Path Tracing Demo

This demo implements a unidirectional Monte Carlo path tracing technique using hardware accelerated ray tracing support in Unity. There is no rasterization based rendering in the demo and a render pipeline is not used (the Camera doesn't render any geometry).

Prerequisites

• Windows 10 version 1809 and above.
• Any NVIDIA GPU starting from GTX 1060 with 6 GB of VRAM. AMD 6000 series GPUs based on RDNA 2 and above will also run the demo.
• Unity 6.3 and above.

Recommendations

• Use winver in a command prompt to see exactly which Windows version you are using.
• NVIDIA RTX 2000 or AMD Radeon 6000 series GPUs and above are recommended because they have hardware units dedicated to ray tracing. A GTX 1060 GPU was 10 to 15 times slower in tests because ray tracing support is emulated and runs entirely on compute units.
• Alternatively use a tool like GPU-Z to check if your system supports ray tracing. The demo will print an error in the console window if ray tracing is not supported.

Setup and Interaction

When in Play Mode, hold right mouse button down and use WASD keys to navigate through the scene.

The image is built by progressively averaging samples across frames, so it converges only while the scene and view hold still. Anything that would change the result discards the accumulated radiance and restarts the average from the first sample. Convergence resets when:

• any material property is edited — this includes only materials used by Renderers in the acceleration structure;
• an object is added, removed, or moved;
• a light is added, removed, or changed (color, type, direction, range, or position);
• the camera moves, rotates, or changes its field of view;
• the environment map is replaced or its texture contents are updated;
• Space is pressed (manual reset);
• the opaque or transparent bounce count changes;
• Debug Single Bounce is toggled, or Debug Bounce Index changes while it is enabled;
• Debug Validate is toggled;
• the render target is resized (the window or camera resolution changes).

Enable Debug Single Bounce in the inspector to isolate the radiance gathered at a single bounce; Debug Bounce Index selects which one, where 0 is the primary ray hit. Every other bounce contributes black, which is useful for inspecting how each bounce builds up the final image.

Enable Debug Validate to locate the source of non-finite samples. A pixel whose sample is NaN or Inf is painted bright magenta instead of being folded into the accumulated average, where one bad sample would otherwise poison the pixel for the rest of convergence. The flag is sticky per pixel until convergence restarts, so even a source that only fires on occasional samples lights up exactly the pixels it originates from rather than spreading. It is a development aid; the renderer otherwise prevents non-finite values at their source.

Materials

Only materials derived from PathTracingStandard.shader and PathTracingStandardGlass.shader are supported by the path tracer. These are the only shaders that provide a RayTracing pass.

PathTracingStandard.shader exposes a Surface Type setting in the material inspector:

• Opaque — every triangle is a solid occluder. The acceleration structure marks these submeshes closest hit only, so no any hit shader runs and traversal is at its fastest.
• Cutout — the albedo texture alpha is compared against the Alpha Cutoff threshold per pixel. Texels below the cutoff are skipped, leaving a hard edged hole; texels at or above it shade as opaque. This drives the ALPHATEST_ON keyword, which enables an any hit shader that samples the albedo alpha and calls IgnoreHit() below the cutoff. The cutout applies to camera rays and shadow rays alike, so foliage and fences cast correctly perforated shadows.

A Double Sided toggle controls back-face culling. When off, the surface is single sided: rays that strike a back face pass straight through, matching Cull Back in the editor preview. When on, the toggle enables the DOUBLE_SIDED_ON keyword, which RayTracingInstanceTriangleCullingConfig.optionalDoubleSidedShaderKeywords registers as double sided, so the acceleration structure traces both faces and the closest hit shader flips the shading normal on back-face hits. Enable it for thin geometry such as leaves and cloth that needs to be lit from either side.

Assigning a Normal Map texture enables tangent space normal mapping on both PathTracingStandard.shader and PathTracingStandardGlass.shader — the NORMAL_MAP_ON keyword follows the texture assignment, with no separate toggle (emission works the same way on the Standard shader: EMISSION_ON is enabled whenever the emission color is above black). The closest hit shader fetches the vertex tangents, builds an orthonormal frame around the interpolated normal (the bitangent sign accounts for both the mesh handedness and negatively scaled instances), and perturbs the shading normal with the sampled map; a Scale slider stretches or flattens the perturbation. The glass shader perturbs the outward normal and applies its front/back orientation flip afterwards, so a ray refracting through picks up exactly opposite normals on entry and exit, and it offsets continuation rays along the unperturbed geometric normal so self intersection stays robust.

On the opaque Standard shader, two further corrections keep strong perturbations artifact free:

• The BRDF contribution is multiplied by the terminator shadowing factor of Chiang et al. (2019): it leaves the look untouched while the light is above the shading normal and ramps the contribution smoothly to zero at the surface horizon, replacing the harsh black band that perturbed shading normals produce at the shadow terminator. The interpolated vertex normal serves as the reference surface, so the factor corrects only the deviation added by the normal map and stays continuous across triangle edges — referencing the per triangle face normal instead would make the tessellation of coarse smooth meshes visible as faceting. The factor is applied to both the next event estimation sample and the sampled bounce direction, so direct and indirect lighting stay consistent.
• The specular lobe is evaluated and sampled with a consistent shading normal (Keller et al. 2017): the normal is bent, view dependently, so the mirror reflection of the view direction stays above the surface. Without it, strong perturbations reflect view rays into the surface; those samples are rejected and show up as black patches in sharp reflections. The reference surface is the interpolated vertex normal here too, for the same continuity reason.

Direct lighting

Unity lights placed in the scene are sampled with next event estimation (NEE): at every opaque surface hit one light is picked uniformly at random, the BRDF is evaluated in its direction, and a single shadow ray tests visibility.

Supported types:

• Directional — Unity Light components with Type = Directional, with color light.color.linear * light.intensity. It is sampled using a cone for soft penumbra shadows, or set the K_DIRECTIONAL_ANGULAR_DIAMETER define to 0 for a pure delta light with sharp shadows.
• Point — Unity Light components with Type = Point, with color light.color.linear * light.intensity.

References

• Trowbridge, T. S., & Reitz, K. P. (1975). Average irregularity representation of a rough surface for ray reflection. JOSA 65(5). — Original GGX normal distribution.
• Schlick, C. (1994). An Inexpensive BRDF Model for Physically-Based Rendering. Computer Graphics Forum 13(3). — Fresnel approximation.
• Walter, B., Marschner, S. R., Li, H., & Torrance, K. E. (2007). Microfacet Models for Refraction through Rough Surfaces. EGSR 2007. — GGX in a rendering context; rough-refraction BTDF.
• Burley, B. (2012). Physically-Based Shading at Disney. SIGGRAPH 2012 Course: Practical Physically Based Shading in Film and Game Production. — Perceptual smoothness → α = (1 − s)² convention.
• Frisvad, J. R. (2012). Building an Orthonormal Basis from a 3D Unit Vector Without Normalization. Journal of Graphics Tools 16(3).
• Heitz, E. (2014). Understanding the Masking-Shadowing Function in Microfacet-Based BRDFs. Journal of Computer Graphics Techniques 3(2). — Smith G1 and height-correlated G2.
• Heitz, E., & d'Eon, E. (2014). Importance Sampling Microfacet-Based BSDFs using the Distribution of Visible Normals. EGSR 2014.
• Duff, T., Burgess, J., Christensen, P., Hery, C., Kensler, A., Liani, M., & Villemin, R. (2017). Building an Orthonormal Basis, Revisited. Journal of Computer Graphics Techniques 6(1). — Branchless tangent-space basis.
• Keller, A., Wächter, C., Raab, M., Seibert, D., van Antwerpen, D., Korndörfer, J., & Kettner, L. (2017). The Iray Light Transport Simulation and Rendering System. arXiv:1705.01263. https://arxiv.org/abs/1705.01263 — Shading normal adaptation: the specular shading normal is bent so the reflected view direction stays above the reference surface (ComputeConsistentShadingNormal).
• Kulla, C., & Conty, A. (2017). Revisiting Physically Based Shading at Imageworks. SIGGRAPH 2017 Course: Physically Based Shading in Theory and Practice. — Multiple-scattering energy compensation that restores the energy single-scatter GGX loses at high roughness (SpecularEnergyCompensation, baked into g_EnergyCompLUT).
• Heitz, E. (2018). Sampling the GGX Distribution of Visible Normals. Journal of Computer Graphics Techniques 7(4). — VNDF importance sampling.
• Wächter, C., & Binder, N. (2019). A Fast and Robust Method for Avoiding Self-Intersection. In Ray Tracing Gems, Chapter 6. Apress. https://www.realtimerendering.com/raytracinggems/ — Adaptive ray origin offset along the geometric normal that scales with the magnitude of the hit coordinate (OffsetRayOrigin).
• Chiang, M. J.-Y., Li, Y. K., & Burley, B. (2019). Taming the Shadow Terminator. SIGGRAPH 2019 Talks. https://doi.org/10.1145/3306307.3328172 — Shadowing factor that removes the harsh shadow terminator caused by normal mapped shading normals (ShadowTerminatorTerm).
• Turquin, E. (2019). Practical multiple scattering compensation for microfacet models. Unity Technologies. — The 1 + F_avg (1 − E_ss)/E_ss multiplier form used for the specular energy compensation.
• van Antwerpen, D. G. (2023). Solving Self-Intersection Artifacts in DirectX Raytracing. NVIDIA Developer Blog. https://developer.nvidia.com/blog/solving-self-intersection-artifacts-in-directx-raytracing/ — Precise barycentric position interpolation and a deferred translation in the object to world transform for stable world space hit points.
• ITU-R Recommendation BT.709. Parameter values for the HDTV standards for production and international programme exchange. — Rec. 709 luminance weights.
• Ertl, O. (2010). Numerical Methods for Topography Simulation. PhD thesis, TU Wien, §5.3.4, eq. (5.53). https://www.iue.tuwien.ac.at/phd/ertl/node100.html — normalize(N + random_unit_vector) cosine-weighted hemisphere sampling.
• Veach, E. (1997). Robust Monte Carlo Methods for Light Transport Simulation. PhD thesis, Stanford University. — Next event estimation foundations.
• Lagarde, S., & de Rousiers, C. (2014). Moving Frostbite to Physically Based Rendering 3.0. SIGGRAPH 2014 Course: Physically Based Shading in Theory and Practice. — Smooth windowed inverse square attenuation for punctual lights.