Recent Advances in Adaptive Sampling and Reconstruction for Monte Carlo Rendering

Monte Carlo integration is firmly established as the basis for most practical realistic image synthesis algorithms because of its flexibility and generality. However, the visual quality of rendered images often suffers from estimator variance, which appears as visually distracting noise. Adaptive sampling and reconstruction algorithms reduce variance by controlling the sampling density and aggregating samples in a reconstruction step, possibly over large image regions. In this paper we survey recent advances in this area. We distinguish between “a priori” methods that analyze the light transport equations and derive sampling rates and reconstruction filters from this analysis, and “a posteriori” methods that apply statistical techniques to sets of samples to drive the adaptive sampling and reconstruction process. They typically estimate the errors of several reconstruction filters, and select the best filter locally to minimize error. We discuss advantages and disadvantages of recent state-of-the-art techniques, and provide visual and quantitative comparisons. Some of these techniques are proving useful in real-world applications, and we aim to provide an overview for practitioners and researchers to assess these approaches. In addition, we discuss directions for potential further improvements.

Denoising Your Monte Carlo Renders: Recent Advances in Image Space-Adaptive Sampling and Reconstruction

With the ongoing shift in the computer graphics industry toward Monte Carlo rendering, there is a need for effective, practical noise-reduction techniques that are applicable to a wide range of rendering effects and easily integrated into existing production pipelines. This course surveys recent advances in image-space adaptive sampling and reconstruction algorithms for noise reduction, which have proven very effective at reducing the computational cost of Monte Carlo techniques in practice. These approaches leverage advanced image-filtering techniques with statistical methods for error estimation. They are attractive because they can be integrated easily into conventional Monte Carlo rendering frameworks, they are applicable to most rendering effects, and their computational overhead is modest.

Image Space Adaptive Rendering

In this thesis, we develop an adaptive framework for Monte Carlo rendering, and more specifically for Monte Carlo Path Tracing (MCPT) and its derivatives. MCPT is attractive because it can handle a wide variety of light transport effects, such as depth of field, motion blur, indirect illumination, participating media, and others, in an elegant and unified framework. However, MCPT is a sampling-based approach, and is only guaranteed to converge in the limit, as the sampling rate grows to infinity. At finite sampling rates, MCPT renderings are often plagued by noise artifacts that can be visually distracting. The adaptive framework developed in this thesis leverages two core strategies to address noise artifacts in renderings: adaptive sampling and adaptive reconstruction. Adaptive sampling consists in increasing the sampling rate on a per pixel basis, to ensure that each pixel value is below a predefined error threshold. Adaptive reconstruction leverages the available samples on a per pixel basis, in an attempt to have an optimal trade-off between minimizing the residual noise artifacts and preserving the edges in the image. In our framework, we greedily minimize the relative Mean Squared Error (rMSE) of the rendering by iterating over sampling and reconstruction steps. Given an initial set of samples, the reconstruction step aims at producing the rendering with the lowest rMSE on a per pixel basis, and the next sampling step then further reduces the rMSE by distributing additional samples according to the magnitude of the residual rMSE of the reconstruction. This iterative approach tightly couples the adaptive sampling and adaptive reconstruction strategies, by ensuring that we only sample densely regions of the image where adaptive reconstruction cannot properly resolve the noise. In a first implementation of our framework, we demonstrate the usefulness of our greedy error minimization using a simple reconstruction scheme leveraging a filterbank of isotropic Gaussian filters. In a second implementation, we integrate a powerful edge aware filter that can adapt to the anisotropy of the image. Finally, in a third implementation, we leverage auxiliary feature buffers that encode scene information (such as surface normals, position, or texture), to improve the robustness of the reconstruction in the presence of strong noise.

Robust Denoising using Feature and Color Information

We propose a method that robustly combines color and feature buffers to denoise Monte Carlo renderings. On one hand, feature buffers, such as per pixel normals, textures, or depth, are effective in determining denoising filters because features are highly correlated with rendered images. Filters based solely on features, however, are prone to blurring image details that are not well represented by the features. On the other hand, color buffers represent all details, but they may be less effective to determine filters because they are contaminated by the noise that is supposed to be removed. We propose to obtain filters using a combination of color and feature buffers in an NL-means and cross-bilateral filtering framework. We determine a robust weighting of colors and features using a SURE-based error estimate. We show significant improvements in subjective and quantitative errors compared to the previous state-of-the-art. We also demonstrate adaptive sampling and space-time filtering for animations.