DATA-AUGMENTATION FOR REDUCING DATASET BIAS IN PERSON RE-IDENTIFICATION

Name That Dataset! Can you name the above re-identification datasets? Answers shown at the bottom of the page.

 

Abstract

In this paper we explore ways to address the issue of dataset bias in person re-identification by using data augmentation to increase the variability of the available datasets, and we introduce a novel data  augmentation method for re-identification based on changing the image background. We show that use of data augmentation can improve the cross-dataset generalisation of convolutional network based re-identification systems, and that changing the image background yields further improvements

 

Data Augmentation using Simulated Backgrounds

In this paper we introduce a novel data augmentation method for re-identification based on changing the image background. This increases the diversity of the images presented to the network during training and increases cross-dataset re-identification performance.

 

We compare the cross-dataset re-identification performance of the network trained using images with simulated background and standard data augmentation, with the network trained using standard data augmentation only

The average cross-dataset rank-1 CMC as the number of training images with simulated backgrounds is varied 

 

Citation

Data-Augmentation for Reducing Dataset Bias in Person Re-identification

N McLaughlin, J Martinez Del Rincon, P Miller

IEEE AVSS, 2015, AMMDS workshop

  

bibtex
@inproceedings{McLaughlin2015,
	Author = {McLaughlin, N. and Martinez Del Rincon, J. and Miller, P.},
	Booktitle = {Activity Monitoring by Multiple Distributed Sensing (AMMDS), 3rd Workshop on},
	Title = {Data-Augmentation for Reducing Dataset Bias in Person Re-identification},
	Year = {2015},
        Month={Aug}
}

 

Additional Material

AVSS presentation

 

 

Solutions to "Name the dataset"

 1. Viper 2. CAVIAR 3. i-Lids 4. ETHZ 5. CUHK 6. PRID