For the algorithm, we started from 20 LDR images of increasing brightness (that I created using HDRshop from Paul Debevec's HDRs).Basically, we took bright regions from dark images, and the dark regions from the bright images, and midranges on the images in between... and put them into one LDR image. It worked fairly well but not better than state of the art methods.

I thought of the subdivision scheme independently - but unfortunately it already had a name... simplest. The simplest algorithm smooths out the cube(left) and the result isn't very smooth (middle). After two iterations of the simplest algorithm (right), the edge topology is the same as Doo-Sabin subdivision.

Dlfldemo uses the dlfl (doubly-linked face list) data structure to elegantly handle topology changes. It makes it very easy to create models that have complex topology. To see the models that I created with the program click here or on the images below.

Dlfldemo was written by Vinod Srinivasan. It was well coded and easy to add to. For more info, see Vinod's page about the program.