BVH Refit in Reverse DFS Order for Real-Time Ray Tracing

The patent segments the rendering process into two distinct phases: a rendering phase that generates raw rendered pixels with full computational accuracy, and a denoising phase that processes these pixels to produce final clean images. This segmentation allows the computationally intensive ray tracing to be separated from the post-processing denoising operations, enabling efficient resource utilization across distributed systems.

The patent introduces rendered pixels as an intermediary data structure that bridges the rendering phase and denoising phase. These rendered pixels serve as a compressed representation of the scene that can be efficiently transmitted and processed across distributed devices, mediating between the high computational cost of ray tracing and the need for efficient distributed processing.

The patent performs preliminary rendering actions on individual devices to generate rendered pixels before the denoising process begins. By pre-computing the rendered pixels and making them available for subsequent denoising operations, the system establishes a clear data flow sequence that simplifies coordination across distributed devices and enables efficient parallel processing.

The patent discards the need for complex inter-device coordination during the rendering phase by allowing each device to independently render pixels that are then recovered and processed collectively during the denoising phase. This approach simplifies the distributed system architecture by separating independent rendering operations from coordinated denoising operations.