Fast Ranking via Metric Learning (FRML-1.1) AUTHORS: Daryl Lim <dklim@ucsd.edu> This code is distributed under the GNU GPL license. See LICENSE for details or http://www.gnu.org/licenses/gpl-3.0.txt INTRODUCTION ------------ This package contains the MATLAB code for Fast Ranking via Metric Learning (FRML). The latest version of this software can be found at the URL above. The software included here implements the algorithm described in Lim, D.K.H., and Lanckriet, G.R.G. Efficient Learning of Mahalanobis Metrics for Ranking. Proceedings of the 31st International Conference on Machine Learning (ICML), 2014 Please cite this paper if you use this code. INSTALLATION ------------ 1. Requirements This software requires MATLAB R2007a or later. Because it makes extensive use of the "bsxfun" function, earlier versions of Matlab will not work. If you have an older Matlab installation, you can install an alternative bsxfun implementation by going to http://www.mathworks.com/matlabcentral/fileexchange/23005, however, this may not work, and it will certainly be slower than the native bsxfun implementation. INCLUDED FILES -------------- get_sim_diff.m - Generates similar/dissimilar index sets for use with frml_warp.m for various cases. frml_demo.m - Demo script that shows how to use frml_warp. mlr_test_largescale.m - Script to evaluate retrieval performance using a low-rank/factored metric IN300folds.mat - Data from 3 classes from ImageNet, split into folds ready for use with the algorithm. Each image is encoded using vector quantization with SIFT features. README - This file Documentation for each function is provided within each file. FEEDBACK -------- Please send any queries to Daryl Lim <dklim@ucsd.edu>.