Stochastic Ray Tracing

Stochastic Ray Tracing in action.

Stochastic raytracing accommodates several different methods like path tracing, final gathering and a photon map second pass. Several options influence accuracy of the solution (and required computation time) and can be tuned for particular types of scenes.

Pixel sampling

• Samples Per Pixel: Determines the number of rays per pixel that are used to compute the image. The same number is used for each pixel. Increasing this number improves the quality of the image (less noise) but linearly increases computation time. Set this to a huge number and go on a holiday if you want the perfect image...
• Progressive Image Refinement: If set pixels are not handled scanline by scanline but progressively (32x32 blocks first, then 16X16, 8X8, ...) This does not influence quality of the final image...

Radiance Contributions

• No storage use: Do not use precomputed info
• Direct visualization: Use precomputed info directly, meaning that everywhere, also where eye rays hit, the solution is used.    Specular or glossy components not present in the stored solution are still computed using stochastic raytracing. (This is like a standard radiosity-raytracing second pass)
• Indirect visualization: Only use the stored solution after a diffuse or glossy bounce was already encountered. (Similar to final gathering)
• Photon map specific: special second pass for the photon map. Caustic map is used. (Cfr. Jensen, EGWR 96)

Light sampling

• Explicit Light Sampling: If on, explicit shadow rays are sent to some or all light sources. If not on, you will need a lot of samples/pixel to get a decent image...
• Light samples: the number of shadow rays used. The actual  total number depends on the mode (button below)
• Light mode:

Power based: randomly choose lights based on their emitted power

Importance based: randomly choose lights based on their (non-occluded) contribution to the point under consideration

All lights: send shadow rays to all the lights (light samples * number of lights) shadow rays used.

Surface scattering

• Particles per Scatter: The 'spawning factor'. This determines how many reflected/refracted rays are used to compute scattered    incoming light. Beware that number of traced rays may explode if set too high (1 eye-ray -> 10 scattered rays -> 100 scattered indirect rays -> ...  wait forever)
• Different for first D|G scatter (Toggle): If set, a different spawning factor is used for the first diffuse/glossy bounce. This improves indirect diffuse/glossy illumination without having an exponential ray explosion.
• Particles For This Scatter: The spawning factor for the first D|G scatter.
• Samples specular separately: Normally when a scattered ray is traced a choice is made whether to use specular/glossy or diffuse scattering. When this option is set, specular reflection and refraction are    always traced separately, which is beneficial for mixed surfaces  (diffuse/glossy and specular). The sharp specular reflections will be better. This option requires more computation.
• BSDF Sampling: Use (normal) sampling based on a full bi-directional scattering distribution function (bsdf). Note that the photon map uses a bit different bsdf model with infinitely sharp specular scattering and Fresnel  factors (-> nicer caustics...)
• Classical Sampling: Use classical raytracing sampling. Glossy and specular components are treated as perfect mirror/refractors. ONLY direct diffuse illumination is computed as in the real classical raytracing, except of course when precomputed radiance is used.
• Minimum Path Length: Normally russian roulette is used to stochastically terminate paths. This option determines how long the path must be BEFORE russian roulette is used.
• Maximum Path Length: Limits the length of traced light paths and as such the number of interreflections calculated. (Length = number of segments in the path)