We propose a geometric antithetic sampling strategy for efficiently rendering spatiotemporally modulated light, encompassing continuous-wave time-of-flight (CW-ToF) and structured light systems. This is motivated by observing that modulated light transport can be decomposed into a slowly varying geometric term and a rapidly oscillating modulation term, where the latter dominates variance. Our key idea is to construct antithetic sample pairs by geometrically deforming a light path so that the modulation term changes sign while preserving similarity in the underlying transport. In both CW-ToF and structured light, antithetic samples lie on an antithetic manifold—the set of configurations with opposite modulation—and we search along it to find the path most similar to a given reference path, thereby maintaining strong negative correlation between paired samples. For CW-ToF, this corresponds to shifting paths in optical path-length space, while for structured light, it involves deforming emitter samples along the spatial modulation manifold in projector-coordinate space. By unifying both cases under this geometric deformation framework, our method cancels high-frequency modulation components through paired samples, leading to significant variance reduction and enabling more efficient simulation of coded illumination and sensing systems in computational imaging.
@inproceedings{kim26antithetic,
author = {Kim, Juhyeon and Cui, Jack and Jarosz, Wojciech and Pediredla, Adithya},
title = {Geometric Antithetic Sampling for Rendering Spatiotemporally Modulated Light},
booktitle = {ACM SIGGRAPH Asia Conference Papers},
year = {2026},
month = dec,
doi = {10.1145/3829340.3842310},
pubstate = {Accepted}
}