Temporal-Gradient Illumination Resampling for Ray-Traced Denoising
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing rendering algorithms based on ray tracing and light transport simulation produce noisy images with lagging shadows and streaks from moving lights, and current denoisers react slowly to abrupt changes, leading to suboptimal noise reduction.
Innovation Solution
Compute temporal gradients using ReSTIR reservoirs instead of random numbers, comparing lighting information between consecutive frames to determine confidence values for history rejection and blending, using a bilateral blur process to smooth lighting effects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If spatiotemporal filters accumulate lighting information over multiple frames, then noise reduction is improved, but the output signal reacts slowly to abrupt changes causing lagging shadows and streaks
Solution Approach 1:
The patent implements dynamic history rejection by computing temporal gradients and using confidence values to adaptively control the blending between historical and current frame data. The denoiser transitions from static accumulation to dynamic adjustment based on detected lighting changes, allowing fast response to abrupt changes while maintaining noise reduction quality during stable periods.
Solution Approach 2:
The system computes temporal gradients as feedback signals to detect lighting changes between frames. This feedback mechanism enables the denoiser to identify when lighting conditions have changed and adjust its behavior accordingly, rejecting historical data when changes are detected and accumulating data when conditions are stable.
2Stability of the object's composition
If random number sequences are reused for gradient computation, then temporal consistency is improved, but false positive gradients occur due to persistent state changes in light reservoirs
Solution Approach 1:
The patent extracts only the essential temporal gradient information needed for confidence computation, separating it from the problematic persistent state in light reservoirs. By computing gradients based on ReSTIR reservoirs rather than reusing random number sequences, the system obtains temporal consistency without the false positives caused by reservoir state changes.
Solution Approach 2:
The patent introduces ReSTIR reservoirs as an intermediary between frame-to-frame comparisons. These reservoirs provide a stable basis for gradient computation that is less sensitive to scene changes, acting as a mediator that enables reliable temporal gradient calculation without directly comparing volatile random number sequences.
3Reliability
If modern denoisers use heuristic methods to mitigate lag effects, then some improvement is achieved, but the effects are only somewhat effective and noise artifacts remain
Solution Approach 1:
The patent changes the fundamental parameter used for controlling denoising behavior from heuristic rules to computed temporal gradients. By using mathematically derived gradient values and confidence scores, the system achieves more effective lag mitigation without requiring complex heuristic logic, simplifying the algorithm while improving performance.
Data Source
AI summary
Systems and methods described relate to the generation of image content. In order to provide for smoothing between sequential images, but avoid introducing lag into lighting effects, light information can be compared for regions between consecutive rendered frames. Shading can be performed and the results compared for tiles of pixels to compute gradient values, such as by using a single light sample for each tile. A filtering pass can be performed with respect to these gradients, and this filtered, lower-resolution grid version can be upscaled into a full resolution, screen-sized image and the gradients transformed into confidence values. These confidence values can be used to determine an extent to which to keep lighting data from the previous frame with respect to the current frame. For example, less lighting information can be used from the prior frame for a given pixel location if the confidence for that location is lower.