Please note: This PhD seminar will take place online.
Weijie Zhou, PhD candidate
David R. Cheriton School of Computer Science
Supervisor: Professor Toshiya Hachisuka
Recent progress in Monte Carlo rendering has highlighted the importance of statistically grounded methods for combining noisy and biased estimates. “Neural James–Stein Combiner for Unbiased and Biased Renderings” introduces a data-driven extension of the James–Stein estimator that fuses unbiased path-traced images with biased denoised results, achieving lower error than either source alone. Building upon this idea of optimally balancing bias and variance, “DSCombiner: Double Shrinkage for Combining Biased and Unbiased Monte Carlo Renderings” further explores shrinkage-based combination strategies, showing that multiple shrinkage stages can more effectively suppress noise and improve image quality. I will introduce these two papers in the seminar.