Massively Parallel Path Space Filtering

Restricting path tracing to a small number of paths per pixel in order render images faster rarely achieves satisfactory image quality for scenes interest. However, space filtering may dramatically improve the visual by sharing information across vertices classified as proximate. Unlike screen approaches, these neither need be present on screen, nor is restricted first intersection with scene. While searching proximate had been more expensive than space, we greatly over this performance penalty storing, updating, and looking up required hash table. The keys are constructed from jittered quantized information, such that only single query very likely replaces costly neighborhood searches. A massively parallel implementation algorithm demonstrated graphics processing unit (GPU).