WaRCO

Introduction

This is an implementation of a multi-class classifier and regressor based on a patch based model termed WARCO (Weighted ARray of COvariance matrices) able to deal with low resolution objects. WARCO is fully implemented in Matlab. The software was tested on different versions of Linux and Windows using Matlab versions R2010a/b (x64) and R2011a(x64). There may be compatibility issues with other versions of Matlab.

Requirements

• Matlab 2010 or higher.
• Several algorithms require the Images toolbox by MathWorks.
• Libsvm http://www.csie.ntu.edu.tw/~cjlin/libsvm/
• Liblinear http://www.csie.ntu.edu.tw/~cjlin/liblinear/
• Piotr Dollar Toolbox http://vision.ucsd.edu/~pdollar/toolbox/doc/

Installation

• Create a directory called 'WARCO' and copy the code there.
• Put all the additional tooboxes (Libsvm, Liblinear, Piotr Dollar Toolbox) in './utils'.

Datasets

• CAVIARShoppingCenterFull
• CAVIARShoppingCenterFullOccl
• HIIT6HeadPose
• HOC
• HOCoffee
• IHDPHeadPose
• QMUL4PoseHeads
• QMUL5PoseHeads

How to start

• Create a the directory WARCO\database and unzip all the datasets in that folder.
• Add the paths of WARCO to Matlab with addpath(genpath(WARCO\path));
• Type help WARCO to know how WARCO is organized.
• Switch to the WARCO folder typing cd WARCO\path;
• Use the test_* scripts to test WARCO. Be sure to set the right data path into the testing scripts.

Reference

• D. Tosato, M. Spera, M. Cristani, V. Murino, Characterizing humans on Riemannian manifolds, IEEE Trans. PAMI, Preprint 2011.