Collaborative Enhancement Network for Low-quality Multi-spectral Vehicle Re-identification

TL;DR

The paper tackles low-quality multi-spectral vehicle ReID by introducing CoEN, a framework that builds a fused proxy from RGB, NIR, and TIR spectra using a Proxy Generator, then dynamically ranks spectrum quality with a Dynamic Quality Sort Module to select a high-quality primary spectrum. It couples primary-based and proxy-based enhancement in a Collaborative Enhancement Module to recover missing cues and align multi-spectral features, supervised by joint identity and triplet losses. Empirical results on WMVeID863, RGBNT100, and MSVR310 show CoEN outperforms state-of-the-art methods, with notable gains in mAP and Rank-1, and ablations confirm the contributions of PG, DQSM, and CEM. The approach offers robust cross-spectral fusion under varying spectral quality and demonstrates strong potential for real-world, all-weather vehicle re-identification scenarios.

Abstract

The performance of multi-spectral vehicle Re-identification (ReID) is significantly degraded when some important discriminative cues in visible, near infrared and thermal infrared spectra are lost. Existing methods generate or enhance missing details in low-quality spectra data using the high-quality one, generally called the primary spectrum, but how to justify the primary spectrum is a challenging problem. In addition, when the quality of the primary spectrum is low, the enhancement effect would be greatly degraded, thus limiting the performance of multi-spectral vehicle ReID. To address these problems, we propose the Collaborative Enhancement Network (CoEN), which generates a high-quality proxy from all spectra data and leverages it to supervise the selection of primary spectrum and enhance all spectra features in a collaborative manner, for robust multi-spectral vehicle ReID. First, to integrate the rich cues from all spectra data, we design the Proxy Generator (PG) to progressively aggregate multi-spectral features. Second, we design the Dynamic Quality Sort Module (DQSM), which sorts all spectra data by measuring their correlations with the proxy, to accurately select the primary spectra with the highest correlation. Finally, we design the Collaborative Enhancement Module (CEM) to effectively compensate for missing contents of all spectra by collaborating the primary spectra and the proxy, thereby mitigating the impact of low-quality primary spectra. Extensive experiments on three benchmark datasets are conducted to validate the efficacy of the proposed approach against other multi-spectral vehicle ReID methods. The codes will be released at https://github.com/yongqisun/CoEN.

Figures (10)

• Figure 1: Performance of different spectra in complex scenes. RGB and NIR spectra perform better in normal illumination scenes, TIR spectra contain more identity information in flare scenes, while NIR and TIR spectra show superior performance in low illumination scenes
• Figure 2: (a) Mutual-based enhancement method topreidwang2025DeMowang2025MambaPro. (b) Specific spectral-based primary enhancement method facenet. (c) Our method.
• Figure 3: The framework of our proposed Collaborative Enhancement Network (CoEN). First, features ($F_R$, $F_N$, $F_T$) from different spectra are extracted using a ViT dosovitskiy2020image backbone network with shared parameters. These features are fed into the Proxy Generator (PG) to generate fused proxy feature $F_P$. Next, the Dynamic Quality Sort Module (DQSM) sorts and selects high-quality spectra feature $F_{\textit{1st}}$. Finally, the Collaborative Enhancement Module (CEM) enhances low-quality spectra by complementing missing discriminative cues and detail information, yielding multi-spectral features with rich discriminative information for vehicle ReID.
• Figure 4: Illustration of Collaborative Enhancement Module.
• Figure 5: Effect of $\gamma$ on mAP (%) and Rank-1 (%) on WMVeID863