ceres-solver

Why we included this project

Most people doing 3D reconstruction end up fitting camera poses and 3D points from image correspondences with Ceres. It is a mature C++ library: you define a cost function, add residual blocks, and hand the problem to a nonlinear least-squares solver that handles sparse structure, bounds constraints, and a range of algorithms from Gauss-Newton to L-BFGS. Teams working on SLAM, structure-from-motion, camera calibration, or any vision geometry that boils down to minimizing a large residual objective will find a direct modeling API and production-grade performance. It has been in production at Google since 2010, and for bundle adjustment and other large sparse optimization problems it is a dependable, well-documented choice rather than a research prototype.