Game Development Reference
In-Depth Information
Figure 1.6. DSMs without filtering (left) exhibit resampling artifacts. Smoother shad-
ows can be achieved with PCF (center), while ESM provides even higher quality (right).
In Table 1.1 we compare the timings for a special version of binary shadow
mapping (simulated by using only the first fragment of a DSM for shading),
DSMs, and DSMs using a 3
5 PCF kernel size, respectively, and
show the comparison for several shadow map resolutions. The comparison to this
version of binary shadow mapping demonstrates the overhead of the DSM lookups
(using both optimizations, i.e., neighbor links and truncation of the transmittance
function). While the overhead is more pronounced for small shadow maps, it
becomes small in relation to the DSM creation for increasing shadow map size.
In Figure 1.6 we compare the quality of DSMs without filtering (left), with
PCF (center), and using ESM (right). As can be seen, PCF performs solid
antialiasing, while EMS improves the rendering quality even more. Furthermore,
while all DSM methods require a depth bias to avoid Z-fighting artifacts, ESM
needs significantly less bias for artifact-free rendering than unfiltered and PCF
We presented an optimized implementation of deep shadow maps for complex hair
models that achieves real-time frame rates by employing new features of current
graphics hardware. Note that our implementation requires only Direct3D 11
shader features and compute shaders, which makes our algorithm attractive in
environments where GPUs from different vendors are used. In our experiments
it turned out that the best DSM quality can be achieved by combining it with
ESM. While interactive applications are the main target for this algorithm, we
also see applications in the movie industry, where such a real-time DSM imple-
mentation could save valuable production time and provide immediate feedback
to the artists.
Search Nedrilad ::

Custom Search