This is a Matlab toolkit that provides the necessary modules to construct a typical bag of words pipeline for object recognition and categorisation purposes.

It is also an evaluation toolkit for the comparison of the different algorithms that comprise the pipeline modules.

Version 0.1 (05/2014)

• Jose Rivera (jose.rivera@imperial.ac.uk)
• Ioannis Alexiou (i.alexiou@qmul.ac.uk)
• Anil Bharath (a.bharath@imperial.ac.uk)

Biologically Inspired Computer Vision Group

Web: http://www.bicv.org

SIFT, DSIFT, VLAD and kernel implementations require VLFEAT (http://www.vlfeat.org/) Clustering requires kmeans_bo.m by Liefeng Bo (http://research.cs.washington.edu/istc/lfb/). LibSVM is used to implement SVM classification (http://www.csie.ntu.edu.tw/~cjlin/libsvm/). LLC_coding_appr.m is part of MATLAB's LLC package (http://www.ifp.illinois.edu/~jyang29/LLC.htm)

A downloadable version of the required libraries and code can be found in the Downloads section of the repository. Link to libraries:

This toolkit can be used with any dataset. However, a 4-object version of Caltech-101 is included in the 'Downloads' section of the repository. Link to 4-object Caltech-101.

1. Run setup.m for the installation of the 3rd party libraries.
2. Run main.m

A beginners version of the toolkit, consisiting of a simple Hard Assignment bag-of-visual-words model for object classification is included as BOVWdemo.m

0. Download and unzip the bovw_pipeline_lib.zip from the Downloads section.
1. Run setup.m
2. Run BOVWdemo.m