LoBCoD

A Local Block Coordinate Descent Algorithm for the CSC Model

This is the Matlab package that implements the LoBCoD algorithm.

E. Zisselman, J. Sulam and M. Elad, "A Local Block Coordinate Descent Algorithm
for the Convolutional Sparse Coding Model". CVPR 2019.

[paper] [supp] [arXiv]

Citing LoBCoD

@InProceedings{Zisselman_2019_CVPR,
    author = {Zisselman, Ev and Sulam, Jeremias and Elad, Michael},
    title = {A Local Block Coordinate Descent Algorithm for the CSC Model},
    booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
    month = {June},
    year = {2019}
}

Installation

Download the package to a local folder (e.g. ~/LoBCoD/) by running:

git clone https://github.com/EvZissel/LoBCoD.git

Open Matlab and navigate to the folder (~/LoBCoD/).

Dependencies

This code uses the following packages:

vlfeat - An open library of computer vision algorithms.
SPAMS optimization toolbox - For its implementation of the batch LARS algorithm.

For Windows

This code is self-contained and includes all the precompiled packages.

Description

This package contains the following main modules:

Module	Description
LoBCoD.m	The main function that implements the batch LoBCoD algorithm
Demo.m	A demo script that applies the function `LoBCoD.m` on the Fruit dataset
LoBCoD_online.m	A function that implements the online LoBCoD algorithm
Demo_online.m	A demo script that applies `LoBCoD_online.m` on a subset of mirflickr dataset
inpainting_LoBCoD.m	A function that implements inpainting using `LoBCoD.m`
Demo_inpainting.m	A demo script that applies `inpainting_LoBCoD.m`
Demo_Multi_Focus_Fusion.m	A demo script for implementing multi-focus image fusion
Demo_Multi_Exposure_Fusion.m	A demo script for implementing multi-exposure image fusion
Demo_Text_Image_Denoising.m	A demo script for implementing salt-and-pepper text image denoising

Examples

The training curves of the LoBCoD algorithm using Demo.m (trained on the Fruit dataset):

The converging objective value of the test set using Demo_online.m (trained on a subset of mirflickr dataset):

Example of inpainting of the corrupted Barbara image using Demo_inpainting.m:

Example of multi-focus image fusion of the images Bird (background and foreground in-focus) using Demo_Multi_Focus_Fusion.m:

Example of multi-exposure image fusion of the images Window using Demo_Multi_Exposure_Fusion.m:

Example of salt-and-pepper text image denoising using Demo_Text_Image_Denoising.m:

wangjinjia1/LoBCoD