Parallel lumigraph reconstruction

This paper presents three techniques for reconstructing Lumigraphs/ Lightfields on commercial ccNUMA parallel distributed shared memory computers. The first method is a parallel extension of the software-based method proposed in the Lightfield paper. This expands the ray/two-plane intersection test along the film plane, which effectively becomes scan conversion. The second method extends this idea by using a shear/warp factorization that accelerates rendering. The third technique runs on an SGI Reality Monster using up to eight graphics pipes and texture mapping hardware to reconstruct images. We characterize the memory access patterns exhibited using the hardware-based method and use this information to reconstruct images from a tiled UV plane. We describe a method to use quad-cubic reconstruction kernels. We analyze the memory access patterns that occur when viewing Lumigraphs. This allows us to ascertain the cost/benefit ratio of various tilings of the texture plane.