HOME | DD

Description
Rendered with a multi-spectral photon mapper I'm currently working on. There is half a million photons in this scene, about 7000 final gathers, 2000 rays each. 40 photons were used in each final gather. Direct lighting uses 128 uniformly randomized shadow rays.

The total rendering time was 4 hours, most of which was spent calculating the indirect term.

There is a rectangular light source at the ceiling as in this reference picture: [link]

All photons have uniform power. Russian roulette is used everywhere instead of power scaling, in order to make density estimation simple.

Photons are shot from the light source using the light's spectral emission data as the probability density function. In the end, density estimation converts the local light spectrum to CIE XYZ, changes the illuminant to D65, transforms to linear RGB and finally to sRGB for display.

Ward's irradiance caching with translational gradients is used to speed up the computation, about 30 times with the particular settings used to generate this image. A kD-Tree with Havran's TA-B-rec traversal algorithm (a modified version of Jacco Bikker's implementation) and Wald's tri-ray intersection algorithm is used at the core.

The photon mapper is written in the D programming language and uses the Tango library.

There are some errors near edges, due to the lack of secondary final gathering. Anti-aliasing and stratified shadow ray sampling is TODO as well.

Description
Rendered with a multi-spectral photon mapper I'm currently working on. There is half a million photons in this scene, about 7000 final gathers, 2000 rays each. 40 photons were used in each final gather. Direct lighting uses 128 uniformly randomized shadow rays.

The total rendering time was 4 hours, most of which was spent calculating the indirect term.

There is a rectangular light source at the ceiling as in this reference picture: [link]

All photons have uniform power. Russian roulette is used everywhere instead of power scaling, in order to make density estimation simple.

Photons are shot from the light source using the light's spectral emission data as the probability density function. In the end, density estimation converts the local light spectrum to CIE XYZ, changes the illuminant to D65, transforms to linear RGB and finally to sRGB for display.

Ward's irradiance caching with translational gradients is used to speed up the computation, about 30 times with the particular settings used to generate this image. A kD-Tree with Havran's TA-B-rec traversal algorithm (a modified version of Jacco Bikker's implementation) and Wald's tri-ray intersection algorithm is used at the core.

The photon mapper is written in the D programming language and uses the Tango library.

There are some errors near edges, due to the lack of secondary final gathering. Anti-aliasing and stratified shadow ray sampling is TODO as well.