Neural-Guided RANSAC
NG-RANSAC for Epipolar Geometry from Sparse Correspondences
Introduction
Neural-Guided RANSAC (NG-RANSAC) is a general method for fitting parametric models to a set of data points that might contain outliers and noise, i.e. it is a robust estimator. This code provides an implementation of NG-RANSAC for fitting epipolar geometry, i.e. a fundamental matrix or an essential matrix, to a set of sparse correspondences between a pair of images.