Triplet Loss with Channel Attention for Person Re-identification

Organisciak, Daniel, Riachy, Chirine, Aslam, Nauman and Shum, Hubert (2019) Triplet Loss with Channel Attention for Person Re-identification. Journal of WSCG, 27 (2). pp. 161-169. ISSN 1213-6972

[img]
Preview
Text
Organisciak - Triplet Loss with Channel Attention for Person Re-identification AAM.pdf - Accepted Version

Download (300kB) | Preview
Official URL: https://doi.org/10.24132/jwscg.2019.27.2.9

Abstract

The triplet loss function has seen extensive use within person re-identification. Most works focus on either improving the mining algorithm or adding new terms to the loss function itself. Our work instead concentrates on two other core components of the triplet loss that have been under-researched. First, we improve the standard Euclidean distance with dynamic weights, which are selected based on the standard deviation of features across the batch. Second, we exploit channel attention via a squeeze and excitation unit in the backbone model to emphasise important features throughout all layers of the model. This ensures that the output feature vector is a better representation of the image, and is also more suitable to use within our dynamically weighted Euclidean distance function. We demonstrate that our alterations provide significant performance improvement across popular reidentification data sets, including almost 10% mAP improvement on the CUHK03 data set. The proposed model attains results competitive with many state-of-the-art person re-identification models.

Item Type: Article
Uncontrolled Keywords: Person Re-identification, Squeeze and Excitation, Triplet Loss, Metric Learning, Siamese Network, Channel Attention, Weighted Euclidean
Subjects: G400 Computer Science
Department: Faculties > Engineering and Environment > Computer and Information Sciences
Depositing User: Paul Burns
Date Deposited: 15 Apr 2019 15:11
Last Modified: 18 Nov 2019 16:00
URI: http://nrl.northumbria.ac.uk/id/eprint/38991

Actions (login required)

View Item View Item

Downloads

Downloads per month over past year

View more statistics