Accelerating animation through verification of shooting walks

Abstract

Until recently, research on global illumination and photon maps in particular considered only static scenes. As hardware capabilities and acceleration techniques allow for more rapid image generation, the application of these methods for animation becomes more relevant. Techniques that retain more information about rendered frames are able to exploit coherence between frames to a larger extent. Storing all of the shooting walks, as does the photon map, is a good basis for generating a modified image more rapidly. In this paper, we describe a fast, but conservative method to verify the stored shooting walks. As a result, the cost of updating of the modified walks becomes comparable to the cost of verification, making nonconservative methods and their inherent assumptions unnecessary. In order to achieve this, we use a separation of dynamic and static objects that allows for occlusion tests limited to potentially moving objects.

SATO: Surface Area Traversal Order for Shadow Ray Tracing

“…In terms of future work, we would like to combine the SATO with other methods and measure the effects of the combination. First, if static and dynamic scenes are handled separately [SBSK03, KNPY13], the SRDH can be used for high‐quality static tree/subtree construction, as well as TO determination, while the SATO can be used for dynamically updating trees or subtrees. Additionally, we are interested in combining the SATO with lazy tree construction [DHW*11] and AAC BVH construction [GHFB13].…”

“…In partial updates, only dynamic parts of the scene are updated. These methods either use a separate dynamic tree [SBSK03] or multi‐level hierarchies [KNPY13]. The lazy tree construction method [DHW*11] computes the subtree associated with a node only when a ray enters that node.…”

Filtering After Shading With Stochastic Texture Filtering

“…Although lighting integrals are evaluated stochastically, their integrands are usually evaluated analytically. Integrands that are themselves stochastic have been used for complex BSDF models [Guo et al 2018;Heitz et al 2016], multi-lobe BSDF evaluation [Szécsi et al 2003] and for many light sampling [Estevez and Kulla 2018;Shirley et al 1996].…”

On the Fast Construction of Spatial Hierarchies for Ray Tracing

“…However, this solution is not suited for the unstructured motion of individual object primitives. In the same year, Szecsi et al [37] proposed an interesting extension to the "sequential" method of Parker et al They suggest to construct two kd-trees, one kd-tree over static data and one kd-tree over dynamic objects. Instead of traversing the trees in sequential order and later combining results, they propose to traverse the two trees simultaneously by a special traversal algorithm over two trees.…”