Multi-scale Decomposed Convolution Refinement Network for Visible-Infrared Person Re-Identification

Mingsheng Zheng; Zirui Jiang; Bo Liu; Yupeng Chen; Jun Zhang; Kai Zhao

cs.CVAug 17, 2026

Multi-scale Decomposed Convolution Refinement Network for Visible-Infrared Person Re-Identification

Authors: Mingsheng Zheng, Zirui Jiang, Bo Liu, Yupeng Chen, Jun Zhang, Kai Zhao

Organizations: School of Computer Science and Technology, Xinjiang University, Urumqi, Xinjiang 830046, China · Joint International Research Laboratory of Silk Road Multilingual Cognitive Computing, Xinjiang University, Urumqi, Xinjiang 830046, China

Abstract

Visible-infrared person re-identification (VI-ReID) suffers from cross-modal discrepancies and limited discriminative capabilities, leading to suboptimal recognition performance. Current approaches exhibit limitations in semantic mining, cross-modal fusion and feature constraints. To tackle these challenges, we propose MDCRNet, a Multi-scale Decomposed Convolution Refinement Network that enhances cross-modal feature learning and discriminative metric learning. Specifically, we introduce a Hierarchical Learning Module (HLM) containing four Hierarchical Decomposed Convolution Attention (HDCA) modules, each equipped with lightweight channel attention and multi-scale spatial perception blocks to capture multi-scale spatial dependencies. Moreover, we develop a Joint Discriminative Metric Loss (JDML) incorporating a novel Granularity Discriminative Loss (GDL) that simultaneously optimizes intra-identity compactness and inter-identity separability across modalities. Extensive experiments on SYSU-MM01 and RegDB datasets demonstrate that MDCRNet achieves state-of-the-art performance on both benchmarks. Code is available at https://github.com/Kevin-zms/MDCRNet.

Figures & tables

Figure 1: Motivation of MDCRNet. (a) Existing methods adopt fixed-scale perception and capture only partial cues. (b) Our progressive multi-scale kernels jointly capture fine-grained (hands, watch, texture), mid-grained (face, clothing shape, upper-body pose), and coarse-grained (overall silhouette) information across the visible, infrared, and channel-augmented streams.

Figure 2: MDCRNet employs a quad-stream architecture that processes cross-modal images along with their channel-augmented variants, then utilizes ResNet backbone for hierarchical feature extraction. The HLM integrates four HDCA modules, each containing CA and MSPB with three PDPA components to capture multi-scale spatial dependencies for effective cross-modal semantic learning. The JDML combines identity loss, QCT loss, and our proposed GDL.

Method	Publish	All Search			Indoor Search
Method	Publish	R-1	R-10	mAP	R-1	R-10	mAP
AGW [ 31 ]	TPAMI 21	47.50	84.39	47.65	54.17	91.14	62.97
MPANet [ 25 ]	CVPR 21	70.58	96.21	68.24	76.74	98.21	80.95
SPOT [ 1 ]	TIP 22	65.34	92.73	62.25	69.42	96.22	74.63
CAL [ 24 ]	ICCV 23	42.40	85.00	40.70	45.90	87.60	54.30
PMT [ 15 ]	AAAI 23	67.53	95.36	51.86	71.66	96.73	76.52

Table 1: Comparison with SOTA methods on SYSU-MM01. R-

k

recognition accuracy (%) and mAP (%) are presented.

Method	Publish	V–I		I–V
Method	Publish	R-1	mAP	R-1	mAP
AGW [ 31 ]	TPAMI 21	70.00	66.30	70.40	65.90
SPOT [ 1 ]	TIP 22	80.30	72.40	79.30	72.20
PMT [ 15 ]	AAAI 23	84.81	76.55	84.16	75.13
SFANet [ 14 ]	TNNLS 21	76.31	68.00	70.20	63.80
LCNL [ 28 ]	IJCV 24	85.60	60.60	84.00	76.90

Table 2: Comparison with SOTA methods on RegDB. R-

k

recognition accuracy (%) and mAP (%) are presented.

HDCA		$\mathcal{L}_{JDM}$		All Search			Indoor Search
CA	PDPA	$(\mathcal{L}_{ID}+\mathcal{L}_{QCT})$	$\mathcal{L}_{GD}$	R-1	R-10	mAP	R-1	R-10	mAP
		✓		76.29	97.20	71.84	81.55	98.27	84.35
	✓	✓		77.28	97.03	73.24	82.21	98.73	84.95
✓	✓	✓		76.60	97.56	73.47	83.51	98.28	85.95
		✓	✓	74.86	97.02	71.41	81.20	98.76	84.47
	✓	✓	✓	77.16	97.48	73.54	82.62	98.42	85.34

Table 3: Ablation study of HDCA and GDL on SYSU-MM01. R-

k

accuracy (%) and mAP (%) are presented.

Configuration	mAP (%)
Full GDL	74.95
w/o $\mathcal{L}_{adaptive}$	73.27
w/o $\mathcal{L}_{center}$	73.64
w/o $\mathcal{L}_{cosine}$	73.77

Table 4: Leave-one-out ablation of GDL components on SYSU-MM01 (all-search).

Figure 3: Ablation study for the number of PDPA module ( $i$ ) in Eq. 2 and hyperparameter analysis of $\lambda_{1}$ , $\lambda_{2}$ and $\lambda_{3}$ in Eq. 11 .

Figure 4: Visualization of R10 retrieval results for visible-infrared images on SYSU-MM01. Red for wrong and green for correct matches.

Visible-Infrared Person Re-Identification

See all 13 papers

BIT: Matching-based Bi-directional Interaction Transformation Network for Visible-Infrared Person Re-Identification
Mar 15, 2026Haoxuan Xu, Guanglin NiuVisible-Infrared Person Re-Identification Person Re-Identification
School of Artificial Intelligence, Beihang University
Learning Language-Driven Sequence-Level Modal-Invariant Representations for Video-Based Visible-Infrared Person Re-Identification
Jan 17, 2026Xiaomei Yang, Antai Liu, Xizhan Gao +3Invariant Representation Learning Visible-Infrared Person Re-Identification
Shandong Key Laboratory of Ubiquitous Intelligent Computing, School of Information Science and Engineering, University of Jinan, Jinan 250022, China · College of Information Science and Technology & Artificial Intelligence, Nanjing Forestry University, Nanjing 210037, China · Department of Informatics, Modeling, Electronics, and Systems, University of Calabria, Via P. Bucci, cubo 41C, 87036, Rende (CS), Italy
Dual-Edged Homogeneous-Modality Similarity: Towards Visible-Infrared Modality-Incomplete Person Re-Identification with Modality Adaptive Matching
Jul 21, 2026Xin Xu, Shuhao Zhan, Wei Liu +3Missing-Modality Learning Cross-Modal Representation Learning
School of Computer Science and Technology, Wuhan University of Science and Technology, Wuhan, 430065, Hubei, China. · Hubei Province Key Laboratory of Intelligent Information Processing and Real-time Industrial System, Wuhan University of Science and Technology, Wuhan, 430081, Hubei, China. · School of Computer Science, Wuhan University, Wuhan, 430079, Hubei, China. +2

Mar 15, 2026J/K move · Enter open · S save

BIT: Matching-based Bi-directional Interaction Transformation Network for Visible-Infrared Person Re-Identification

Haoxuan Xu, Guanglin Niu

School of Artificial Intelligence, Beihang University

Visible-Infrared Person Re-Identification Person Re-Identification Cross-Modal Representation Learning Cross-Modal Feature Matching

Open paper AskPDF arXiv

Visible-Infrared Person Re-Identification (VI-ReID) is a challenging retrieval task due to the substantial modality gap between visible and infrared images. While existing methods attempt to bridge this gap by learning modality-invariant features within a shared embedding space, they often overlook the complex and implicit correlations between modalities. This limitation becomes more severe under distribution shifts, where infrared samples are often far fewer than visible ones. To address these challenges, we propose a novel network termed Bi-directional Interaction Transformation (BIT). Instead of relying on rigid feature alignment, BIT adopts a matching-based strategy that explicitly models the interaction between visible and infrared image pairs. Specifically, BIT employs an encoder-decoder architecture where the encoder extracts preliminary feature representations, and the decoder performs bi-directional feature integration and query aware scoring to enhance cross-modality correspondence. To our best knowledge, BIT is the first to introduce such pairwise matching-driven interaction in VI-ReID. Extensive experiments on several benchmarks demonstrate that our BIT achieves state-of-the-art performance, highlighting its effectiveness in the VI-ReID task.

Abstract

Visible-Infrared Person Re-Identification

Explore similar work

BIT: Matching-based Bi-directional Interaction Transformation Network for Visible-Infrared Person Re-Identification

Learning Language-Driven Sequence-Level Modal-Invariant Representations for Video-Based Visible-Infrared Person Re-Identification

Dual-Edged Homogeneous-Modality Similarity: Towards Visible-Infrared Modality-Incomplete Person Re-Identification with Modality Adaptive Matching