Blue Noise Dithering for Reservoir-based Spatio-temporal Importance Resampling.

Area of Science:

• Computer Graphics
• Image Synthesis
• Computational Imaging

Background:

• Monte Carlo path tracing is standard for photorealistic images but suffers from high variance.
• Reservoir-based Spatiotemporal Importance Resampling (ReSTIR) offers efficient sample reuse in real-time path tracing.
• ReSTIR's stochastic resampling conflicts with optimal sample distributions like blue noise.

Purpose of the Study:

• To reinterpret ReSTIR through dithered sampling and connect it to blue-noise sampling.
• To develop a blue-noise-driven spatiotemporal resampling scheme for improved perceptual quality.
• To integrate this scheme into a real-time path tracer for efficient rendering.

Main Methods:

• Reinterpreted ReSTIR as a dithered sampling process.
• Established a link between reservoir resampling and ordered blue-noise sampling.
• Developed and integrated a blue-noise-driven spatiotemporal resampling method into a real-time path tracer.

Main Results:

• The proposed method preserves perceptually favorable high-frequency error distribution during sample reuse.
• Integrating blue noise into ReSTIR significantly improves perceptual quality for low-sample-count images.
• Quantitative evaluations show reduced perceptual errors with minimal runtime overhead.

Conclusions:

• Blue-noise-driven spatiotemporal resampling enhances ReSTIR for real-time path tracing.
• The method maintains rendering efficiency suitable for real-time applications.
• This approach offers a viable solution for improving image quality in demanding graphics scenarios.