Deep light clustering for denoising Monte Carlo renderings

Abstract

Monte Carlo (MC) path tracing has been widely used to synthesize realistic images. However, it takes extensive time to render a high-quality image since it requires to sample numerous light paths for each pixel. Hence, MC image reconstruction methods have been actively studied to remove rendering noises and recover clean images. This study proposes a light path clustering framework to further improve MC image reconstruction. Though image-space features (e.g., surface normal, depth, texture maps) have significantly contributed to MC denoising, direct utilization of high-dimensional light paths has not yet been sufficiently explored. This paper proposes a contrastive manifold learning framework that reduces the dimensionality of path space for MC reconstruction models to exploit path-space features effectively. Conventional contrastive approaches utilize discrete data labels to discriminate and cluster the data. Yet, discrete and exact labeling for light paths remains ill-defined due to the continuity and complexity of path space. We circumvent this issue by weakly-supervised learning

we use reference pixel colors for continuous pseudo labeling. We apply our framework to the recent reconstruction models and demonstrate considerable improvements. This thesis is written on the basis of the journal published papers in which the current candidate participated as a first author [12].