $q$ the lens and vignetting attenuation (typically $q = 0.65$[^lensAttenuation]). This definition does not take the sensor sensitivity into account. To do so, we must use one of the three ways to relate photometric exposure and sensitivity: saturation-based speed, noise-based speed and standard output sensitivity. We choose the saturation-based speed relation, which gives us $H_{sat}$, the maximum possible exposure that does not lead to clipped or bloomed camera output (equation $\ref{hSat}$). $$$$\label{hSat} H_{sat} = \frac{78}{S_{sat}}$$$$ We combine equations $\ref{hSat}$ and $\ref{photometricExposure}$ in equation $\ref{lmax}$ to compute the maximum luminance $L_{max}$ that will saturate the sensor given exposure settings $S$, $N$ and $t$. $$$$\label{lmax} L_{max} = \frac{N^2}{q \cdot t} \frac{78}{S}$$$$ This maximum luminance can then be used to normalize incident luminance $L$ as shown in equation $\ref{normalizedLuminance}$. $$$$\label{normalizedLuminance} L' = L \frac{1}{L_{max}}$$$$ $L_{max}$ can be simplified using equation $\ref{ev}$, $S = 100$ and $q = 0.65$: \begin{align*} L_{max} &= \frac{N^2}{t} \frac{78}{q \cdot S} \\ L_{max} &= 2^{EV_{100}} \frac{78}{q \cdot S} \\ L_{max} &= 2^{EV_{100}} \times 1.2 \end{align*} Listing [fragmentExposure] shows how the exposure term can be applied directly to the pixel color computed in a fragment shader. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ // Computes the camera's EV100 from exposure settings // aperture in f-stops // shutterSpeed in seconds // sensitivity in ISO float exposureSettings(float aperture, float shutterSpeed, float sensitivity) { return log2((aperture * aperture) / shutterSpeed * 100.0 / sensitivity); } // Computes the exposure normalization factor from // the camera's EV100 float exposure(ev100) { return 1.0 / (pow(2.0, ev100) * 1.2); } float ev100 = exposureSettings(aperture, shutterSpeed, sensitivity); float exposure = exposure(ev100); vec4 color = evaluateLighting(); color.rgb *= exposure; ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ [Listing [fragmentExposure]: Implementation of exposure in GLSL] In practice the exposure factor can be pre-computed on the CPU to save shader instructions. [^lensAttenuation]: See *Film Speed, Measurements and calculations* on Wikipedia (https://en.wikipedia.org/wiki/Film_speed) ### Automatic exposure The process described above relies on artists setting the camera exposure settings manually. This can prove cumbersome in practice since camera movements and/or dynamic effects can greatly affect the scene's luminance. Since we know how to compute the exposure value from a given luminance (see section [Exposure value and luminance]), we can transform our camera into a spot meter. To do so, we need to measure the scene's luminance. There are two common techniques used to measure the scene's luminance: - **Luminance downsampling**, by downsampling the previous frame successively until obtaining a 1x1 log luminance buffer that can be read on the CPU (this could also be achieved using a compute shader). The result is the average log luminance of the scene. The first downsampling must extract the luminance of each pixel first. This technique can be unstable and its output should be smoothed over time. - **Using a luminance histogram**, to find the average log luminance. This technique has an advantage over the previous one as it allows to ignore extreme values and offers more stable results. Note that both methods will find the average luminance after multiplication by the albedo. This is not entirely correct but the alternative is to keep a luminance buffer that contains the luminance of each pixel before multiplication by the surface albedo. This is expensive both computationally and memory-wise. These two techniques also limit the metering system to average metering, where each pixel has the same influence (or weight) over the final exposure. Cameras typically offer 3 modes of metering: Spot metering : In which only a small circle in the center of the image contributes to the final exposure. That circle is usually 1 to 5% of the total image size. Center-weighted metering : Gives more influence to scene luminance values located in the center of the screen. Multi-zone or matrix metering : A metering mode that differs for each manufacturer. The goal of this mode is to prioritize exposure for the most important parts of the scene. This is often achieved by splitting the image into a grid and by classifying each cell (using focus information, min/max luminance, etc.). Advanced implementations attempt to compare the scene to a known dataset to achieve proper exposure (backlit sunset, overcast snowy day, etc.). #### Spot metering The weight $w$ of each luminance value to use when computing the scene luminance is given by equation $\ref{spotMetering}$. $$$$\label{spotMetering} w(x,y) = \begin{cases} 1 & \left| p_{x,y} - s_{x,y} \right| \le s_r \\ 0 & \left| p_{x,y} - s_{x,y} \right| \gt s_r \end{cases}$$$$ Where $p$ is the position of the pixel, $s$ the center of the spot and $s_r$ the radius of the spot. #### Center-weighted metering $$$$\label{centerMetering} w(x,y) = smooth(\left| p_{x,y} - c \right| \times \frac{2}{width} )$$$$ Where $c$ is the center of the time and $smooth()$ a smoothing function such as GLSL's smoothstep(). #### Adaptation To smooth the result of the metering, we can use equation $\ref{adaptation}$, an exponential feedback loop as described by Pattanaik et al. in [Pattanaik00]. $$$$\label{adaptation} L_{avg} = L_{avg} + (L - L_{avg}) \times (1 - e^{-\Delta t \cdot \tau})$$$$ Where $\Delta t$ is the delta time from the previous frame and $\tau$ a constant that controls the adaptation rate. ### Bloom Because the EV scale is almost perceptually linear, the exposure value is also often used as a light unit. This means we could let artists specify the intensity of lights or emissive surfaces using exposure compensation as a unit. The intensity of emitted light would therefore be relative to the exposure settings. Using exposure compensation as a light unit should be avoided whenever possible but can be useful to force (or cancel) a bloom effect around emissive surfaces independently of the camera settings (for instance, a lightsaber in a game should always bloom). ![Figure [bloom]: Saturated photosites on a sensor create a blooming effect in the bright parts of the scene](images/screenshot_bloom.jpg) With $c$ the bloom color and $EV_{100}$ the current exposure value, we can easily compute the luminance of the bloom value as show in equation $\ref{bloomEV}$. $$$$\label{bloomEV} EV_{bloom} = EV_{100} + EC \\ L_{bloom} = c \times 2^{EV_{bloom} - 3}$$$$ Equation $\ref{bloomEV}$ can be used in a fragment shader to implement emissive blooms, as shown in listing [fragmentEmissive]. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ vec4 surfaceShading() { vec4 color = evaluateLights(); // rgb = color, w = exposure compensation vec4 emissive = getEmissive(); color.rgb += emissive.rgb * pow(2.0, ev100 + emissive.w - 3.0); color.rgb *= exposure; return color; } ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ [Listing [fragmentEmissive]: Implementation of emissive bloom in GLSL] ## Optics post-processing ### Color fringing [TODO] ![Figure [fringing]: Example of color fringing: look at the ear on the left or the chin at the bottom.](images/screenshot_fringing.jpg) ### Lens flares [TODO] Notes: there is a physically based approach to generating lens flares, by tracing rays through the optical assembly of the lens, but we are going to use an image-based approach. This approach is cheaper and has a few welcome benefits such as free emitters occlusion and unlimited light sources support. ## Filmic post-processing [TODO] Perform post-processing on the scene referred data (linear space, before tone-mapping) as much as possible It is important to provide color correction tools to give artists greater artistic control over the final image. These tools are found in every photo or video processing application, such as Adobe Photoshop or Adobe After Effects. ### Contrast ### Curves ### Levels ### Color grading ## Light path The light path, or rendering method, used by the engine can have serious performance implications and may impose strong limitations on how many lights can be used in a scene. There are traditionally two different rendering methods used by 3D engines forward and deferred rendering. Our goal is to use a rendering method that obeys the following constraints: - Low bandwidth requirements - Multiple dynamic lights per pixel Additionally, we would like to easily support: - MSAA - Transparency - Multiple material models Deferred rendering is used by many modern 3D rendering engines to easily support dozens, hundreds or even thousands of light source (amongst other benefits). This method is unfortunately very expensive in terms of bandwidth. With our default PBR material model, our G-buffer would use between 160 and 192 bits per pixel, which would translate directly to rather high bandwidth requirements. Forward rendering methods on the other hand have historically been bad at handling multiple lights. A common implementation is to render the scene multiple times, once per visible light, and to blend (add) the results. Another technique consists in assigning a fixed maximum of lights to each object in the scene. This is however impractical when objects occupy a vast amount of space in the world (building, road, etc.). Tiled shading can be applied to both forward and deferred rendering methods. The idea is to split the screen in a grid of tiles and for each tile, find the list of lights that affect the pixels within that tile. This has the advantage of reducing overdraw (in deferred rendering) and shading computations of large objects (in forward rendering). This technique suffers however from depth discontinuities issues that can lead to large amounts of extraneous work. The scene displayed in figure [sponza] was rendered using clustered forward rendering. ![Figure [sponza]: Clustered forward rendering with dozens of dynamic lights and MSAA](images/screenshot_sponza.jpg) Figure [sponzaTiles] shows the same scene split in tiles (in this case, a 1280x720 render target with 80x80px tiles). ![Figure [sponzaTiles]: Tiled shading (16x9 tiles)](images/screenshot_sponza_tiles.jpg) ### Clustered Forward Rendering We decided to explore another method called Clustered Shading, in its forward variant. Clustered shading expands on the idea of tiled rendering but adds a segmentation on the 3rd axis. The “clustering” is done in view space, by splitting the frustum into a 3D grid. The frustum is first sliced on the depth axis as show in figure [sponzaSlices]. ![Figure [sponzaSlices]: Depth slicing (16 slices)](images/screenshot_sponza_slices.jpg) And the depth slices are then combined with the screen tiles to "voxelize" the frustum. We call each cluster a froxel as it makes it clear what they represent (a voxel in frustum space). The result of the "froxelization" pass is shown in figure [froxel1] and figure [froxel2]. ![Figure [froxel1]: Frustum voxelization (5x3 tiles, 8 depth slices)](images/screenshot_sponza_froxels1.jpg) ![Figure [froxel2]: Frustum voxelization (5x3 tiles, 8 depth slices)](images/screenshot_sponza_froxels2.jpg) Before rendering a frame, each light in the scene is assigned to any froxel it intersects with. The result of the lights assignment pass is a list of lights for each froxel. During the rendering pass, we can compute the ID of the froxel a fragment belongs to and therefore the list of lights that can affect that fragment. The depth slicing is not linear, but exponential. In a typical scene, there will be more pixels close to the near plane than to the far plane. An exponential grid of froxels will therefore improve the assignment of lights where it matters the most. Figure [froxelDistribution] shows how much world space unit each depth slice uses with exponential slicing. ![Figure [froxelDistribution]: Near: 0.1m, Far: 100m, 16 slices](images/diagram_froxels1.png) A simple exponential voxelization is unfortunately not enough. The graphic above clearly illustrates how world space is distributed across slices but it fails to show what happens close to the near plane. If we examine the same distribution in a smaller range (0.1m to 7m) we can see an interesting problem appear as shown in figure [froxelDistributionClose]. ![Figure [froxelDistributionClose]: Depth distribution in the 0.1-7m range](images/diagram_froxels2.png) This graphic shows that a simple exponential distribution uses up half of the slices very close to the camera. In this particular case, we use 8 slices out of 16in the first 5 meters. Since dynamic world lights are either point lights (spheres) or spot lights (cones), such a fine resolution is completely unnecessary so close to the near plane. Our solution is to manually tweak the size of the first froxel depending on the scene and the near and far planes. By doing so, we can better distribute the remaining froxels across the frustum. Figure [froxelDistributionExp] shows for instance what happens when we use a special froxel between 0.1m and 5m. ![Figure [froxelDistributionExp]: Near: 0.1, Far: 100m, 16 slices, Special froxel: 0.1-5m](images/diagram_froxels3.png) This new distribution is much more efficient and allows a better assignment of the lights throughout the entire frustum. ### Implementation notes Lights assignment can be done in two different ways, on the GPU or on the CPU. #### GPU lights assignment This implementation requires OpenGL ES 3.1 and support for compute shaders. The lights are stored in Shader Storage Buffer Objects (SSBO) and passed to a compute shader that assigns each light to the corresponding froxels. The frustum voxelization can be executed only once by a first compute shader (as long as the projection matrix does not change), and the lights assignment can be performed each frame by another compute shader. The threading model of compute shaders is particularly well suited for this task. We simply invoke as many workgroups as we have froxels (we can directly map the X, Y and Z workgroup counts to our froxel grid resolution). Each workground will in turn be threaded and traverse all the lights to assign. Intersection tests imply simple sphere/frustum or cone/frustum tests. See the annex for the source code of a GPU implementation (point lights only). #### CPU lights assignment On non-OpenGL ES 3.1 devices, lights assignment can be performed efficiently on the CPU. The algorithm is different from the GPU implementation. Instead of iterating over every light for each froxel, the engine will “rasterize” each light as froxels. For instance, given a point light’s center and radius, it is trivial to compute the list of froxels it intersects with. This technique has the added benefit of providing tighter culling than in the GPU variant. The CPU implementation can also more easily generate a packed list of lights. #### Shading The list of lights per froxel can be passed to the fragment shader either as an SSBO (OpenGL ES 3.1) or a texture. #### From depth to froxel Given a near plane $n$, a far plane $f$, a maximum number of depth slices $m$ and a linear depth value $z$ in the range [0..1], equation $\ref{zToCluster}$ can be used to compute the index of the cluster for a given position. $$$$\label{zToCluster} zToCluster(z,n,f,m)=floor \left( max \left( log2(z) \frac{m}{-log2(\frac{n}{f})} + m, 0 \right) \right)$$$$ This formula suffers however from the resolution issue mentioned previously. We can fix it by introducing $sn$, a special near value that defines the extent of the first froxel (the first froxel occupies the range [n..sn], the remaining froxels [sn..f]). $$$$\label{zToClusterFix} zToCluster(z,n,sn,f,m)=floor \left( max \left( log2(z) \frac{m-1}{-log2(\frac{sn}{f})} + m, 0 \right) \right)$$$$ Equation $\ref{linearZ}$ can be used to compute a linear depth value from gl_FragCoord.z (assuming a standard OpenGL projection matrix). $$$$\label{linearZ} linearZ(z)=\frac{n}{f+z(n-f)}$$$$ This equation can be simplified by pre-computing two terms $c0$ and $c1$, as shown in equation $\ref{linearZFix}$. $$$$\label{linearZFix} c1 = \frac{f}{n} \\ c0 = 1 - c1 \\ linearZ(z)=\frac{1}{z \cdot c0 + c1}$$$$ This simplification is important because we pass the linear z value to a log2 in $\ref{zToClusterFix}$. Since the division becomes a negation under a logarithmic, we can avoid a division by using $-log2(z \cdot c0 + c1)$ instead. All put together, computing the froxel index of a given fragment can be implemented fairly easily as shown in listing [fragCoordToFroxel]. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ #define MAX_LIGHT_COUNT 16 // max number of lights per froxel uniform uvec4 froxels; // res x, res y, count y, count y uniform vec4 zParams; // c0, c1, index scale, index bias uint getDepthSlice() { return uint(max(0.0, log2(zParams.x * gl_FragCoord.z + zParams.y) * zParams.z + zParams.w)); } uint getFroxelOffset(uint depthSlice) { uvec2 froxelCoord = uvec2(gl_FragCoord.xy) / froxels.xy; froxelCoord.y = (froxels.w - 1u) - froxelCoord.y; uint index = froxelCoord.x + froxelCoord.y * froxels.z + depthSlice * froxels.z * froxels.w; return index * MAX_FROXEL_LIGHT_COUNT; } uint slice = getDepthSlice(); uint offset = getFroxelOffset(slice); // Compute lighting... ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ [Listing [fragCoordToFroxel]: GLSL implementation to compute a froxel index from a fragment's screen coordinates] Several uniforms must be pre-computed for perform the index evaluation efficiently. The code used to pre-compute these uniforms can be found in listing [froxelIndexPrecomputation]. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ froxels[0] = TILE_RESOLUTION_IN_PX; froxels[1] = TILE_RESOLUTION_IN_PX; froxels[2] = numberOfTilesInX; froxels[3] = numberOfTilesInY; zParams[0] = 1.0f - Z_FAR / Z_NEAR; zParams[1] = Z_FAR / Z_NEAR; zParams[2] = (MAX_DEPTH_SLICES - 1) / log2(Z_SPECIAL_NEAR / Z_FAR); zParams[3] = MAX_DEPTH_SLICES; ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ [Listing [froxelIndexPrecomputation]] #### From froxel to depth Given a froxel index $i$, a special near plane $sn$, a far plane $f$ and a maximum number of depth slices $m$, equation $\ref{clusterToZ}$ computes the minimum depth of a given froxel. $$$$\label{clusterToZ} clusterToZ(i \ge 1,sn,f,m)=2^{(i-m) \frac{-log2(\frac{sn}{f})}{m-1}}$$$$ For $i=0$, the z value is 0. The result of this equation is in the [0..1] range and should be multiplied by $f$ to get a distance in world units. The compute shader implementation should use exp2 instead of a pow. The division can be precomputed and passed as a uniform. ## Validation Given the complexity of our lighting system, it is important to validate our implementation. We will do so in several ways: using reference renderings, light measurements and data visualization. [TODO] Explain light measurement validation (reading EV from the render target and comparing against values measure with light meters/cameras, etc.) ### Scene referred visualization A quick and easy way to validate a scene's lighting is to modify the shader to output colors that provide an intuitive mapping to relevant data. This can easily be done by using a custom debug tone-mapping operator that outputs fake colors. #### Luminance stops With emissive materials and IBLs, it is fairly easy to obtain a scene in which specular highlights are brighter than their apparent caster. This type of issue can be difficult to observe after tone-mapping and quantization but is fairly obvious in the scene-referred space. Figure [luminanceViz] shows how the custom operator described in listing [tonemapLuminanceViz] is used to show the exposed luminance of a scene. ![Figure [luminanceViz]: Visualizing luminance by color coding the stops: cyan is middle gray, blue is 1 stop darker, green 1 stop brighter, etc.](images/screenshot_luminance_debug.png) ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ vec3 Tonemap_DisplayRange(const vec3 x) { // The 5th color in the array (cyan) represents middle gray (18%) // Every stop above or below middle gray causes a color shift float v = log2(luminance(x) / 0.18); v = clamp(v + 5.0, 0.0, 15.0); int index = int(floor(v)); return mix(debugColors[index], debugColors[min(15, index + 1)], fract(v)); } const vec3 debugColors[16] = vec3[]( vec3(0.0, 0.0, 0.0), // black vec3(0.0, 0.0, 0.1647), // darkest blue vec3(0.0, 0.0, 0.3647), // darker blue vec3(0.0, 0.0, 0.6647), // dark blue vec3(0.0, 0.0, 0.9647), // blue vec3(0.0, 0.9255, 0.9255), // cyan vec3(0.0, 0.5647, 0.0), // dark green vec3(0.0, 0.7843, 0.0), // green vec3(1.0, 1.0, 0.0), // yellow vec3(0.90588, 0.75294, 0.0), // yellow-orange vec3(1.0, 0.5647, 0.0), // orange vec3(1.0, 0.0, 0.0), // bright red vec3(0.8392, 0.0, 0.0), // red vec3(1.0, 0.0, 1.0), // magenta vec3(0.6, 0.3333, 0.7882), // purple vec3(1.0, 1.0, 1.0) // white ); ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ [Listing [tonemapLuminanceViz]: GLSL implementation of a custom debug tone-mapping operator for luminance visualization] ### Reference renderings To validate our implementation against reference renderings, we will use a commercial-grade Open Source physically based offline path tracer called Mitsuba. Mitsuba offers many different integrators, samplers and material models, which should allow us to provide fair comparisons with our real-time renderer. This path tracer also relies on a simple XML scene description format that should be easy to automatically generate from our own scene descriptions. Figure [mitsubaReference] and figure [filamentReference] show a simple scene, a perfectly smooth dielectric sphere, rendered respectively with Mitsuba and Filament. ![Figure [mitsubaReference]: Rendered in 2048x1440 in 1 minute and 42 seconds on a 12 core 2013 MacPro](images/screenshot_ref_mitsuba.jpg) ![Figure [filamentReference]: Rendered in 2048x1440 with MSAA 4x at 60 fps on a Nexus 9 device (Tegra K1 GPU)](images/screenshot_ref_filament.jpg) The parameters used to render both scenes are the following: **Filament** - Material - Base color: sRGB 0.81, 0, 0 - Metallic: 0 - Roughness: 0 - Reflectance: 0.5 - Indirect light: IBL - 256x256 cubemap generated by cmgen from office.exr - Multiplier: 35,000 - Direct light: directional light - Linear color: 1.0, 0.96, 0.95 - Intensity: 120,000 lux - Exposure - Aperture: f/16 - Shutter speed: 1/125s - ISO: 100 **Mitsuba** - BSDF: roughplastic - Distribution: GGX - Alpha: 0 - Diffuse reflectance: sRGB 0.81, 0, 0 - Emitter: environment map - Source: office.exr - Scale: 35,000 - Emitter: directional - Irradiance: linear RGB 120,000 115,200 114,000 - Film: LDR - Exposure: -15.23, computed from log2(filamentExposure) - Integrator: path - Sampler: ldsampler - Sample count: 256 The full Mitsuba scene can be found as an annex. Both scenes were rendered at the same resolution (2048x1440). #### Comparison The slight differences between the two renderings come from the various approximations used by Filament: RGBM 256x256 reflection probe, RGBM 1024x1024 background map, Lambert diffuse, split-sum approximation, analytical approximation of the DFG term, etc. Figure [referenceComparison] shows the luminance gradient of the images produced by both engines. The comparison was performed on LDR images. ![Figure [referenceComparison]: Luminance gradients from Mitsuba (left) and Filament (right)](images/screenshot_ref_comparison.png) The biggest difference is visible at grazing angles, which is most likely explained by Filament's use of a Lambertian diffuse term. The Disney diffuse term and its grazing retro-reflections would move Filament closer to Mitsuba. ## Coordinates systems ### World coordinates system Filament uses a Y-up, right-handed coordinate system. ![Figure [coordinates]: Red +X, green +Y, blue +Z (rendered in Marmoset Toolbag).](images/screenshot_coordinates.jpg) ### Camera coordinates system Filament's Camera looks towards its local -Z axis. That is, when placing a camera in the world without any transform applied to it, the camera looks down the world's -Z axis. ### Cubemaps coordinates system All cubemaps used in Filament follow the OpenGL convention for face alignment shown in figure [cubemapCoordinates]. ![Figure [cubemapCoordinates]: Horizontal cross representation of a cubemap following the OpenGL faces alignment convention.](images/screenshot_cubemap_coordinates.png) Note that environment background and reflection probes are mirrored (see section [Mirroring]). #### Mirroring To simplify the rendering of reflections, IBL cubemaps are stored mirrored on the X axis. This is the default behaviour of the cmgen tool. This means that an IBL cubemap used as environment background needs to be mirrored again at runtime. An easy way to achieve this for skyboxes is to use textured back faces. Filament does this by default. #### Equirectangular environment maps To convert equirectangular environment maps to horizontal/vertical cross cubemaps we position the +Z face in the center of the source rectilinear environment map. #### World space orientation of environment maps and Skyboxes When specifying a skybox or an IBL in Filament, the specified cubemap is oriented such that its -Z face points towards the +Z axis of the world (this is because filament assumes mirrored cubemaps, see section [Mirroring]). However, because environments and skyboxes are expected to be pre-mirrored, their -Z (back) face points towards the world's -Z axis as expected (and the camera looks toward that direction by default, see section [Camera coordinates system]). # Annex ## Specular color The specular color of a metallic surface, or $\fNormal$, can be computed directly from measured spectral data. Online databases such as [Refractive Index](https://refractiveindex.info/?shelf=3d&book=metals&page=brass) provide tables of complex IOR measured at different wavelengths for various materials. Earlier in this document, we presented equation $\ref{fresnelEquation}$ to compute the Fresnel reflectance at normal incidence for a dielectric surface given its IOR. The same equation can be rewritten for conductors by using complex numbers to represent the surface's IOR: $$$$c_{ior} = n_{ior} + ik$$$$ Equation $\ref{fresnelComplexIOR}$ presents the resulting Fresnel formula, where $c^*$ is the conjugate of the complex number $c$: $$$$\label{fresnelComplexIOR} \fNormal(c_{ior}) = \frac{(c_{ior} - 1)(c_{ior}^* - 1)}{(c_{ior} + 1)(c_{ior}^* + 1)}$$$$ To compute the specular color of a material we need to evaluate the complex Fresnel equation at each spectral sample of complex IOR over the visible spectrum. For each spectral sample, we obtain a spectral reflectance sample. To find the RGB color at normal incidence, we must multiply each sample by the CIE XYZ CMFs (color matching functions) and the spectral power distribution of the desired illuminant. We choose the standard illuminant D65 because we want to compute a color in the sRGB color space. We then sum (integrate) and normalize all the samples to obtain $\fNormal$ in the XYZ color space. From there, a simple color space conversion yields a linear sRGB color or a non-linear sRGB color after applying the opto-electronic transfer function (OETF, commonly known as "gamma" curve). Note that for some materials such as gold the final sRGB color might fall out of gamut. We use a simple normalization step as a cheap form of gamut remapping but it would be interesting to consider computing values in a color space with a wider gamut (for instance BT.2020). To achieve the desired result we used the ICE 1931 2 degrees CMFs, from 360nm to 830nm at 1nm intervals ([source](http://cvrl.ioo.ucl.ac.uk/cmfs.htm)), and the CIE Standard Illuminant D65 relative spectral power distribution, from 300nm to 830nm, at 5nm intervals ([source](http://files.cie.co.at/204.xls)). Our implementation is presented in listing [specularColorImpl], with the actual data omitted for brevity. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ // CIE 1931 2-deg color matching functions (CMFs), from 360nm to 830nm, // at 1nm intervals // // Data source: // http://cvrl.ioo.ucl.ac.uk/cmfs.htm // http://cvrl.ioo.ucl.ac.uk/database/text/cmfs/ciexyz31.htm const size_t CIE_XYZ_START = 360; const size_t CIE_XYZ_COUNT = 471; const float3 CIE_XYZ[CIE_XYZ_COUNT] = { ... }; // CIE Standard Illuminant D65 relative spectral power distribution, // from 300nm to 830, at 5nm intervals // // Data source: // https://en.wikipedia.org/wiki/Illuminant_D65 // https://cielab.xyz/pdf/CIE_sel_colorimetric_tables.xls const size_t CIE_D65_INTERVAL = 5; const size_t CIE_D65_START = 300; const size_t CIE_D65_END = 830; const size_t CIE_D65_COUNT = 107; const float CIE_D65[CIE_D65_COUNT] = { ... }; struct Sample { float w = 0.0f; // wavelength std::complex ior; // complex IOR, n + ik }; static float illuminantD65(float w) { auto i0 = size_t((w - CIE_D65_START) / CIE_D65_INTERVAL); uint2 indexBounds{i0, std::min(i0 + 1, CIE_D65_END)}; float2 wavelengthBounds = CIE_D65_START + float2{indexBounds} * CIE_D65_INTERVAL; float t = (w - wavelengthBounds.x) / (wavelengthBounds.y - wavelengthBounds.x); return lerp(CIE_D65[indexBounds.x], CIE_D65[indexBounds.y], t); } // For std::lower_bound bool operator<(const Sample& lhs, const Sample& rhs) { return lhs.w < rhs.w; } // The wavelength w must be between 360nm and 830nm static std::complex findSample(const std::vector& samples, float w) { auto i1 = std::lower_bound( samples.begin(), samples.end(), Sample{w, 0.0f + 0.0if}); auto i0 = i1 - 1; // Interpolate the complex IORs float t = (w - i0->w) / (i1->w - i0->w); float n = lerp(i0->ior.real(), i1->ior.real(), t); float k = lerp(i0->ior.imag(), i1->ior.imag(), t); return { n, k }; } static float fresnel(const std::complex& sample) { return (((sample - (1.0f + 0if)) * (std::conj(sample) - (1.0f + 0if))) / ((sample + (1.0f + 0if)) * (std::conj(sample) + (1.0f + 0if)))).real(); } static float3 XYZ_to_sRGB(const float3& v) { const mat3f XYZ_sRGB{ 3.2404542f, -0.9692660f, 0.0556434f, -1.5371385f, 1.8760108f, -0.2040259f, -0.4985314f, 0.0415560f, 1.0572252f }; return XYZ_sRGB * v; } // Outputs a linear sRGB color static float3 computeColor(const std::vector& samples) { float3 xyz{0.0f}; float y = 0.0f; for (size_t i = 0; i < CIE_XYZ_COUNT; i++) { // Current wavelength float w = CIE_XYZ_START + i; // Find most appropriate CIE XYZ sample for the wavelength auto sample = findSample(samples, w); // Compute Fresnel reflectance at normal incidence float f0 = fresnel(sample); // We need to multiply by the spectral power distribution of the illuminant float d65 = illuminantD65(w); xyz += f0 * CIE_XYZ[i] * d65; y += CIE_XYZ[i].y * d65; } // Normalize so that 100% reflectance at every wavelength yields Y=1 xyz /= y; float3 linear = XYZ_to_sRGB(xyz); // Normalize out-of-gamut values if (any(greaterThan(linear, float3{1.0f}))) linear *= 1.0f / max(linear); return linear; } ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ [Listing [specularColorImpl]: C++ implementation to compute the base color of a metallic surface from spectral data] Special thanks to Naty Hoffman for his valuable help on this topic. ## Importance sampling for the IBL In the discrete domain, the integral can be approximated with sampling as defined in equation $\ref{iblSampling}$. $$$$\label{iblSampling} \Lout(n,v,\Theta) \equiv \frac{1}{N} \sum_{i}^{N} f(l_{i}^{uniform},v,\Theta) L_{\perp}(l_i) \left< n \cdot l_i^{uniform} \right>$$$$ Unfortunately, we would need too many samples to evaluate this integral. A technique commonly used is to choose samples that are more "important" more often, this is called _importance sampling_. In our case we'll use the distribution of micro-facets normals, $D_{ggx}$, as the distribution of important samples. The evaluation of $\Lout(n,v,\Theta)$ with importance sampling is presented in equation $\ref{annexIblImportanceSampling}$. $$$$\label{annexIblImportanceSampling} \Lout(n,v,\Theta) \equiv \frac{1}{N} \sum_{i}^{N} \frac{f(l_{i},v,\Theta)}{p(l_i,v,\Theta)} L_{\perp}(l_i) \left< n \cdot l_i \right>$$$$ In equation $\ref{annexIblImportanceSampling}$, $p$ is the probability density function (PDF) of the distribution of _important direction samples_ $l_i$. These samples depend on $h_i$, $v$ and $\alpha$. The definition of the PDF is shown in equation $\ref{iblPDF}$. $h_i$ is given by the distribution we chose, see section [Choosing important directions] for more details. The _important direction samples_ $l_i$ are calculated as the reflection of $v$ around $h_i$, and therefore **do not** have the same PDF as $h_i$. The PDF of a transformed distribution is given by: $$$$p(T_r(x)) = p(x) |J(T_r)|$$$$ Where $|J(T_r)|$ is the determinant of the Jacobian of the transform. In our case we're considering the transform from $h_i$ to $l_i$ and the determinant of its Jacobian is given in \ref{iblPDF}. $$$$\label{iblPDF} p(l,v,\Theta) = D(h,\alpha) \left< \NoH \right> |J_{h \rightarrow l}| \\ |J_{h \rightarrow l}| = \frac{1}{4 \left< \VoH \right>}$$$$ ### Choosing important directions Refer to section [Choosing important directions for sampling the BRDF] for more details. Given a uniform distribution $(\zeta_{\phi},\zeta_{\theta})$ the important direction $l$ is defined by equation $\ref{importantDirection}$. $$$$\label{importantDirection} \phi = 2 \pi \zeta_{\phi} \\ \theta = cos^{-1} \sqrt{\frac{1 - \zeta_{\theta}}{(\alpha^2 - 1)\zeta_{\theta}+1}} \\ l = \{ cos \phi sin \theta, sin \phi sin \theta, cos \theta \}$$$$ Typically, $(\zeta_{\phi},\zeta_{\theta})$ are chosen using the Hammersley uniform distribution algorithm described in section [Hammersley sequence]. ### Pre-filtered importance sampling Importance sampling considers only the PDF to generate important directions; in particular, it is oblivious to the actual content of the IBL. If the latter contains high frequencies in areas without a lot of samples, the integration won’t be accurate. This can be somewhat mitigated by using a technique called _pre-filtered importance sampling_, in addition this allows the integral to converge with many fewer samples. Pre-filtered importance sampling uses several images of the environment increasingly low-pass filtered. This is typically implemented very efficiently with mipmaps and a box filter. The LOD is selected based on the sample importance, that is, low probability samples use a higher LOD index (more filtered). This technique is described in details in [#Krivanek08]. The cubemap LOD is determined in the following way: \begin{align*} lod &= log_4 \left( K\frac{\Omega_s}{\Omega_p} \right) \\ K &= 4.0 \\ \Omega_s &= \frac{1}{N \cdot p(l_i)} \\ \Omega_p &\approx \frac{4\pi}{6 \cdot width \cdot height} \end{align*} Where $K$ is a constant determined empirically, $p$ the PDF of the BRDF, $\Omega_{s}$ the solid angle associated to the sample and $\Omega_p$ the solid angle associated with the texel in the cubemap. Cubemap sampling is done using seamless trilinear filtering. It is extremely important to sample the cubemap correctly across faces using OpenGL's seamless sampling feature or any other technique that avoids/reduces seams. Table [importanceSamplingViz] shows a comparison between importance sampling and pre-filtered importance sampling when applied to figure [importanceSamplingRef]. ![Figure [importanceSamplingRef]: Importance sampling image reference](images/image_is_original.png) Samples | Importance sampling | Pre-filtered importance sampling ---------|-------------------------------|--------------------------------------- 4096 | ![](images/image_is_4096.png) |   1024 | ![](images/image_is_1024.png) | ![](images/image_fis_1024.png) 32 | ![](images/image_is_32.png) | ![](images/image_fis_32.png) [Table [importanceSamplingViz]: Importance sampling vs pre-filtered importance sampling with $\alpha = 0.4$] The reference renderer used in the comparison below performs no approximation. In particular, it does not assume $v = n$ and does not perform the split sum approximation. The pre-filtered renderer uses all the techniques discussed in this section: pre-filtered cubemaps, the analytic formulation of the DFG term, and of course the split sum approximation. Left: reference renderer, right: pre-filtered importance sampling. ![](images/image_is_ref_1.png) ![](images/image_filtered_1.png) ![](images/image_is_ref_2.png) ![](images/image_filtered_2.png) ![](images/image_is_ref_3.png) ![](images/image_filtered_3.png) ![](images/image_is_ref_4.png) ![](images/image_filtered_4.png) ## Choosing important directions for sampling the BRDF For simplicity we use the $D$ term of the BRDF as the PDF, however the PDF must be normalized such that the integral over the hemisphere is 1: $$$$\int_{\Omega}p(m)dm = 1 \\ \int_{\Omega}D(m)(n \cdot m)dm = 1 \\ \int_{\phi=0}^{2\pi}\int_{\theta=0}^{\frac{\pi}{2}}D(\theta,\phi) cos \theta sin \theta d\theta d\phi = 1 \\$$$$ The PDF of the BRDF can therefore be expressed as in equation $\ref{importantPDF}$: $$$$\label{importantPDF} p(\theta,\phi) = \frac{\alpha^2}{\pi(cos^2\theta (\alpha^2-1) + 1)^2} cos\theta sin\theta$$$$ The term $sin\theta$ comes from the differential solid angle $sin\theta d\phi d\theta$ since we integrate over a sphere. We sample $\theta$ and $\phi$ independently: \begin{align*} p(\theta) &= \int_0^{2\pi} p(\theta,\phi) d\phi = \frac{2\alpha^2}{(cos^2\theta (\alpha^2-1) + 1)^2} cos\theta sin\theta \\ p(\phi) &= \frac{p(\theta,\phi)}{p(\phi)} = \frac{1}{2\pi} \end{align*} The expression of $p(\phi)$ is true for an isotropic distribution of normals. We then calculate the cumulative distribution function (CDF) for each variable: \begin{align*} P(s_{\phi}) &= \int_{0}^{s_{\phi}} p(\phi) d\phi = \frac{s_{\phi}}{2\pi} \\ P(s_{\theta}) &= \int_{0}^{s_{\theta}} p(\theta) d\theta = 2 \alpha^2 \left( \frac{1}{(2\alpha^4-4\alpha^2+2) cos(s_{\theta})^2 + 2\alpha^2 - 2} - \frac{1}{2\alpha^4-2\alpha^2} \right) \end{align*} We set $P(s_{\phi})$ and $P(s_{\theta})$ to random variables $\zeta_{\phi}$ and $\zeta_{\theta}$ and solve for $s_{\phi}$ and $s_{\theta}$ respectively: \begin{align*} P(s_{\phi}) &= \zeta_{\phi} \rightarrow s_{\phi} = 2\pi\zeta_{\phi} \\ P(s_{\theta}) &= \zeta_{\theta} \rightarrow s_{\theta} = cos^{-1} \sqrt{\frac{1-\zeta_{\theta}}{(\alpha^2-1)\zeta_{\theta}+1}} \end{align*} So given a uniform distribution $(\zeta_{\phi},\zeta_{\theta})$, our important direction $l$ is defined as: \begin{align*} \phi &= 2\pi\zeta_{\phi} \\ \theta &= cos^{-1} \sqrt{\frac{1-\zeta_{\theta}}{(\alpha^2-1)\zeta_{\theta}+1}} \\ l &= \{ cos\phi sin\theta,sin\phi sin\theta,cos\theta \} \end{align*} ## Hammersley sequence ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ vec2f hammersley(uint i, float numSamples) { uint bits = i; bits = (bits << 16) | (bits >> 16); bits = ((bits & 0x55555555) << 1) | ((bits & 0xAAAAAAAA) >> 1); bits = ((bits & 0x33333333) << 2) | ((bits & 0xCCCCCCCC) >> 2); bits = ((bits & 0x0F0F0F0F) << 4) | ((bits & 0xF0F0F0F0) >> 4); bits = ((bits & 0x00FF00FF) << 8) | ((bits & 0xFF00FF00) >> 8); return vec2f(i / numSamples, bits / exp2(32)); } ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ [C++ implementation of a Hammersley sequence generator] ## Precomputing L for image-based lighting The term $L_{DFG}$ is only dependent on $\NoV$. Below, the normal is arbitrarily set to $n=\left[0, 0, 1\right]$ and $v$ is chosen to satisfy $\NoV$. The vector $h_i$ is the $D_{GGX}(\alpha)$ important direction sample $i$. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ float GDFG(float NoV, float NoL, float a) { float a2 = a * a; float GGXL = NoV * sqrt((-NoL * a2 + NoL) * NoL + a2); float GGXV = NoL * sqrt((-NoV * a2 + NoV) * NoV + a2); return (2 * NoL) / (GGXV + GGXL); } float2 DFG(float NoV, float a) { float3 V; V.x = sqrt(1.0f - NoV*NoV); V.y = 0.0f; V.z = NoV; float2 r = 0.0f; for (uint i = 0; i < sampleCount; i++) { float2 Xi = hammersley(i, sampleCount); float3 H = importanceSampleGGX(Xi, a, N); float3 L = 2.0f * dot(V, H) * H - V; float VoH = saturate(dot(V, H)); float NoL = saturate(L.z); float NoH = saturate(H.z); if (NoL > 0.0f) { float G = GDFG(NoV, NoL, a); float Gv = G * VoH / NoH; float Fc = pow(1 - VoH, 5.0f); r.x += Gv * (1 - Fc); r.y += Gv * Fc; } } return r * (1.0f / sampleCount); } ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ [C++ implementation of the $L_{DFG}$ term] ## Spherical Harmonics Symbol | Definition :---------------------------:|:---------------------------| $K^m_l$ | Normalization factors $P^m_l(x)$ | Associated Legendre polynomials $y^m_l$ | Spherical harmonics bases, or SH bases $L^m_l$ | SH coefficients of the $L(s)$ function defined on the unit sphere [Table [shSymbols]: Spherical harmonics symbols definitions] ### Basis functions Spherical parameterization of points on the surface of the unit sphere: $$$$\{ x, y, z \} = \{ cos \phi sin \theta, sin \phi sin \theta, cos \theta \}$$$$ The complex spherical harmonics bases are given by: $$$$Y^m_l(\theta, \phi) = K^m_l e^{im\theta} P^{|m|}_l(cos \theta), l \in N, -l <= m <= l$$$$ However we only need the real bases: \begin{align*} y^{m > 0}_l &= \sqrt{2} K^m_l cos(m \phi) P^m_l(cos \theta) \\ y^{m < 0}_l &= \sqrt{2} K^m_l sin(m \phi) P^{|m|}_l(cos \theta) \\ y^0_l &= K^0_l P^0_l(cos \theta) \end{align*} The normalization factors are given by: $$$$K^m_l = \sqrt{\frac{(2l + 1)(l - |m|)!}{4 \pi (l + |m|)!}}$$$$ The associated Legendre polynomials $P^{|m|}_l$ can be calculated from the following recursions: $$$$\label{shRecursions} P^0_0(x) = 1 \\ P^0_1(x) = x \\ P^l_l(x) = (-1)^l (2l - 1)!! (1 - x^2)^{\frac{l}{2}} \\ P^m_l(x) = \frac{((2l - 1) x P^m_{l - 1} - (l + m - 1) P^m_{l - 2})}{l - m} \\$$$$ Computing $y^{|m|}_l$ requires to compute $P^{|m|}_l(z)$ first. This can be accomplished fairly easily using the recursions in equation $\ref{shRecursions}$. The third recursion can be used to "move diagonally" in table [basisFunctions], i.e. calculating $y^0_0$, $y^1_1$, $y^2_2$ etc. Then, the fourth recursion can be used to move vertically. Band index | Basis functions $-l <= m <= l$ :-----------:|:---------------------------------:| $l = 0$ | $y^0_0$ $l = 1$ | $y^{-1}_1$ $y^0_1$ $y^1_1$ $l = 2$ | $y^{-2}_2$ $y^{-1}_2$ $y^0_2$ $y^1_2$ $y^2_2$ [Table [basisFunctions]: Basis functions per band] It’s also fairly easy to compute the trigonometric terms recursively: \begin{align*} C_m &\equiv cos(m \phi) \\ S_m &\equiv sin(m \phi) \\ \{ x, y, z \} &= \{ cos \phi sin \theta, sin \phi sin \theta, cos \theta \} \end{align*} Using the angle sum trigonometric identities: \begin{align*} cos(m \phi + \phi) &= cos(m \phi) cos(\phi) - sin(m \phi) sin(\phi) \Leftrightarrow C_{m + 1} = \frac{(x C_m - y S_m)}{sin(\theta)^{|m + 1|}} \\ sin(m \phi + \phi) &= sin(m \phi) sin(\phi) + cos(m \phi) sin(\phi) \Leftrightarrow S_{m + 1} = \frac{(x S_m - y C_m)}{sin(\theta)^{|m + 1|}} \end{align*} The equations above have an extra term $sin(\theta)^{-|m + 1|}$ but we can compensate for that in the $P^{|m|}_l(z)$ recursion by multiplying $P^l_l(z)$ by $sin(\theta)^{|m + 1|}$ which greatly simplifies the third equation in $\ref{shRecursions}$ because $P^l_l(cos \theta) sin(\theta)^{-l} = (-1)^l(2l - 1)!!$. Listing [nonNormalizedSHBasis] shows the C++ code to compute the non-normalized SH basis $\frac{y^m_l(s)}{\sqrt{2} K^m_l}$: ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ static inline size_t SHindex(ssize_t m, size_t l) { return l * (l + 1) + m; } void computeShBasis( double* const SHb, size_t numBands, const vec3& s) { // handle m=0 separately, since it produces only one coefficient double Pml_2 = 0; double Pml_1 = 1; SHb[0] = Pml_1; for (ssize_t l = 1; l < numBands; l++) { double Pml = ((2 * l - 1) * Pml_1 * s.z - (l - 1) * Pml_2) / l; Pml_2 = Pml_1; Pml_1 = Pml; SHb[SHindex(0, l)] = Pml; } double Pmm = 1; for (ssize_t m = 1; m < numBands ; m++) { Pmm = (1 - 2 * m) * Pmm; double Pml_2 = Pmm; double Pml_1 = (2 * m + 1)*Pmm*s.z; // l == m SHb[SHindex(-m, m)] = Pml_2; SHb[SHindex( m, m)] = Pml_2; if (m + 1 < numBands) { // l == m+1 SHb[SHindex(-m, m + 1)] = Pml_1; SHb[SHindex( m, m + 1)] = Pml_1; for (ssize_t l = m + 2; l < numBands; l++) { double Pml = ((2 * l - 1) * Pml_1 * s.z - (l + m - 1) * Pml_2) / (l - m); Pml_2 = Pml_1; Pml_1 = Pml; SHb[SHindex(-m, l)] = Pml; SHb[SHindex( m, l)] = Pml; } } } double Cm = s.x; double Sm = s.y; for (ssize_t m = 1; m <= numBands ; m++) { for (ssize_t l = m; l < numBands ; l++) { SHb[SHindex(-m, l)] *= Sm; SHb[SHindex( m, l)] *= Cm; } double Cm1 = Cm * s.x - Sm * s.y; double Sm1 = Sm * s.x + Cm * s.y; Cm = Cm1; Sm = Sm1; } } ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ [Listing [nonNormalizedSHBasis]: C++ implementation to compute a non-normalized SH basis] Normalized SH basis functions $y^m_l(s)$ for the first 3 bands: Band | $m = -2$ | $m = -1$ | $m = 0$ | $m = 1$ | $m = 2$ | :-------:|:------------------------------------:|:-------------------------------------:|:---------------------------------------------------:|:-------------------------------------:|:---------------------------------------------:| $l = 0$ | | | $\frac{1}{2}\sqrt{\frac{1}{\pi}}$ | | | $l = 1$ | | $-\frac{1}{2}\sqrt{\frac{3}{\pi}}y$ | $\frac{1}{2}\sqrt{\frac{3}{\pi}}z$ | $-\frac{1}{2}\sqrt{\frac{3}{\pi}}x$ | | $l = 2$ | $\frac{1}{2}\sqrt{\frac{15}{\pi}}xy$ | $-\frac{1}{2}\sqrt{\frac{15}{\pi}}yz$ | $\frac{1}{4}\sqrt{\frac{5}{\pi}}(2z^2 - x^2 - y^2)$ | $-\frac{1}{2}\sqrt{\frac{15}{\pi}}xz$ | $\frac{1}{4}\sqrt{\frac{15}{\pi}}(x^2 - y^2)$ | [Table [basisFunctions]: Normalized basis functions per band] ### Decomposition and reconstruction A function $L(s)$ defined on a sphere is projected to the SH basis as follows: $$$$L^m_l = \int_\Omega L(s) y^m_l(s) ds \\ L^m_l = \int_{\theta = 0}^{\pi} \int_{\phi = 0}^{2\pi} L(\theta, \phi) y^m_l(\theta, \phi) sin \theta d\theta d\phi$$$$ Note that each $L^m_l$ is a vector of 3 values, one for each RGB color channel. The inverse transformation, or reconstruction, or rendering, from the SH coefficients is given by: $$$$\hat{L}(s) = \sum_l \sum_{m = -l}^l L^m_l y^m_l(s)$$$$ ### Decomposition of $\left< cos \theta \right>$ Since $\left< cos \theta \right>$ does not depend on $\phi$ (azimuthal independence), the integral simplifies to: \begin{align*} C^0_l &= 2\pi \int_0^{\pi} \left< cos \theta \right> y^0_l(\theta) sin \theta d\theta \\ C^0_l &= 2\pi K^)_l \int_0^{\frac{\pi}{2}} P^0_l(cos \theta) cos \theta sin \theta d\theta \\ C^m_l &= 0, m != 0 \end{align*} In [#Ramamoorthi01] an analytical solution to the integral is described: \begin{align*} C_1 &= \sqrt{\frac{\pi}{3}} \\ C_{odd} &= 0 \\ C_{l, even} &= 2\pi \sqrt{\frac{2l + 1}{4\pi}} \frac{(-1)^{\frac{l}{2} - 1}}{(l + 2)(l - 1)} \frac{l!}{2^l (\frac{l!}{2})^2} \end{align*} The first few coefficients are: \begin{align*} C_0 &= +0.88623 \\ C_1 &= +1.02333 \\ C_2 &= +0.49542 \\ C_3 &= +0.00000 \\ C_4 &= -0.11078 \end{align*} Very few coefficients are needed to reasonably approximate $\left< cos \theta \right>$, as shown in figure [shCosThetaApprox]. ![Figure [shCosThetaApprox]: Approximation of $cos \theta$ with SH coefficients](images/chart_sh_cos_thera_approx.png) ### Convolution Convolutions by a kernel $h$ that has a circular symmetry can be applied directly and easily in SH space: $$$$(h * f)^m_l = \sqrt{\frac{4\pi}{2l + 1}} h^0_l(s) f^m_l(s)$$$$ Conveniently, $\sqrt{\frac{4\pi}{2l + 1}} = \frac{1}{K^0_l}$, so in practice we pre-multiply $C_l$ by $\frac{1}{K^0_l}$ and we get a simpler expression: $$$$\hat{C}_{l, even} = 2\pi \frac{(-1)^{\frac{l}{2} - 1}}{(l + 2)(l - 1)} \frac{l!}{2^l (\frac{l!}{2})^2} \\ \hat{C}_1 = \frac{2\pi}{3}$$$$ Here is the C++ code to compute $\hat{C}_l$: ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ static double factorial(size_t n, size_t d = 1); // < cos(theta) > SH coefficients pre-multiplied by 1 / K(0,l) double computeTruncatedCosSh(size_t l) { if (l == 0) { return M_PI; } else if (l == 1) { return 2 * M_PI / 3; } else if (l & 1) { return 0; } const size_t l_2 = l / 2; double A0 = ((l_2 & 1) ? 1.0 : -1.0) / ((l + 2) * (l - 1)); double A1 = factorial(l, l_2) / (factorial(l_2) * (1 << l)); return 2 * M_PI * A0 * A1; } // returns n! / d! double factorial(size_t n, size_t d ) { d = std::max(size_t(1), d); n = std::max(size_t(1), n); double r = 1.0; if (n == d) { // intentionally left blank } else if (n > d) { for ( ; n>d ; n--) { r *= n; } } else { for ( ; d>n ; d--) { r *= d; } r = 1.0 / r; } return r; } ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ## Sample validation scene for Mitsuba ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ <scene version="0.5.0"> <integrator type="path"/> <shape type="serialized" id="sphere_mesh"> <string name="filename" value="plastic_sphere.serialized"/> <integer name="shapeIndex" value="0"/> <bsdf type="roughplastic"> <string name="distribution" value="ggx"/> <float name="alpha" value="0.0"/> <srgb name="diffuseReflectance" value="0.81, 0.0, 0.0"/> </bsdf> </shape> <emitter type="envmap"> <string name="filename" value="../../environments/office/office.exr"/> <float name="scale" value="35000.0" /> <boolean name="cache" value="false" /> </emitter> <emitter type="directional"> <vector name="direction" x="-1" y="-1" z="1" /> <rgb name="irradiance" value="120000.0, 115200.0, 114000.0" /> </emitter> <sensor type="perspective"> <float name="farClip" value="12.0"/> <float name="focusDistance" value="4.1"/> <float name="fov" value="45"/> <string name="fovAxis" value="y"/> <float name="nearClip" value="0.01"/> <transform name="toWorld"> <lookat target="0, 0, 0" origin="0, 0, -3.1" up="0, 1, 0"/> </transform> <sampler type="ldsampler"> <integer name="sampleCount" value="256"/> </sampler> <film type="ldrfilm"> <integer name="height" value="1440"/> <integer name="width" value="2048"/> <float name="exposure" value="-15.23" /> <rfilter type="gaussian"/> </film> </sensor> </scene> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ## Light assignment with froxels Assigning lights to froxels can be implemented on the GPU using two compute shaders. The first one, shown in listing [froxelGeneration], creates the froxels data (4 planes + a min Z and max Z per froxel) in an SSBO and needs to be run only once. The shader requires the following uniforms: Projection matrix : The projection matrix used to render the scene (view space to clip space transformation). Inverse projection matrix : The inverse of the projection matrix used to render the scene (clip space to view space transformation). Depth parameters : $-log2(\frac{z_{lighnear}}{z_{far}}) \frac{1}{maxSlices-1}$, maximum number of depth slices, Z near and Z far. Clip space size : $\frac{F_x \times F_r}{w} \times 2$, with $F_x$ the number of tiles on the X axis, $F_r$ the resolution in pixels of a tile and w the width in pixels of the render target. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ #version 310 es precision highp float; precision highp int; #define FROXEL_RESOLUTION 80u layout(local_size_x = 1, local_size_y = 1, local_size_z = 1) in; layout(location = 0) uniform mat4 projectionMatrix; layout(location = 1) uniform mat4 projectionInverseMatrix; layout(location = 2) uniform vec4 depthParams; // index scale, index bias, near, far layout(location = 3) uniform float clipSpaceSize; struct Froxel { // NOTE: the planes should be stored in vec4[4] but the // Adreno shader compiler has a bug that causes the data // to not be read properly inside the loop vec4 plane0; vec4 plane1; vec4 plane2; vec4 plane3; vec2 minMaxZ; }; layout(binding = 0, std140) writeonly restrict buffer FroxelBuffer { Froxel data[]; } froxels; shared vec4 corners[4]; shared vec2 minMaxZ; vec4 projectionToView(vec4 p) { p = projectionInverseMatrix * p; return p / p.w; } vec4 createPlane(vec4 b, vec4 c) { // standard plane equation, with a at (0, 0, 0) return vec4(normalize(cross(c.xyz, b.xyz)), 1.0); } void main() { uint index = gl_WorkGroupID.x + gl_WorkGroupID.y * gl_NumWorkGroups.x + gl_WorkGroupID.z * gl_NumWorkGroups.x * gl_NumWorkGroups.y; if (gl_LocalInvocationIndex == 0u) { // first tile the screen and build the frustum for the current tile vec2 renderTargetSize = vec2(FROXEL_RESOLUTION * gl_NumWorkGroups.xy); vec2 frustumMin = vec2(FROXEL_RESOLUTION * gl_WorkGroupID.xy); vec2 frustumMax = vec2(FROXEL_RESOLUTION * (gl_WorkGroupID.xy + 1u)); corners[0] = vec4( frustumMin.x / renderTargetSize.x * clipSpaceSize - 1.0, (renderTargetSize.y - frustumMin.y) / renderTargetSize.y * clipSpaceSize - 1.0, 1.0, 1.0 ); corners[1] = vec4( frustumMax.x / renderTargetSize.x * clipSpaceSize - 1.0, (renderTargetSize.y - frustumMin.y) / renderTargetSize.y * clipSpaceSize - 1.0, 1.0, 1.0 ); corners[2] = vec4( frustumMax.x / renderTargetSize.x * clipSpaceSize - 1.0, (renderTargetSize.y - frustumMax.y) / renderTargetSize.y * clipSpaceSize - 1.0, 1.0, 1.0 ); corners[3] = vec4( frustumMin.x / renderTargetSize.x * clipSpaceSize - 1.0, (renderTargetSize.y - frustumMax.y) / renderTargetSize.y * clipSpaceSize - 1.0, 1.0, 1.0 ); uint froxelSlice = gl_WorkGroupID.z; minMaxZ = vec2(0.0, 0.0); if (froxelSlice > 0u) { minMaxZ.x = exp2((float(froxelSlice) - depthParams.y) * depthParams.x) * depthParams.w; } minMaxZ.y = exp2((float(froxelSlice + 1u) - depthParams.y) * depthParams.x) * depthParams.w; } if (gl_LocalInvocationIndex == 0u) { vec4 frustum[4]; frustum[0] = projectionToView(corners[0]); frustum[1] = projectionToView(corners[1]); frustum[2] = projectionToView(corners[2]); frustum[3] = projectionToView(corners[3]); froxels.data[index].plane0 = createPlane(frustum[0], frustum[1]); froxels.data[index].plane1 = createPlane(frustum[1], frustum[2]); froxels.data[index].plane2 = createPlane(frustum[2], frustum[3]); froxels.data[index].plane3 = createPlane(frustum[3], frustum[0]); froxels.data[index].minMaxZ = minMaxZ; } } ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ [Listing [froxelGeneration]: GLSL implementation of froxels data generation (compute shader)] The second compute shader, shown in listing [froxelEvaluation], runs every frame (if the camera and/or lights have changed) and assigns all the lights to their respective froxels. This shader relies only on a couple of uniforms (the number of point/spot lights and the view matrix) and four SSBOs: Light index buffer : For each froxel, the index of each light that affects said froxel. The indices for point lights are written first and if there is enough space left, the indices for spot lights are written as well. A sentinel of value 0x7fffffffu separates point and spot lights and/or marks the end of the froxel's list of lights. Each froxel has a maximum number of lights (point + spot). Point lights buffer : Array of structures describing the scene's point lights. Spot lights buffer : Array of structures describing the scene's spot lights. Froxels buffer : The list of froxels represented by planes, created by the previous compute shader. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ #version 310 es precision highp float; precision highp int; #define LIGHT_BUFFER_SENTINEL 0x7fffffffu #define MAX_FROXEL_LIGHT_COUNT 32u #define THREADS_PER_FROXEL_X 8u #define THREADS_PER_FROXEL_Y 8u #define THREADS_PER_FROXEL_Z 1u #define THREADS_PER_FROXEL (THREADS_PER_FROXEL_X * \ THREADS_PER_FROXEL_Y * THREADS_PER_FROXEL_Z) layout(local_size_x = THREADS_PER_FROXEL_X, local_size_y = THREADS_PER_FROXEL_Y, local_size_z = THREADS_PER_FROXEL_Z) in; // x = point lights, y = spot lights layout(location = 0) uniform uvec2 totalLightCount; layout(location = 1) uniform mat4 viewMatrix; layout(binding = 0, packed) writeonly restrict buffer LightIndexBuffer { uint index[]; } lightIndexBuffer; struct PointLight { vec4 positionFalloff; // x, y, z, falloff vec4 colorIntensity; // r, g, b, intensity vec4 directionIES; // dir x, dir y, dir z, IES profile index }; layout(binding = 1, std140) readonly restrict buffer PointLightBuffer { PointLight lights[]; } pointLights; struct SpotLight { vec4 positionFalloff; // x, y, z, falloff vec4 colorIntensity; // r, g, b, intensity vec4 directionIES; // dir x, dir y, dir z, IES profile index vec4 angle; // angle scale, angle offset, unused, unused }; layout(binding = 2, std140) readonly restrict buffer SpotLightBuffer { SpotLight lights[]; } spotLights; struct Froxel { // NOTE: the planes should be stored in vec4[4] but the // Adreno shader compiler has a bug that causes the data // to not be read properly inside the loop vec4 plane0; vec4 plane1; vec4 plane2; vec4 plane3; vec2 minMaxZ; }; layout(binding = 3, std140) readonly restrict buffer FroxelBuffer { Froxel data[]; } froxels; shared uint groupLightCounter; shared uint groupLightIndexBuffer[MAX_FROXEL_LIGHT_COUNT]; float signedDistanceFromPlane(vec4 p, vec4 plane) { // plane.w == 0.0, simplify computation return dot(plane.xyz, p.xyz); } void synchronize() { memoryBarrierShared(); barrier(); } void main() { if (gl_LocalInvocationIndex == 0u) { groupLightCounter = 0u; } memoryBarrierShared(); uint froxelIndex = gl_WorkGroupID.x + gl_WorkGroupID.y * gl_NumWorkGroups.x + gl_WorkGroupID.z * gl_NumWorkGroups.x * gl_NumWorkGroups.y; Froxel current = froxels.data[froxelIndex]; uint offset = gl_LocalInvocationID.x + gl_LocalInvocationID.y * THREADS_PER_FROXEL_X; for (uint i = 0u; i < totalLightCount.x && groupLightCounter < MAX_FROXEL_LIGHT_COUNT && offset + i < totalLightCount.x; i += THREADS_PER_FROXEL) { uint currentLight = offset + i; vec4 center = pointLights.lights[currentLight].positionFalloff; center.xyz = (viewMatrix * vec4(center.xyz, 1.0)).xyz; float r = inversesqrt(center.w); if (-center.z + r > current.minMaxZ.x && -center.z - r <= current.minMaxZ.y) { if (signedDistanceFromPlane(center, current.plane0) < r && signedDistanceFromPlane(center, current.plane1) < r && signedDistanceFromPlane(center, current.plane2) < r && signedDistanceFromPlane(center, current.plane3) < r) { uint index = atomicAdd(groupLightCounter, 1u); groupLightIndexBuffer[index] = currentLight; } } } synchronize(); uint pointLightCount = groupLightCounter; offset = froxelIndex * MAX_FROXEL_LIGHT_COUNT; for (uint i = gl_LocalInvocationIndex; i < pointLightCount; i += THREADS_PER_FROXEL) { lightIndexBuffer.index[offset + i] = groupLightIndexBuffer[i]; } if (gl_LocalInvocationIndex == 0u) { if (pointLightCount < MAX_FROXEL_LIGHT_COUNT) { lightIndexBuffer.index[offset + pointLightCount] = LIGHT_BUFFER_SENTINEL; } } } ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ [Listing [froxelEvaluation]: GLSL implementation of assigning lights to froxels (compute shader)] # Revisions February 20, 2019: Cloth shading - Removed Fresnel term from the cloth BRDF - Removed cloth DFG approximations, replaced with a new channel in the DFG LUT August 21, 2018: Multiscattering - Added section [Energy loss in specular reflectance] on how to compensate for energy loss in single scattering BRDFs August 17, 2018: Specular color - Added section [Specular color] to explain how the base color of various metals is computed August 15, 2018: Fresnel - Added a description of the Fresnel effect in section [Fresnel (specular F)] August 9, 2018: Lighting - Added explanation about pre-exposed lights August 7, 2018: Cloth model - Added description of the "Charlie" NDF August 3, 2018: First public version # Bibliography [#Ashdown98]: Ian Ashdown. 1998. Parsing the IESNA LM-63 photometric data file. http://lumen.iee.put.poznan.pl/kw/iesna.txt [#Ashikhmin00]: Michael Ashikhmin, Simon Premoze and Peter Shirley. A Microfacet-based BRDF Generator. *SIGGRAPH '00 Proceedings*, 65-74. [#Ashikhmin07]: Michael Ashikhmin and Simon Premoze. 2007. Distribution-based BRDFs. [#Burley12]: Brent Burley. 2012. Physically Based Shading at Disney. *Physically Based Shading in Film and Game Production, ACM SIGGRAPH 2012 Courses*. [#Estevez17]: Alejandro Conty Estevez and Christopher Kulla. 2017. Production Friendly Microfacet Sheen BRDF. *ACM SIGGRAPH 2017*. [#Hammon17]: Earl Hammon. 217. PBR Diffuse Lighting for GGX+Smith Microsurfaces. *GDC 2017*. [#Heitz14]: Eric Heitz. 2014. Understanding the Masking-Shadowing Function in Microfacet-Based BRDFs. *Journal of Computer Graphics Techniques*, 3 (2). [#Heitz16]: Eric Heitz et al. 2016. Multiple-Scattering Microfacet BSDFs with the Smith Model. *ACM SIGGRAPH 2016*. [#Hill12]: Colin Barré-Brisebois and Stephen Hill. 2012. Blending in Detail. http://blog.selfshadow.com/publications/blending-in-detail/ [#Karis13]: Brian Karis. 2013. Specular BRDF Reference. http://graphicrants.blogspot.com/2013/08/specular-brdf-reference.html [#Karis14]: Brian Karis. 2014. Physically Based Shading on Mobile. https://www.unrealengine.com/blog/physically-based-shading-on-mobile [#Kelemen01]: Csaba Kelemen et al. 2001. A Microfacet Based Coupled Specular-Matte BRDF Model with Importance Sampling. *Eurographics Short Presentations*. [#Krystek85]: M. Krystek. 1985. An algorithm to calculate correlated color temperature. *Color Research & Application*, 10 (1), 38–40. [#Krivanek08]: Jaroslave Krivànek and Mark Colbert. 2008. Real-time Shading with Filtered Importance Sampling. *Eurographics Symposium on Rendering 2008*, Volume 27, Number 4. [#Kulla17]: Christopher Kulla and Alejandro Conty. 2017. Revisiting Physically Based Shading at Imageworks. *ACM SIGGRAPH 2017* [#Lagarde14]: Sébastien Lagarde and Charles de Rousiers. 2014. Moving Frostbite to PBR. *Physically Based Shading in Theory and Practice, ACM SIGGRAPH 2014 Courses*. [#Lagarde18]: Sébastien Lagarde and Evgenii Golubev. 2018. The road toward unified rendering with Unity’s high definition rendering pipeline. *Advances in Real-Time Rendering in Games, ACM SIGGRAPH 2018 Courses*. [#Lazarov13]: Dimitar Lazarov. 2013. Physically-Based Shading in Call of Duty: Black Ops. *Physically Based Shading in Theory and Practice, ACM SIGGRAPH 2013 Courses*. [#McAuley15]: Stephen McAuley. 2015. Rendering the World of Far Cry 4. *GDC 2015*. [#McGuire10]: Morgan McGuire. 2010. Ambient Occlusion Volumes. *High Performance Graphics*. [#Narkowicz14]: Krzysztof Narkowicz. 2014. Analytical DFG Term for IBL. https://knarkowicz.wordpress.com/2014/12/27/analytical-dfg-term-for-ibl [#Neubelt13]: David Neubelt and Matt Pettineo. 2013. Crafting a Next-Gen Material Pipeline for The Order: 1886. *Physically Based Shading in Theory and Practice, ACM SIGGRAPH 2013 Courses*. [#Oren94]: Michael Oren and Shree K. Nayar. 1994. Generalization of lambert's reflectance model. *SIGGRAPH*, 239–246. ACM. [#Pattanaik00]: Sumanta Pattanaik00 et al. 2000. Time-Dependent Visual Adaptation For Fast Realistic Image Display. *SIGGRAPH '00 Proceedings of the 27th annual conference on Computer graphics and interactive techniques*, 47-54. [#Ramamoorthi01]: Ravi Ramamoorthi and Pat Hanrahan. 2001. On the relationship between radiance and irradiance: determining the illumination from images of a convex Lambertian object. *Journal of the Optical Society of America*, Volume 18, Number 10, October 2001. [#Revie12]: Donald Revie. 2012. Implementing Fur in Deferred Shading. *GPU Pro 2*, Chapter 2. [#Russell15]: Jeff Russell. 2015. Horizon Occlusion for Normal Mapped Reflections. http://marmosetco.tumblr.com/post/81245981087 [#Schlick94]: Christophe Schlick. 1994. An Inexpensive BRDF Model for Physically-Based Rendering. *Computer Graphics Forum*, 13 (3), 233–246. [#Walter07]: Bruce Walter et al. 2007. Microfacet Models for Refraction through Rough Surfaces. *Proceedings of the Eurographics Symposium on Rendering*.