Optimized geometry compression for real-time rendering

Most existing visualization applications use 3D geometry as their basic rendering primitive. As users demand more complex data sets, the memory requirements for retrieving and storing large 3D models are becoming excessive. In addition, the current 3D rendering hardware is facing a large memory bus bandwidth bottleneck at the processor to graphics pipeline interface. Rendering 1 million triangles with 24 bytes per triangle at 30 Hz requires as much as 720 MB/sec memory bus bandwidth. This transfer rate is well beyond the current low-cost graphics systems. A solution is to compress the static 3D geometry as an off-line pre-process. Then, only the compressed geometry needs to be stored in main memory and sent down to the graphics pipeline for real-time decompression and rendering. The author presents several new techniques for compression of 3D geometry that produce 2 to 3 times better compression ratios than existing methods. They first introduce several algorithms for the efficient encoding of the original geometry as generalized triangle meshes. This encoding allows most of the mesh vertices to be reused when forming new triangles. Their second contribution allows various parts of a geometric model to be compressed with different precision depending on the level of details present. Together, the meshifying algorithms and the variable compression method achieve compression ratios of 30 and 37 to one over ASCII encoded formats and 10 and 15 to one over binary encoded triangle strips. The experimental results show a dramatically lowered memory bandwidth required for real-time visualization of complex data sets.