Optimized shift mapping operators for rendering

Abstract

Realistic physically-based rendering relies on efficient light transport simulation, for which expensive path tracing algorithms are the only effective solution. This cost arises from the large number of path samples required to generate a usable image, which makes it infeasible for real-time applications. ReSTIR performs spatiotemporal path reuse, drastically reducing the required number of samples to within the realm of modern graphical hardware. Its efficiency is conditional on the quality of its shift mappings, which transform paths between the distinct path domains of different pixels and frames. However, current shift mappings do not consider the quality of their source samples, resulting in reuse of non-contributing samples and increased noise. In this thesis, I present an optimized shift mapping operator based on the machinery of stochastic texture filtering, which significantly increases the quality of path samples available for spatiotemporal path reuse.

Date	10 Aug 2026
Original language	American English
Awarding Institution	École de technologie supérieure
Supervisor	Adrien Gruson (Supervisor) & Derek Nowrouzezahrai (Co-supervisor)