Converging Algorithm-Agnostic Denoising for Monte Carlo Rendering

Abstract

Denoising Monte Carlo rendered images is a critical challenge in computer graphics, traditionally addressed in post-processing phases. Recent advances have shifted focus towards integrating denoising directly into progressive rendering. This approach not only blends denoised outputs with noisy inputs to enhance visual quality but also guides adaptive sampling through variance estimation of denoised outputs. In this paper, we introduce a novel method that predicts the blending weight of denoised images directly during progressive rendering. Our technique adjusts the blending weight for each pixel based on error estimates, effectively 'skipping' certain iterations of Monte Carlo Path Tracing (MCPT) and mimicking adaptive sampling a posteriori. A key innovation of our approach is an analytical method that ensures the blended output converges accurately to the reference image. Our method does not rely on deep-learning techniques, making it immediately applicable to any denoising algorithm. We demonstrate that our method enhances visual quality and allows for blending between noisy and denoised images, even those obtained at different MCPT iterations. This not only streamlines the rendering process but also improves efficiency and output fidelity.