 |
|
Fig.: Time Control Unit of the InVis-System
The quality of each pipeline-step is scalable in order to produce interactive frame rates. Every step can be controlled by a parameter, wich determines quality and influences the execution time.
A lookup-table with estimated start-values for the parameter ensures that the pipeline terminates quickly enough at the first iteration.
The parameter influences execution-time of each step nearly in a linear way. There may be an oscillation effect because of dependencies within the pipeline-steps (for example, fewer reduction in the data reduction step causes longer rendering time): First the measured time is too long and therefore the parameter is decreased, then the measured time is too short and the parameter is increased and so on... But this oscillation effekt is very low, the system is designed to converge to a constant value as soon as possible.
Time-Control Consequences on Pipeline-Steps:
data acquisition |
There is fixed in this first step of the pipeline which part of information will be shown and which will be omitted. Now data consist of a visible and an invisible part. If there has to be shown less the following steps will run faster, of course. |
data reduction |
Volume information can be changed into a scalable form, for example into
multiresolution wavelets. If there is less execution time left data will be supplied in a
lower resolution. This reduction saves time at all following pipeline-steps, but the
disadvantage of this will be a coarser visualization. Geometric objects are devided into two groups: polygon objects (triangles) and CSG objects. The vertices can be combined at polygons with a lower curvature. This reduces the number of vertices in a significant way and also reduces the execution time of all following pipeline-steps. More reduction causes coarser visualization. It is generally not necessary for CSG objects to apply data reduction algorithms. |
visibility transformation |
This stage can only be skipped in order
to speed up the visualization. In this case, a standard color and transparency will be
applied. If the view point does not change much during a rendering sequence, a hidden object removal algorithm can be applied: Previously hidden objects are removed until there will be more time left. |
viewing transformation & rendering |
"Progressive refinement"-techniques allow much influence on rendering time. First, a coarse visualization is created, which can be manipulated by the user. Later, if no user action happens, a refined image will be calculated and shown in hierarchical time steps. |