Hierarchical Structures For Dynamic Polygonal Simplification


This was the first paper, to my knowledge, that introduced dynamic view-dependent polygonal simplification. It was submitted to SIGGRAPH 96, but justly rejected for being (1) not well-written enough, and (2) containing "dropouts", or transient artifacts characterized by polygons that disappear for a frame. My second attempt, a year later, addressed both of these concerns and more, and was accepted into SIGGRAPH 97. So though this paper may be of historical interest, I recommend readers interested in view-dependent simplification check out our 1997 SIGGRAPH paper, a better written, more informative, and up-to-date presentation of the algorithms described here.

Developers interested in view-dependent simplification should also check out VDSlib, a public- domain library that implements the latest version of these algorithms.


Still under construction!

To be exact, I never fully webified the paper. Here is a not-quite-finished conversion of the paper. The formatting is a bit strange and the one diagram is missing but you may find it useful. Or you can download the paper as postscript or PDF. Either way you'll want to look at the images below. Clicking on an image will download a (very) high-resolution TIFF file.

In the meantime, here's the abstract:


This paper presents a novel technique for simplifying polygonal environments. The technique is unique from previous multi-resolution methods in that it operates dynamically, simplifying the scene on-the-fly as the user's viewing position shifts, and adaptively, simplifying the entire database without first decomposing the environment into individual objects. Each frame the simplification process queries a spatial subdivision of the model to generate a scene containing only polygons "important" given the current viewpoint. This spatial subdivision, an octree, classifies the polygons of the scene according to which regions of space they intersect. When the volume of space associated with an octree node occupies less than a user-specified amount of the screen, all vertices within that node are collapsed together and degenerate polygons filtered out. An active list of visible polygons is maintained for rendering. Since frame-to-frame movements typically involve small changes in viewpoint, and therefore modify the active list by only a few polygons, the method can take advantage of temporal coherence for greater speed. The algorithm has been implemented and tested successfully on a wide range of models, providing a 2x to 4x increase in rendering performance with only slight degradation of image quality. On larger models, or in situations where greater degradation is acceptable, the algorithm should perform even better.