← All resources

Quadric-Based Silhouette Sampling for Differentiable Rendering

Auto-discovered

Paper Advanced

Mariia Soroka, Christoph Peters, Steve Marschner

Visit resource ↗

Description

Quadric-Based Silhouette Sampling for Differentiable Rendering presents a SIGGRAPH 2025 ACM TOG paper by Mariia Soroka, Christoph Peters, and Steve Marschner on importance sampling silhouette edges for physically based differentiable rendering.

Geometry derivatives require integrating over silhouette discontinuities, but prior tree-based edge sampling admitted many zero-contribution samples. The authors propose a tighter rejection test using dual-space bounding boxes and dual quadrics over mesh patches, plus improved stochastic traversal heuristics. In a unidirectional path tracer, the method reduces variance versus the original edge sampler and outperforms reported area-sampling baselines.

It is for researchers and engineers working on inverse rendering and differentiable path tracers. Background in physically based rendering and differential light transport is expected.

The author page summarizes the abstract and links the official TOG version and interactive supplemental results, noting the August 2025 SIGGRAPH presentation.

The work helps keep silhouette edge sampling competitive by cutting wasted zero-contribution samples. Link last verified accessible on 2026-09-29. Link last verified accessible on 2026-09-28.