MicroSDF: microfacet-driven neural SDFs for mixed-reflectance 3D reconstruction
ssh4net · x · 2026-10-15
MicroSDF is a neural implicit framework using microfacet theory to reconstruct geometry and reflectance of mixed reflective/non-reflective surfaces. Key ideas: multi-scale normals from an SDF with microfacet normal consistency loss, normal-variance-based specular detection, and detection-guided blending of dual color branches—yielding notably better geometry on complex reflective surfaces than prior dual-branch methods.
More from Research
- RLVR fine-tuning collapses solution space, replay buffer fix boosts diversity and pass@8 — manoelribeiro · 2026-10-15
- ModeBench shows RLVR collapses LLMs into single-solution mode; Re:Max training preserves diversity — manoelribeiro · 2026-10-15
- NYU's SCOOP makes LLMs better explorers with a strategist-solver-optimizer loop — jaseweston · 2026-10-15
- Google's WeatherNext outforecast NHC physics models on Hurricane Isaias, adding ~1 day of lead time — PeterWBattaglia · 2026-10-15
- MIT shows simple local synaptic rules let neural networks build their own learning machinery — ChuckDBrooks · 2026-10-15
- VASPO Combines Self-Distillation and Privileged Value Estimation to Boost PPO Reasoning — ustc-community · 2026-10-15