Adaptive Shadow Denoising in Ray Tracing With Variance-Guided Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional ray tracing techniques for rendering soft shadows require a large number of shadow rays to converge to an accurate result, leading to noisy render data that is inefficient and costly, and existing denoising methods rely on bandwidth-intensive hit distance data or global illumination techniques.
Innovation Solution
Adaptive spatiotemporal filtering that adjusts filter values based on variance in temporally accumulated ray-traced samples, excluding samples outside a defined range and using moments to conserve computing resources, thereby eliminating the need for hit distance data and reducing noise.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a large number of shadow rays are cast to sample lighting conditions, then shadow rendering accuracy is improved, but computing resources and rendering time increase significantly
Solution Approach 1:
The patent applies partial sampling by casting a limited number of shadow rays per pixel rather than exhaustive sampling. This partial action approach accepts some noise in exchange for dramatically reduced computing resources and faster rendering, which is then resolved through temporal denoising accumulation
Solution Approach 2:
The patent performs preliminary denoising by accumulating shadow ray samples across multiple frames before final rendering. By pre-accumulating temporal data and performing denoising in advance, the system reduces the computational burden during real-time rendering while maintaining accuracy
2Productivity
If shadow rays are sparsely sampled to conserve computing resources, then rendering time is reduced, but noise in render data increases
Solution Approach 1:
The patent maintains continuous accumulation of shadow ray samples across multiple frames rather than processing each frame independently. This continuous temporal integration allows sparse per-frame sampling to converge to accurate results over time, maintaining both speed and accuracy
Solution Approach 2:
The patent uses temporal feedback by comparing and accumulating shadow samples across frames. The denoising process uses feedback from previously accumulated samples to guide the filtering of current frame samples, improving accuracy progressively while maintaining low per-frame computational cost
3Measurement precision
If conventional shadow denoising filters use ray-hit distance to guide filtering, then denoising accuracy is improved, but memory bandwidth requirements increase
Solution Approach 1:
The patent extracts and eliminates the dependency on hit distance data for guiding the denoising filter. By removing this requirement, the system significantly reduces memory bandwidth consumption while maintaining effective denoising through alternative guidance mechanisms based on shadow ray visibility patterns
4Measurement precision
If global illumination techniques are used to provide shadowing, then illumination accuracy is improved, but computational cost increases significantly
Solution Approach 1:
The patent uses inexpensive, computationally lightweight shadow ray sampling instead of expensive global illumination techniques. While individual shadow rays provide limited information, their cumulative effect through temporal accumulation achieves accurate shadowing at a fraction of the computational cost of global illumination
Data Source
AI summary
In examples, a filter used to denoise shadows for a pixel(s) may be adapted based at least on variance in temporally accumulated ray-traced samples. A range of filter values for a spatiotemporal filter may be defined based on the variance and used to exclude temporal ray-traced samples that are outside of the range. Data used to compute a first moment of a distribution used to compute variance may be used to compute a second moment of the distribution. For binary signals, such as visibility, the first moment (e.g., accumulated mean) may be equivalent to a second moment (e.g., the mean squared). In further respects, spatial filtering of a pixel(s) may be skipped based on comparing the mean of variance of the pixel(s) to one or more thresholds and based on the accumulated number of values for the pixel.