Matlab Wrappers for wide-baseline matching and 3-D reconstruction.
Matlab code contains:
(a) RepMatch pose and correspondence; (b) CODE matching; (c) Robust Pose estimation;
(d) Wrappers for sending match information to Visual SfM. Warning it is extremely slow. We are working on the speed now.
(2) CODE: Coherence based Decision Boundaries for Feature Matching
Code for illumination robust image alignment, written with Liu Linlin.
(4) Robust Non-parametric Data Fitting for Correspondence Modeling
Code for robust non-linear regression that provides the basic engine for many other papers.
(5) Dimensionality's Blessing: Clustering Images by Underlying Distribution
Code for clustering data by their distribution.
Bits and Pieces
Video by Wang Fan on integrating matching with Visual SfM.
Interface to "Let there be color!", written by Wang Fan and used in the above video,
Robust two view pose estimation. Difference from conventional
techniques: (a) Uses a larger minimal set, 20+ instead of 5 points. The larger point set gives
additional stability and leverages on RepMatch's very clean matches. (b) A different bundle
adjustment formulation which does not require 3-D points as variables. (c) A number of iterative steps
to ensure stability. See (1) for code.