|
I developed a demo
of shadow maps for my fourth-term directed focus
study at the Guildhall. I researched,
implemented, and compared three separate shadow
mapping algorithms involving various spatial and
filtering methods. In addition to standard
shadow maps, I implemented trapezoidal shadow
maps for improved spatial accuracy and variance
shadow maps to facilitate Gaussian blurring for
soft shadows.
Initially, I
planned to implement perspective shadow maps
instead of trapezoidal shadow maps.
However, I found this technique to be poorly
documented, and after spending several days
struggling with it, I chose to implement
trapezoidal shadow maps instead, as it was not
only well-documented, but also supposedly a
better technique with fewer problem cases to
be resolved.
An unexpected
and unfortunate result of combining trapezoidal
and variance shadow maps can be seen to a degree
in the screenshots below, and becomes very
apparent in scenes with a dynamic camera.
Because the blur effect is performed in texture
space, the perceived amount of blur changes
depending on the camera's position, orientation,
and field-of-vision relative to the light
source. This is not an issue with standard
shadow maps because it simply means that shadows
further from the light source will appear to
have blurrier edges, which is actually a
desirable effect.
Another problem
I discovered with the variance shadow mapping
technique is that, even when blurred, the
shadows will never exceed their initial
silhouettes. This means that the blurring
effect can only go "inwards" from the silhouette
edges, which can make the shadow appear too
skinny or make it disappear completely depending on the
size of the object casting it.
This directed
focus study was presented at the Guildhall
Summer Exhibition on June 16, 2006. |
