Efficient-Kernel-XQDA-Python

Paper: Cross-View Kernel Similarity Metric Learning Using Pairwise Constraints for Person Re-identification

Accepted in Indian Conference on Computer Vision, Graphics and Image Processing (ICVGIP) 20-21, Jodhpur, India T M Feroz Ali, Subhasis Chaudhuri https://arxiv.org/abs/1909.11316

This repository contains the complete code of Efficient Kernel Cross-view Quadratic Discriminant Analysis (EK-XQDA). Using this code you can reproduce our result in Table 1 (CUHK01 dataset) of our paper. GOG + k-XQDA : R1 = 62.23% R5 = 83.09% R10 = 89.46% R20 = 94.43%

Code setup:

1. You need to download the GOG features for CUHK01 dataset (available at http://www.i.kyushu-u.ac.jp/~matsukawa/ReID_files/GOG_CUHK01.zip) and place the following files inside the folder 'Features': a) CUHK01_feature_all_GOGyMthetaHSV.mat b) CUHK01_feature_all_GOGyMthetaLab.mat c) CUHK01_feature_all_GOGyMthetanRnG.mat d) CUHK01_feature_all_GOGyMthetaRGB.mat

2. Edit config.m file: Chage the path 'directry' according to the location of code in your system.

3. Run demo_EK_XQDA.m

If you find this work useful, please kindly cite our paper.

@article{ali2019cross, title={Cross-View Kernel Similarity Metric Learning Using Pairwise Constraints for Person Re-identification}, author={Ali, TM and Chaudhuri, Subhasis}, journal={arXiv preprint arXiv:1909.11316}, year={2019} }