Paper: Silhouette Shadows

At the Stanford Graphics Lab, they implemented a new way to enhance the quality of shadow-map-generated shadows. In addition to rendering the shadow map (giving the depth, as seen from the light, of each pixel), they render a silhouette map, which gives information on the sub-pixel position of the silhouette edge as seen from the light -- i.e., the shadow-casting edge. This map in essence enables the shadow map to be projected as a piecewise continuous linear edge rather than a raster-sampled grid.

The result is a considerably improved quality for shadow edges (see the illustrations from the paper for more details). However, the implementation on even GeForce FX class hardware is slowed down considerably due to the difficulty of implementing some of the calculations. Nevertheless, their framerates for single-character scenes were in the 10 fps range.

The full title of this paper was "Shadow Silhouette Maps" by Pradeep Sen, Mike Cammarano, and Pat Hanrahan from Stanford University.