Portfolio Model for Adjusting Level-Of-Detail

When there are multiple objects in the scene and the rendering engine cannot render all of them at their best quality, objects are often rendered at a reduced level of detail (LOD). It is not uncommon to model an object with different numbers of polygons. A penguin viewed from afar needs less polygons than when it is close to the camera.


figure 1: measuring risk for a penguin and a lamp

Two problems with using multiple levels of detail are (1) determining who gets reduced or improved and (2) when this change happens. Frequent changes to objects cause them to jitter or "pop" as they go from one level to another. This can be visually distracting.

Portfolio Model
The idea with a portfolio model is to model each object as a stock. Its value is based on how "important" it is -- in this case, how much visual angle it consumes and how close to the center of the scene. Since the camera and objects move, this importance value fluxuates over time. We call this variance or risk.

The portfolio model compares the change in importance over time to evaluate its variance or risk. Using these measures, the system can calculate what objects are returning the most value and which ones are more risky. Combining these numbers produces a risk curve. Setting the system to low-risk means that objects won't change their level-of-detail much, but they may never acheive their highest resolution. By setting the risk high, objects try to achieve their best resolution, but risk popping a lot as they are forced to vary LOD more.

For the developer, this model reduces the problem to optimizing the risk curve and/or setting a single risk number regardless of the number of objects in the scene.

Detailed paper coming soon.