Home

Awesome

Demo---Consensus-maximization-tree-search-revisited

A significantly accelerated tree search method for globally optimal consensus maximization. (Paper link)

Published in ICCV 2019 as oral presentations.

About

alt text

Consensus maximization is an effective tool for robust fitting in computer vision. A* Tree Search is one of the most efficient methods for globally optimal consensus maximization. In this work, we propose two new techniques that significantly accelerate A* Tree Search, making it capable of handling problems with much larger number of outliers.

This demo is free for non-commercial academic use. Any commercial use is strictly prohibited without the authors' consent. Please acknowledge the authors by citing:

@article{cai2019consensus,
  title={Consensus Maximization Tree Search Revisited},
  author={Cai, Zhipeng and Chin, Tat-Jun and Koltun, Vladlen},
  journal={arXiv preprint arXiv:1908.02021},
  year={2019}
}

in any academic publications that have made use of this package or part of it.

Contact

Homepage:https://zhipengcai.github.io/

Email: czptc2h@gmail.com

Do not hesitate to contact the authors if you have any question or find any bugs :)

Getting Started

This demo is implemented using MATLAB 2018b and has been tested on Ubuntu 14.04 LTS 64-bit.

Run the demo

  1. Clone this repository.

  2. Run "demo.m" in MATLAB.

Please refer to "demo.m" file for detailed code explanations.

List of addressed problems in the demo

Linear problem:

  1. Linearized Fundamental matrix estimation (ignoring the rank-2 constraint)

Nonlinear problem (the code of this part can handle problems with pseudo-convex residuals (see the example forms in the paper) ):

  1. Homography estimation

List of included methods

Previous A* tree search variants:

  1. A* tree search (Chin et al. CVPR'15)

  2. A* tree search + True Outlier Detection (TOD) for branch pruning (Chin et al. TPAMI'17)

Variants with our new techniques:

  1. A* tree search + Non-Adjacent Path Avoidance (NAPA)

  2. A* tree search + NAPA + TOD

  3. A* tree search + NAPA + Dimension-Insensitive Branch Pruning (DIBP)