This is a repository for the matrix factorization project published in ICCV 2015.

• Je Hyeong Hong (jhh37@outlook.com)
• Andrew Fitzgibbon (awf@microsoft.com)

Clone the repository using the following code

git clone --recursive https://github.com/jhh37/matrix-factorization

Below is a list of algorithms included.

1. [JE] Our implementation of DRW (Damped Ruhe and Wedin) algorithms
2. [JE] Our Ceres implementation of joint optimisation algorithms
3. [CH] Chen's LM_S, LM_M series (our modification)
4. [CO] Chen's LM_S, LM_M series (original: http://sist.sysu.edu.cn/~chenpei/)
5. [DB] Del Bue's BALM (our implementation but not necessarily fast)
6. [NB] Boumal's RTRMC (original: https://perso.uclouvain.be/nicolas.boumal/RTRMC/)
7. [PG] Gotardo's CSF (original: http://www2.ece.ohio-state.edu/~gotardop/)
8. [RC] Cabral's ALM series (our implementation but not necessarily fast)
9. [TO] Okatani's Damped Wiberg (DW) (original: http://www.vision.is.tohoku.ac.jp/us/download/)

You may run the script [ run_demo.m ] inside the [ Code ] folder.

Before starting, there are some issues you need to watch out for:

1. Prior to running any experiment, you must have run [ Code/setup.m ], which adds relevant paths and compile required mex files.
2. All commands must run in the [ Code ] directory.
3. To run our Ceres-based algorithms (assuming Ceres-solver from http://ceres-solver.org/ is already installed), you need to compile the source (which is pulled as a submodule) present in [ Code/Algorithms/JE/lrmf_ceres ]. 1. The CMake file is given. 2. It may be convenient to compile in the [ Code/Algorithms/JE/lrmf_ceres/build ] directory since this is git-ignored. 3. Place the compiled executable inside the [ Code/Algorithms/JE ] folder with the name [ lrmf_ceres_exec ] or [ lrmf_ceres_exec.exe ]. 4. Reading sparse matrices is currently not supported by [ LRMF-Ceres ].

Our 1-min video spotlight can be found at https://youtu.be/Kk0NIeNnc5M.

• The work was supported by Microsoft and Toshiba Research Europe.
• We thank Roberto Cipolla, Christopher Zach, Bamdev Mishra and the anonymous reviewers for their comments.
• The conference travel was funded by the British Machine Vision Association (BMVA), Cambridge University Engineering Department (Rex Moir Fund), Christ's College (University of Cambridge) and Cambridge Philosophical Society.