There were four papers presented on points at the conference, each of which dealt with handling large numbers of points or re-sampling polygonal data for representation with points. Considering the use of large scanned data sets, this area of study is very important to current practice in film.

Combining Edges and Points

This paper discussed the introduction of EPI, the Edge and Point Image representation. The basic idea is to improve the handling of sharp features by representing them using edges, while using points to represent data efficiently within the edge boundaries.

Space Modeling with Point-sampled Geometry

This work, done at ETH, uses a hybrid point clouds representation to balance the use of implicit and explicit surfaces when representing data. By balancing these two methods, it can provide the advantages of implicit surfaces (like fast boolean operations) with some of the advantages of explicit surfaces (like fast visibility and bounding box calculations). The work will be available this fall as part of Pointshop3D in source and object form.

Interactive Boolean Operations on Surfel-Bounded Solids

The goal of this paper is to do boolean operations with arbitrary free-form solids. By composing the solids of surfels (small disks that have position normal and radius), the system is able to quickly determine whether a point is inside or outside of the surface. Further, as operations are performed, surfels can be split and have their radii reduced in order to accurate represent the new shape without substantial movement.

It's still a bit unclear to me the benefit of surfels as opposed to using a triangle mesh in this case, except for the (rather significant) lack of need for connectivity of the mesh. Read the paper, maybe you can explain it to me. However, it was a very fast and effective technique.

Sequential Point Trees

Sequential Point Trees are intended to take the benefits of a QSplat and implement them so that the GPU can compute them. This implementation uses vertex programming to reposition the vertex being evaluated to infinity if the triangle is not intended to be rendered. Further, by taking advantage of the visible distance sorting on the CPU, only potentially visible points are sent to the GPU in the first place, further increasing efficiency.