QSplat compression

The great advances in the field of 3D scanning technologies have enabled the creation of meshes with hundred millions of polygons. Rendering data sets of that size is time consuming even with commodity graphics hardware. The QSplat technique that has been introduced by S. Rusinkiewics and M. Levoy of Stanford University is used for the inter-active point based visualization of large 3D scenes. Nevertheless, it has some drawbacks like the storage requirement which is still higher. The objective of our work we present in this paper is to improve the per-node storage requirements of QSplat models and to minimize the transmission cost in streaming QSplat models across low-bandwidth networks or bottlenecked networks. To do that, we focus on coding strategies which provide reasonable data reduction at low decoding complexity. In this context, Huffman and relative delta encoding fit well with our purposes. The performances of the compression process are studied and the rendering algorithm is extended in order to be able to work on compressed data without loosing the original system interactivity.