WHOS: Weighted Histograms of Overlapping Stripes

We have designed a discriminative and efficient descriptor of person appearance for re-identification based on coarse, striped pooling of local features. It exploits a simple yet effective center support kernel to approximately segment foreground from background. The entire descriptor construction process is illustrated in the following figure:

 


Downloads:

Dataset #downloads
VIPER and CAVIAR (HS+RGB+HOG) with trials
VIPER (HS+RGB+HOG)
CAVIAR4REID (HS+RGB+HOG)
i-LIDS (HS+RGB+HOG)
ETHZ1 (HS+RGB+HOG)
ETHZ2 (HS+RGB+HOG)
ETHZ3 (HS+RGB+HOG)
3DPeS (HS+RGB+HOG)
VIPER (HS+RGB+Lab+HOG+LBP)
CAVIAR4REID (HS+RGB+Lab+HOG+LBP)
i-LIDS (HS+RGB+Lab+HOG+LBP)
ETHZ1 (HS+RGB+Lab+HOG+LBP)
ETHZ2 (HS+RGB+Lab+HOG+LBP)
ETHZ3 (HS+RGB+Lab+HOG+LBP)
3DPeS (HS+RGB+Lab+HOG+LBP)
CUHK01 (HS+RGB+HOG)
CUHK01 (HS+RGB+Lab+HOG+LBP)
PRID 450s (HS+RGB+HOG)
PRID 450s (HS+RGB+Lab+HOG+LBP)
GRID – underGround Re-IDentification (HS+RGB+HOG)
GRID – underGround Re-IDentification (HS+RGB+Lab+HOG+LBP)
Market-1501 v15.09.15 (HS+RGB+HOG)
Market-1501 v15.09.15 (HS+RGB+Lab+HOG+LBP)

If you want to receive the descriptor extracted on a particular dataset please send me (giuseppe [dot] lisanti [at] unifi [dot] it) a zip file containing all the images with the following format:

PPPPPIIIIIIIIII.EEE

where:

  • PPPPP represents the ID of the individual.
  • IIIIIIIIII represents the number of the image (of the cropped person image).
  • EEE is the extesion; you can send me PNG, JPG, BMP (preferable PNG).

The descriptor takes only 5ms to be extracted!


Please cite our paper if you use this descriptor:

@ARTICLE{LisantiPAMI14, 
author={Lisanti, G. and Masi, I. and Bagdanov, A. and Del Bimbo, A.}, 
journal={Pattern Analysis and Machine Intelligence, IEEE Transactions on}, 
title={Person Re-identification by Iterative Re-weighted Sparse Ranking}, 
year={2014}, 
volume={PP}, 
number={99}, 
pages={1-1}, 
doi={10.1109/TPAMI.2014.2369055}, 
ISSN={0162-8828}
}

Other papers related to this descriptor:

  • G. Lisanti , I. Masi , A. Del Bimbo, “Matching People across Camera Views using Kernel Canonical Correlation Analysis”, Eighth ACM/IEEE International Conference on Distributed Smart Cameras, 2014.
  • S. Karaman, G. Lisanti, A. D. Bagdanov, A. Del Bimbo, “Leveraging local neighborhood topology for large scale person re-identification ,” in Pattern Recognition, 2014.
  • S. Karaman, G. Lisanti, A. D. Bagdanov, A. Del Bimbo, “From re-identification to identity inference: labelling consistency by local similarity constraints”, Advances in Computer Vision and Pattern Recognition: Person Re-Identification, 2014.

Some plots showing the performance of the proposed descriptor with our ISR approach [LisantiPAMI14] for the MvsM scenario on the CAVIAR4REID and iLIDS  datasets, along with an analysis of the contribution of each component to the final descriptor performance:

cmc_caviarcmc_ilidscmc_viper


Changelog:
6 Jun 2017: Released the original (HS+RGB+HOG) and extended (HS+RGB+Lab+HOG+LBP) descriptors for: GRID and Market-1501.
26 Apr 2017: Released the original (HS+RGB+HOG) and extended (HS+RGB+Lab+HOG+LBP) descriptors for: PRID 450s.
22 March 2017: Released the original (HS+RGB+HOG) and extended (HS+RGB+Lab+HOG+LBP) descriptors for: CUHK01.
23 June 2015: Released the extended descriptors (HS+RGB+Lab+HOG+LBP) for: VIPeR, CAVIAR, ETHZ, iLIDS, 3DPeS.
18 June 2015: Released the descriptors (HS+RGB+HOG) for: VIPeR, CAVIAR, ETHZ, iLIDS, 3DPeS.
9 Jan 2015: Added a video explaining the extraction process.
10 Nov 2014: Released the VIPER and CAVIAR4REID descriptors (HS+RGB+HOG) with trials.

% (c) Copyright 2014-15 MICC - Media Integration and Communication Center,
% University of Florence. Giuseppe Lisanti <giuseppe.lisanti@unifi.it> and
% Iacopo Masi <iacopo.masi@unifi.it>

Comments are closed.