Multi-pass Density Estimation for Infrared Rendering

Pacific Graphics 2004 (submitted)

We compute several little particle maps. All these particle maps are global in opposition to the localised particle maps dependent on the geometry of [13] which means that each particle map stores particles within the entire model. This ensures that the illumination is spread correctly over the scene for each pass and thus that density estimations are correct.We benefit from the speedup of computing multiple small particle maps shown in [13] without limitation on the mesh complexity because the particle maps are not dependent on the geometry. Unfortunately, when the image size is large (i.e 1024x768 pixels), the time saved while computing and sorting the particle map is lost when rendering due to the number of density estimations needed. Therefore, it is more interesting to compute fewer passes with bigger particle maps when rendering big images. The memory used is constant from one pass to another.