HdRestir 1 — Resampled Importance Sampling
Resampled Importance Sampling changes the question from “which proposal should generate this sample?” to “which candidate should survive?” HdRestir draws candidates from multiple proposals, assigns scalar weights, and selects one without losing the full spectral contribution it represents.
The article derives the estimator before opening the implementation: candidate generation, target functions, categorical selection, normalization, support conditions, regularization, and the real cost of increasing the candidate count.
