Conclusions

The presented results were generated using a multi-threaded implementation written in C++ under Windows NT. The following table shows the memory requirement for some models:

Model	Triangles	Memory
Terrain	39.300	42 MB
Bunny	69.451	64 MB
Crater	199.369	186 MB

Most scenes with one or more objects achieve good results. The models can be displayed in almost arbitrary fine simplification levels by adjusting the threshold for the Screenspace Error. The user can choose the level of simplification by adjusting the threshold and obtains an appropriate frame rate or visual quality. Figure 8 shows a model of the human foot bones in two simplification levels. The upper figure shows the model without simplification and the figure below shows how the algorithm is able to simplify the model by changing the topology (non-manifold). It can be seen, that certain bones grow together.

  

Figure 8: Non mani-fold Simplification

Using a model from the Crate Lake, USA (Figure 9) the performance of the multi-threaded implementation is shown. The test was executed on a Dual-Pentium III 600 MHz system with 256 MB RAM and a NVidia Erazor² under Windows 2000 (resolution 1600x1200):

  

Figure 9: Model of the Crater Lake, USA (199k polygons)

The dual-thread simplification runnning on a dual-processor-system achieves almost twice the framerate as the single-thread variant. Compared to the rendering without simplification a significant higher frame rate is reached, although the visual quality is reduced almost unnoticeable. Although the number of rendered triangles is reduced below 20%, the visual quality remains high. The model was displayed to be completely visible and to fill the display window as much as possible.

Model: Crater
Testcase	Error	Triangles	avg. framerate
Without simplification	-	199k	1.8
Single-thread simplification	1%	30k	11
Dual-thread simplification	1%	30k	21

Problems occur with scenes with several closely located but separated objects (Figure 8). The vertices of different objects are not distinguished during fold operations. For high levels of simplification (drastic simplification) this effect is welcome, because the whole scene is displayed very simplified and it is allowed that close object merge. If the effect occurs at high resolutions (slight simplification), the human eye is very sensitive to such simplification artefacts. Improvements are possible by implementing a silhouette preservation based on a Cone of Normals [9] or an advanced error metric like a Quadrics Error Metric [2].