DVHN: A Deep Hashing Framework for Large-scale Vehicle Re-identification

TL;DR

This paper introduces DVHN, a deep hashing framework for vehicle re-identification that enhances retrieval efficiency and accuracy by directly learning discrete binary hash codes, outperforming existing methods on key datasets.

Contribution

The paper presents the first integration of deep hash learning with vehicle re-identification, proposing a novel framework that directly learns optimal binary hash codes for improved performance.

Findings

DVHN achieves significant accuracy improvements on VehicleID and VeRi datasets.

The method reduces memory usage and increases retrieval speed.

Outperforms state-of-the-art deep hash methods in vehicle re-identification.

Abstract

In this paper, we make the very first attempt to investigate the integration of deep hash learning with vehicle re-identification. We propose a deep hash-based vehicle re-identification framework, dubbed DVHN, which substantially reduces memory usage and promotes retrieval efficiency while reserving nearest neighbor search accuracy. Concretely,~DVHN directly learns discrete compact binary hash codes for each image by jointly optimizing the feature learning network and the hash code generating module. Specifically, we directly constrain the output from the convolutional neural network to be discrete binary codes and ensure the learned binary codes are optimal for classification. To optimize the deep discrete hashing framework, we further propose an alternating minimization method for learning binary similarity-preserved hashing codes. Extensive experiments on two widely-studied vehicle re-identification datasets- \textbf{VehicleID} and \textbf{VeRi}-~have demonstrated the superiority of our method against the state-of-the-art deep hash methods. \textbf{DVHN} of $2048$ bits can achieve 13.94\% and 10.21\% accuracy improvement in terms of \textbf{mAP} and \textbf{Rank@1} for \textbf{VehicleID (800)} dataset. For \textbf{VeRi}, we achieve 35.45\% and 32.72\% performance gains for \textbf{Rank@1} and \textbf{mAP}, respectively.