Linear Model Analysis of Real-Time Rendering - Real-Time Rendering: Computer Graphics with Control Engineering - page 29

FIGURE 3.4

(See colour insert.) Screenshot of application in Experiment 1.

shading of the 3D object in the application (Figure 3.3). From a causal system per-
spective, this variable is selected because of its independence from the geometry
input variable. No tight coupling exists between the shader complexity variable used
in the raster stage of the rendering pipeline and the input geometry typically used in
the pre-rasterisation stage.
In the data collection process in Experiment 2, we manipulated the vertex input
to the rendering process by changing the tessellation segments and the shader com-
plexity values at various tessellation levels. The corresponding frame rate changes
were registered. Of the 14,000 frames collected, 12,000 frames were used for
model derivation and the remainder for validation. From Experiment 2, the derived
model allowed us to extrapolate the framework to support multiple inputs in more
complicated rendering processes.

3.5.3 e xPeRiment 3: c ontRol f RamewoRk u sing s ystem m odel
The objective was to construct a simple control framework using a system model
derived from the aforementioned modelling process. We selected another appli-
cation from the DirectX toolkit that used a progressive mesh control mechanism
(Figure 3.5). The technique is similar to geometry tessellation and has been adopted
widely in many interactive graphics applications to achieve fine resolution control
of a 3D object's geometry. After a model of the rendering process was derived, we
introduced the concept of a controller to manage the input to the rendering pro-
cess to produce a rendering framework that offered stability and conformed to a
user-defined frame rate.
In Experiment 3, we collected 60,000 frames of data; 50,000 were used for model
derivation and the remainder for validation of the system model. The input data