Software & Data Downloads — SAC-GNC

SAmple Consensus for Adaptive Graduated Non-Convexity for testing and benchmarking various GNC approaches (including the most recent implementations) and a range of robust loss functions.

This publication provides the accompanying code for the paper titled “SAC-GNC: SAmple Consensus for Adaptive Graduated Non-Convexity”, presented at the IEEE International Conference on Computer Vision (ICCV 2025) in Honolulu. Graduated Non-Convexity (GNC) typically relies on fixed annealing factors to update the shape parameter of a robust loss function within a robust least squares formulation. This paper introduces a novel method that adaptively anneals the shape parameter within the GNC framework and proposes new stopping criteria, along with an initialization technique that improves performance across diverse datasets. In addition to the SAC-GNC implementation, we provide a framework for testing and benchmarking various GNC approaches (including the most recent implementations) and a range of robust loss functions