Some different scanline tracer

The result of Irradiance Gradients is very dependent on the used scanline tracer.
If the tracer scans the picture too regular, like you can see at the first image, there
are some strange artefacts due to extrapolation.
To improve the visibility of the difference of the next four pictures, these images
are ray traced only with diffuse lighting.

calulated with a standard (scanline) tracer 
calculated with a PSR (Pixel-Selected Ray-Tracing) tracer 
rendered with an uniformly distributed random tracer 
calculated with an interlace tracer, with an interlace 
step of 8


Numerical Problems in Radiance (with irradiance gradients)

rendered with direct and diffuse lighting (with Radiance) 
Rendered with Radiance and only diffuse lighting, you can see some 
dark spots in the corner. 
The cause for that is, that some rays shot through the wall because of 
numerical problems of the ray tracing algorithm. 

Solution (implemented in Vision):

As you see here, the starting point P is too near at the 
right plane and because of this the Ray S2 shots through 
that plane. 
The solution to that problem is, to move the starting point P 
a little bit away from that plane (in the opposition direction 
of the incoming ray S). As a result, the distance between P0 
and the right plane is now big enough, so no ray will shot 
through that plane.


A still existing problem: diffuse lightsources

The irradiance gradients algorithm cannot detect diffuse lightsources
very good.
Normally, this diffuse lightsource, as seen is the second image, throws
a hard shadow on the left and upper wall. But because of the interpolation
of different directly calculated irradiance values, this hard line is smoothed.

Direct lighting by a point light covered by a box 
only open to the left 


Direct and diffuse lighting of that box - you can see 
some artefacts here, which appear due to the big 
difference in lighting inside and outside of the box. 
Because of  that the possible error of Monte Carlo 
sampling is high.
One improvement of the image with the existing Irradiance Gradients
algorithm would be, to increase the number of rays sent. But you would
need a pretty large number of additional rays to really improve the
calculation with this solution.
Another, and maybe better solution would be, to use the images, created
by the hemicube sampling, to detect these critical regions in the scene.
Then it is only necessary to increase the number of rays in that regions.

Back to the  Main Page