brdf_and_shading_effects.md

BRDF and Shading Effects Cheat Sheet

This document summarizes the main shading / BRDF effects we discussed and how they relate to SDFStudio-style configs (--pipeline.model.sdf-field.*). It also lists representative papers for deeper reading, especially works that try to explicitly model some of these effects in neural fields.

See also: pbr_learning_starter_pack.md — curated video/reading list for learning PBR extraction

---

1. Core Shading and BRDF Concepts

• BRDF (Bidirectional Reflectance Distribution Function)
  Describes how light is reflected at a surface as a function of incident and outgoing directions. A BRDF encodes the “material” independent of lighting and geometry.

  • Classic intros:
    • Pharr, Jakob, Humphreys – Physically Based Rendering (PBRT), 3rd ed.
    • Nicodemus et al. – Geometrical Considerations and Nomenclature for Reflectance.

• Diffuse reflection / Lambertian shading
  Ideal matte reflection; outgoing radiance is independent of view direction, proportional to max(0, n·l).

  • Oren, Nayar – Generalization of Lambert's Reflectance Model (Oren–Nayar BRDF).

• Specular reflection / highlights
  Mirror-like component concentrated near the reflection direction; responsible for shiny highlights.

  • Early microfacet: Torrance, Sparrow – Theory for Off-Specular Reflection.
  • Practical game/PBR overview: Karis – Real Shading in Unreal Engine 4.

• Diffuse vs specular separation (albedo vs highlights)
  Modeling color as a sum of a view-independent “albedo” term and a view-dependent specular term.

  • NeRFactor: Zhang et al. – NeRFactor: Neural Factorization of Shape and Reflectance under Unknown Illumination.
  • Neural Reflectance Fields: Bi et al. – Deep Reflectance Volumes / related NeRF-based reflectance works.

• Specular tint / metallic vs dielectric behavior
  Colored specular (metals) vs nearly white specular (dielectrics); in PBR often controlled by a metallic factor and F₀ (base reflectivity).

  • Burley – Physically-Based Shading at Disney.
  • Burley – Extending the Disney BRDF to a BSDF with Integrated Subsurface Scattering.

• Roughness
  Parameter controlling microfacet orientation distribution. Low roughness → sharp highlights; high roughness → wide, blurred highlights.

  • Heitz – Understanding the Masking-Shadowing Function in Microfacet-Based BRDFs.
  • Heitz – A Survey of Microfacet Models for Rough Surfaces.

• Microfacet BRDF models (GGX, Beckmann, etc.)
  Physically-based specular models using a normal distribution function (NDF), masking–shadowing, and Fresnel.

  • Walter et al. – Microfacet Models for Refraction through Rough Surfaces.
  • PBRT book again for detailed derivations.

• Masking–shadowing (geometry term)
  Microfacets self-occlude each other, reducing visible specular energy especially at grazing angles.

• Energy conservation & reciprocity
  Physical constraints on BRDFs: total reflected energy ≤ incoming, and swapping light/view directions leaves the BRDF unchanged.

---

2. Angular / View-Dependent Effects

• Angle of incidence / n·v term
  Cosine of the angle between surface normal and view direction; used in shading, foreshortening, Fresnel, and orientation losses.

• Foreshortening
  Apparent shrinking / darkening of surfaces as they tilt away from the viewer; mathematically tied to cosines like n·v or n·l.

• Limb darkening
  Generic term for surfaces or objects appearing darker near silhouettes (grazing angles), due to angular effects and transport.

• Fresnel effect
  Reflectance increasing at grazing angles; often approximated in graphics by Schlick’s formula.

  • Schlick – An Inexpensive BRDF Model for Physically-based Rendering.
  • Also discussed extensively in PBRT and Disney/UE4 shading notes.

• View-dependent appearance / view-dependent radiance
  Color changes with view direction even for fixed illumination, encompassing specular, Fresnel, anisotropy, etc.

  • Mildenhall et al. – NeRF: Representing Scenes as Neural Radiance Fields for View Synthesis.
  • Barron et al. – Mip-NeRF and Mip-NeRF 360 (better anti-aliasing and unbounded scenes).

---

3. Reflection, Refraction, Environment

• Reflection law (mirror reflection direction)
  Outgoing direction r = v - 2 (n·v) n; governs where specular highlights and mirror reflections appear.

• Environment reflections / environment mapping
  Reflective objects showing the surrounding scene; often modeled with environment maps or image-based lighting.

  • Ramamoorthi, Hanrahan – An Efficient Representation for Irradiance Environment Maps.
  • Debevec – Rendering Synthetic Objects into Real Scenes (light probes).

• Refraction / transmission
  Light bending when passing between media (Snell’s law), plus internal transmission.

  • Walter et al. – Microfacet Models for Refraction through Rough Surfaces.
  • Neural fields with refraction: e.g. Guo et al. – NeRFReN: Neural Radiance Fields with Reflections and Refractions.

• Transparent / refractive objects
  Glass, water, etc. require modeling multiple interfaces, refraction, and internal reflections. Standard NeRF/SDF pipelines (including SDFStudio/mini-mesh) largely treat these as “out of scope”.

• Subsurface scattering
  Light entering a translucent medium, scattering internally, and exiting at nearby points (skin, wax, marble).

  • Jensen et al. – A Practical Model for Subsurface Light Transport.
  • RenderMan/Disney subsurface models in Burley’s work.

• Volumetric light transport
  Light absorption, scattering, and emission in participating media (fog, smoke, semi-transparent volumes). NeRF’s volumetric rendering is a simplified version of this.

  • Kajiya, Von Herzen – Ray Tracing Volume Densities.
  • NeRF again for the simplified volumetric rendering integral.

---

4. NeRF / SDF-Specific Modeling Topics

• SDF normals (∇SDF) and geometric gradients
  Deriving surface normals from the gradient of a signed distance function; used heavily in NeuS/SDFStudio.

  • Wang et al. – NeuS: Learning Neural Implicit Surfaces by Volume Rendering for Multi-view Reconstruction.
  • Yariv et al. – VolSDF: Volume Rendering of Signed Distance Functions.

• View-direction encoding vs reflection-direction encoding
  Feeding the color network either encoded view directions or encoded reflection directions (or both).

  • Mildenhall et al. – NeRF (view-direction encoding).
  • Verbin et al. – Ref-NeRF: Structured View-Dependent Appearance for Neural Radiance Fields (reflection dirs, diffuse/specular split, tint, n·v).

• Diffuse/specular decomposition in neural fields (Ref-NeRF-style)
  Architectures that explicitly separate diffuse and specular components to stabilize geometry + appearance learning.

  • Verbin et al. – Ref-NeRF.
  • Boss et al. – NeRD: Neural Reflectance Decomposition from Image Collections.
  • Zhang et al. – NeRFactor (factorization of shape, reflectance, and lighting).

• Learned roughness and material parameters in NeRF-like models
  Predicting roughness, metallic, or other material parameters inside a neural field and using them in a BRDF head.

  • Srinivasan et al. – NeRV: Neural Reflectance and Visibility Fields for Relighting and View Synthesis.
  • Boss et al. – NeRD; Hasselgren et al. – Neural Reflectance Fields.
  • Recent “neural PBR” works that combine NeRF with explicit microfacet BRDFs.

• Lighting / environment estimation in neural fields
  Some methods explicitly estimate lighting (environment maps or spherical harmonics) jointly with geometry and reflectance.

  • Martin-Brualla et al. – NeRF in the Wild (NeRFW).
  • Srinivasan et al. – NeRV.
  • Boss et al. – NeRD.

---

5. How This Relates to SDFStudio Flags

In SDFStudio/mini-mesh, the following SDF field flags are loosely inspired by Ref-NeRF and related works:

Core flags

• use_diffuse_color Approximates a diffuse/specular decomposition: diffuse color predicted from geometry features; specular learned in the view-dependent MLP. Conceptually similar to Ref-NeRF's "diffuse color" head.

• use_reflections Uses reflection directions as part of the view encoding, following Ref-NeRF's reflection-direction encoding. This strongly couples specular highlight position to geometry and normals.

• use_n_dot_v Supplies n·v directly to the color MLP, making Fresnel-like ramps and limb-darkening behavior much easier to learn. Ref-NeRF also uses such angular cues.

• use_fresnel_term Appends a Schlick-style Fresnel scalar as an extra input to the color MLP. More explicit than use_n_dot_v for Fresnel effects.

• enable_pred_roughness (requires use_reflections=True) Predicts a roughness in [0, 1] and uses it to mix view-direction and reflection-direction encodings. This is a very lightweight proxy for roughness-dependent specular behavior; it is not a full analytic microfacet BRDF, but nudges the network toward roughness-consistent highlights and provides an interpretable roughness map.

Material-specific flags

• use_specular_tint (metals only) Adds a learned RGB tint to the specular component, analogous to metallic/colored specular behavior in Disney/UE4 shading. Do not use for dielectric materials (plastic, ceramics) which have white/neutral specular.

• specular_exclude_geo_features (requires use_diffuse_color=True) Excludes geometry features from the specular MLP, forcing the specular branch to be purely view-dependent. All spatial color variation goes through diffuse. Recommended for uniform plastic surfaces like lego.

• use_roughness_gated_specular (requires enable_pred_roughness=True + use_diffuse_color=True) Gates the specular contribution by (1 - roughness), so rough surfaces have minimal specular and smooth surfaces have full specular. Enforces physical constraint: rough surfaces scatter diffusely.

Advanced flags

• use_roughness_in_color_mlp (requires enable_pred_roughness=True) Appends the predicted roughness scalar as an extra input to the color MLP, allowing view-dependent appearance to condition on roughness.

• learned_specular_scale (requires use_diffuse_color=True) Replaces the fixed 0.5 specular multiplier with a learned per-point value in [0, 1], allowing the network to control specular intensity spatially.

• roughness_blend_space (requires enable_pred_roughness=True + use_reflections=True) Controls where to mix view/reflection signals: "encoding" (default) encodes separately then blends features; "direction" blends raw directions first then encodes once.

Summary table

Flag	Purpose	Requires	Use for
use_diffuse_color	Diffuse/specular split	—	Most scenes
use_reflections	Reflection-direction encoding	—	Glossy scenes
use_n_dot_v	Angle of incidence input	—	Most scenes
use_fresnel_term	Explicit Schlick Fresnel	—	Shiny surfaces
enable_pred_roughness	Predict roughness [0,1]	use_reflections	Gloss variation
use_specular_tint	Colored specular	—	Metals only
specular_exclude_geo_features	Purely view-dependent specular	use_diffuse_color	Uniform plastic
use_roughness_gated_specular	Gate specular by (1-roughness)	enable_pred_roughness + use_diffuse_color	Plastic
use_roughness_in_color_mlp	Feed roughness to color MLP	enable_pred_roughness	Complex appearance
learned_specular_scale	Per-point specular intensity	use_diffuse_color	Variable specularity
roughness_blend_space	"encoding" or "direction"	enable_pred_roughness + use_reflections	Tuning

These flags give you Ref-NeRF–style signals and biases inside an SDF-based model, but the underlying rendering is still fully learned by an MLP — there is no explicit microfacet BRDF, env lighting, or guaranteed physical correctness.