Final Gathering

Final gathering is a technique for estimating global illumination for a given point by either sampling a number of directions in the hemisphere over that point (such a sample set is called a final gather point), or by averaging a number of final gather points nearby since final gather points are too expensive to compute for every illuminated point.

For diffuse scenes, final gathering often improves the quality of the global illumination solution. Without final gathering, the global illumination on a diffuse surface is computed by estimating the photon density (and energy) near that point. With final gathering, many new rays are sent out to sample the hemisphere above the point to determine the incident illumination. Some of these rays hit diffuse surfaces; the global illumination at those points is then computed by the material shaders at those sample point, using illumination from the globillum photon map if available and other material properties. Other rays hit specular surfaces and do not contribute to the final gather color (since that type of light transport is a secondary caustic). Tracing many rays (each with a photon map lookup) is very time-consuming so it is only done when necessary - in most cases, interpolation and extrapolation from previous nearby final gatherings is sufficient.

Final gathering is useful in scenes where the indirect illumination various slowly, such as purely diffuse scenes. For such scenes, final gathering eliminates photon map artifacts such as low frequency noise and dark corners. Fewer photons are needed in the globillum photon map and lower globillum accuracy is sufficient since each final gather averages over many values of indirect illumination.

In film production work, final gathering is often used rather than photon mapping, except for caustics. Final gathering by default supports a single bounce only. Multiple-bounce effects tend to have far less impact on the final image than the first bounce; these are performed by photons by default and by final gathering only if the shaders adjusts the trace depth. Although physical correctness is lost, this is often "good enough" for film production, and final gathering is easier to control than photons emanating from distant light sources. However, for accurate indoor illumination simulations and other CAD-related applications, photon mapping is preferred.

Final gathering is set off by default, but can be turned on in the options. To change the number of rays shot in each final gather (and optionally the maximum distance at which a final gathering result can be used for interpolation and the minimum distance at which it must be used), specify a finalgather accuracy setting in the options. For example,

finalgather accuracy 1000 1.5 0.25

increases the number of rays and reduces noise in scenes with complex illumination and geometry. The default number of rays is 1000. The default maximum distance depends on the scene extent; decreasing it will reduce noise but increase render time. The default minimum distance is 10% of the maximum distance.

mental ray 3.4 has changed the final gathering algorithm. It normally achieves better quality in less time with approximately one-half the number of rays. It is necessary to modify the accuracy settings though, for example changing 1000 to 500 in the preceding example, or rendering times will increase.

3.8 Final gathering can be combined with importance-based techniques like Importons and Irradiance Particles, to achieve better quality with less artifacts in shorter rendering time. In these cases, importons are shot before FG, which are then used to control the distribution of FG rays according to "importance".

Final Gathering Modes

mental ray offers the following final gathering modes:

"automatic"

This final gathering mode offers greatly simplified usage compared to version 3.4. It no longer requires radii to control final gathering sampling and interpolation. The parameters used to control this mode are:

number of finalgather rays (the first numeric argument of "finalgather accuracy"),
"finalgather points", the approximate number of points used for the interpolation.

"multiframe"

This mode is similar to the "automatic" mode, but has been specifically designed for rendering of camera fly-through animations. To avoid flickering, the rendering should be split into two parts. In the first part, the final gather map is rendered for the set of key frames or for a coarse sequence of frames. In the second part, the frozen final gathering mode is used for the rendering of the full animation. It is possible that some part of the scene will not contain sufficient number of finalgather points for some frames not belonging to the key frame set. To avoid picking up illumination from distant objects which is possible in the automatic mode, the finalgather accuracy max radius limits the maximal distance. If a sufficient number of final gather points is not found within that distance, the illumination will be faded to black smoothly.

This mode is typically useful for animated scenes, but is not necessarily the best method for animations where flickering is not a problem. It may also be well suitable for still images without animation. The parameters used to control this mode are:

number of finalgather rays (the first numeric argument of "finalgather accuracy"),
min and max radius (secondary arguments of "finalgather accuracy"),
optional "view" flag of the finalgather accuracy,
"finalgather points", the approximate number of points used for the interpolation.

"3.4"

This mode is provided for compatibility with mental ray version 3.4 settings, which gives the experienced user additional control over the final gathering computations. It renders results similar to the older version of mental ray but benefits from improvements of the current implementation. It and makes use of all "finalgather accuracy" arguments (optional "view" flag, number of finalgather rays, max and min radii).

"strict 3.4"

This mode is similar to the "3.4" mode, but achieves higher compatibility with the older mental ray version, allowing identical re-rendering of old scenes with the current mental ray. Some rendering improvements are disabled in that mode.

"force"

This mode disables the final gather point cache at all, and will always compute the accurate solution. This takes time but will yield superior results.

The mode can be selected by a scene option or on the mental ray command line.
The default is "3.4" compatibility mode.