/person_re_id_trdp

Person Re-Identification System

Primary LanguageMATLABBSD 3-Clause "New" or "Revised" LicenseBSD-3-Clause

Person Re-identification Benchmark

This repository hosts the codebase of Tutored Research Development Project (TRDP) completed by Bithi Barua and Kazi Ahmed Asif Fuad. This modified repository of State of Art "A Systematic Evaluation and Benchmark for Person Re-Identification: Features, Metrics, and Datasets." done by: Karanam, S., Gou, M., Wu, Z., Rates-Borras, A., Camps, O., & Radke, R. J. (2018). A Systematic Evaluation and Benchmark for Person Re-Identification: Features, Metrics, and Datasets. IEEE Transactions on Pattern Analysis and Machine Intelligence, accepted February 2018.

In out TRDP, we, at first, retrained the CNN features(MATLAB Build-in Models) by Transfer Learning Method. As we did not achieve better result compared to SOA. We later tried two pre-processing techniques: Midway Image Equalization and Luminance Remapping on the VIPeR dataset. Our report and presentation slides are available at: https://www.dropbox.com/sh/qodea83s425pg62/AAAQ3kncY3D4E-EInqOFX-5ca?dl=0 We have visualized the images with their corresponding ranking to understand the issues with different re-id approach.

Quick Start for our repository

  • Clone this repository
  • Copy the VIPeR dataset from this link: https://www.dropbox.com/sh/joa1324b1fkk6w3/AACBK5_0JYOWB3FMlZT0NTtXa?dl=0 and copy to DataOriginal folder.The structure should be ./DataOriginal/VIPeR/(cam1,cam2).
  • Run re_id_trdp.mlapp (which requires MATLAB installed with appdesigner)
  • All the results will be showed in cmd in MATLAB and Visual Failcases will be stored in Visual_FailCasesImagaes folder.

Tested on Windows 10 with MATLAB 2018b(Student License)

For Transfer Learning of MATLAB CNN, please follow "ReadMeForTransferLearning.txt".

Author's Repository is available at https://github.com/RSL-NEU/person-reid-benchmark

##Instructions for Author's repository

Quick Start for Author's repository

  • Clone this repository
  • Run a quick example in run_experiment_benchmark.m
  • Read the results for VIPeR dataset with WHOS feature and XQDA

Run other experiments

  • Download supported dataset, unzip it and put it under the folder ./Data
  • Download corresponding partition file and put it under the folder ./TrainTestSplits
  • Run corresponding prepare_DATANAME.m inside the folder ./Data (if avaliable)
  • Change the parameters in run_experiment_benchmark.m

Check List for supported/tested feature

  • HistLBP
  • WHOS
  • gBiCov
  • LDFV
  • ColorTexture\ELF
  • LOMO (Windows)
  • GOG (Windows)

Check List for supported/tested metric learning

  • FDA
  • LFDA
  • kLFDA-linear/chi2/chi2-rbf/exp
  • XQDA
  • MFA
  • kMFA-linear/chi2/chi2-rbf/exp
  • NFST
  • KISSME
  • PCCA-linear/chi2/chi2-rbf/exp
  • rPCCA-linear/chi2/chi2-rbf/exp
  • kPCCA-linear/chi2/chi2-rbf/exp
  • PRDC
  • SVMML
  • kCCA

Check List for supported/tested multi-shot ranking method

  • rnp
  • srid
  • ahisd

Check List for supported/tested dataset

Reference

Please cite the work appropriately for each used feature/metric learning/ranking/dataset

@ARTICLE{8294254, 
author={S. Karanam and M. Gou and Z. Wu and A. Rates-Borras and O. Camps and R. J. Radke}, 
journal={IEEE Transactions on Pattern Analysis and Machine Intelligence}, 
title={A Systematic Evaluation and Benchmark for Person Re-Identification: Features, Metrics, and Datasets}, 
year={2018}, 
keywords={Benchmark testing;Cameras;Feature extraction;Histograms;Image color analysis;Measurement;Probes}, 
doi={10.1109/TPAMI.2018.2807450}, 
ISSN={0162-8828}, 
}