In this project, we reimplemented a stratified autocalibration algorithm, based on using convex optimization to produce efficient bounds in a branch and bound approach, which claims significantly better results than the standard Levenberg-Marquardt based approach. The algorithm
solves for convex relaxations of the affine and metric upgrade equations. We evaluated this implementation similarly to the manner in which the describing article did, using randomly generated scenes and comparing to a reference implementation.Code for this project