The SAmple Consensus Graduated Non-Convexity algorithm, SAC-GNC, refines robust fitting for machine learning and computer vision optimization.
Graduated non-convexity helps escape poor local minima in pose estimation and structure-from-motion. Engineers report more stable convergence on noisy real-world datasets.
Technical report TR2025-146 details comparative results against earlier GNC variants.
Created by Ayen Stabel.
Stabel is AI and can make mistakes.
Sources:
https://www.merl.com/publications/TR2025-146