MDA framework: multi-architecture domain adaptation for unsupervised person re-identification
Authors: Wei Zhang, Peijun Ye, Dihu Chen, Tao Su
Organizations: School of Electronics and Information Technology, Sun Yat-sen University, Guangzhou, Guangdong, China · Peng Cheng Laboratory, Shenzhen, Guangdong, China · School of Integrated Circuits, Sun Yat-sen University, Shenzhen, Guangdong, China
Unsupervised person re-identification not only faces the challenge of domain gaps, but existing methods also rarely support multiple network architectures. To tackle both issues, this paper proposes a multi-architecture auto-adaptive domain adaptation method: Mamba augmented multi-architecture domain adaptation framework. Specifically, this paper constructs the multi-architecture domain adaptation capability at two levels. At the first level, we propose a composite operator(MDA operator) that encapsulates Mamba as a component for enhancing domain adaptation. At the second level, based on this operator, we propose a Mamba-enhanced multi-architecture Domain Adaptation framework(MDA framework). Comprehensive experiments on person re-identification benchmarks validate the consistency and flexibility of the proposed framework. The MDA framework supports versatile deployment strategies without being confined to a single integration point. Under a single deployment configuration, it surpasses several SOTA methods. Moreover, it demonstrates substantial performance improvements across both CNN and Transformer architectures. When embedded into a CNN backbone, the framework achieves a 10.2% mAP improvement on the adaptation task from CUHK03 to Market-1501. Integrated with a Transformer backbone, it yields a 19.8% mAP improvement on the same task.
Figures & tables
Figure 1: Workflow of domain adaptation
Figure 2: MDA framework and MDA Operator
algorithm
MSMT17 → Market
Market → MSMT17
mAP
R1
mAP
R1
MDJL [ 1 ]
—
—
13.4
34.3
SECRET [ 12 ]
81.0
92.6
31.3
60.4
PPLR [ 3 ]
81.5
92.8
31.4
61.1
CDCL [ 20 ]
81.9
93.1
29.5
59.7
MTNet [ 2 ]
82.7
93.0
32.1
59.6
Table 1: Comparison with the SOTA method on multiple benchmarks
Experimental setup
C to M
M to C
M to MS
MS to M
mAP
R1
mAP
R1
mAP
R1
mAP
R1
Baseline1
61.2
78.9
6.3
5.5
13.0
29.1
71.2
86.5
MDA framework after stage 1
65.0
81.5
7.2
5.4
14.6
32.5
71.2
86.5
MDA framework after stage 2
68.9
84.9
8.2
6.9
15.3
33.7
73.7
88.5
MDA framework after stage 3
66.0
82.5
7.8
6.1
15.0
33.0
74.4
88.6
MDA framework after stage 4
71.4
86.5
8.0
6.4
16.5
35.9
76.2
89.7
Table 2: Ablation experiment of MDA framework deployment on baseline 1.
Setting
C to M
M to C
M to MS
MS to M
mAP
R1
mAP
R1
mAP
R1
mAP
R1
Baseline2
75.3
88.6
10.1
7.9
20.2
41.3
80.1
91.8
MDA framework after stage 1
78.4
90.2
12.5
11.3
22.8
47.1
81.7
92.9
MDA framework after stage 2
77.0
89.8
11.5
10.3
22.8
46.1
80.6
91.7
MDA framework after stage 3
78.4
89.8
12.0
10.6
21.7
44.2
82.0
92.0
MDA framework after stage 4
80.8
91.6
12.7
10.9
24.1
47.4
84.3
93.7
Table 3: Ablation experiment of MDA framework deployment on baseline 2.
Experimental setup
C to M
M to C
M to MS
MS to M
mAP
R1
mAP
R1
mAP
R1
mAP
R1
Baseline3
23.9
48.0
6.9
6.3
4.8
14.6
33.2
59.1
MDA framework after stage 1
40.8
64.6
8.5
7.4
4.3
10.6
48.3
70.6
MDA framework after stage 2
41.5
63.2
8.6
7.5
5.4
13.3
53.0
72.2
MDA framework after stage 3
43.7
64.9
8.3
6.9
4.7
11.2
56.7
76.4
MDA framework after stage 4
36.5
58.9
5.3
4.8
5.7
14.2
46.8
68.2
Table 4: Ablation experiment of MDA framework deployment on baseline 3.
Part of network
Params(M)
FLOPs(G)
ResNet50 stage 1
0.010
0.158
MDA framework after stage 1
0.070
1.048
ResNet50 stage 2
0.216
0.895
MDA framework after stage 2
1.004
3.846
ResNet50 stage 3
1.220
1.358
MDA framework after stage 3
3.930
3.789
Table 5: MDA framework VS ResNet50 stage 1/2/3/4/5
Figure 3: Visualization for MDA framework. Note: From left to right are figure one to six in each group of images. Figure-1 : original image. Figure-2 : heatmap of baseline. Figure-3 : heatmap for MDA framework after stage1. Figure-4: heatmap for MDA framework after stage2. Figure-5: heatmap for MDA framework after stage3. Figure-6: heatmap for MDA framework after stage4.
Person re-identification (Re-ID) aims to match images of the same individual across non-overlapping camera views and remains challenging due to domain shifts caused by variations in illumination, background, camera characteristics, and population distributions. Although supervised models perform well under matched training and testing conditions, their performance degrades significantly when deployed in unseen environments. Existing intermediate domain approaches such as IDM and IDM++ alleviate this gap by constructing bridge feature distributions between domains; however, they rely on fixed mixing strategies and joint source-target access, limiting their applicability to multi-source and source-free settings. To address these limitations, this paper proposes Adaptive Intermediate Domain Adaptation (AIDA), also referred to as Source-Free Multi-Source Intermediate Domain Adaptation (SF-MIDA). The proposed framework treats intermediate-domain learning as a dynamically regulated process, where feature mixing and regularization strength are adaptively controlled using feedback signals derived from model uncertainty and training stability. A multi-source intermediate domain generator synthesizes diverse intermediate representations, while a pseudo-mirror regularization strategy preserves identity consistency under domain perturbations. Extensive experiments across domain generalization and source-free settings demonstrate the effectiveness of the proposed framework.
Sundas Iqbal, Qing Tian, Danish Ali +2
School of Software, Nanjing University of Information Science and Technology, Nanjing 210044, China · Wuxi Institute of Technology, Nanjing University of Information Science and Technology, Wuxi 214000, China · School of Computer Science, Wuhan University, Wuhan 430070, China +2
Unsupervised cross-domain person re-identification remains challenging because domain shifts generate noisy pseudo labels that are reinforced during iterative adaptation. Existing clustering-based methods often rely on limited reliability cues or global confidence criteria and may impose hard identity supervision on ambiguous target samples. To address these limitations, we propose CGPLR, a confidence-guided pseudo-label refinement framework for unsupervised domain-adaptive person re-identification. CGPLR estimates sample-level reliability by jointly modeling centroid compatibility, local neighborhood consistency, separation from the strongest competing cluster and temporal assignment stability. The resulting confidence distribution is used to derive cluster-adaptive upper and lower thresholds for a three-way selective decision. Reliable assignments are preserved, recoverable samples are reassigned only when an alternative cluster receives sufficient geometric and neighborhood support and highly ambiguous samples are deferred from hard identity supervision. The learning objective adapts to these decisions through confidence-weighted supervision for preserved samples, cautious corrected-label weighting for reassigned samples and label-free representation consistency for deferred samples. CGPLR achieves 76.4% Rank-1 accuracy and 47.6% mean average precision (mAP) on Market-1501 to MSMT17, and 95.1% Rank-1 accuracy and 87.7% mAP on MSMT17 to Market-1501, demonstrating effective reliability-aware selective adaptation. The code is available at https://github.com/Sundas058/CGPLR.
Sundas Iqbal, Danish Ali, Qing Tian +3
School of Software, Nanjing University of Information Science and Technology, Nanjing 210044, China · School of Computer Science, Wuhan University, Wuhan 430072, China · School of Computer Science, Nanjing University of Information Science and Technology, Nanjing 210044, China +5
Domain Generalizable (DG) person re-identification (Re-ID) has attracted growing research interest due to its potential for deployment in unseen real-world scenarios. Most existing approaches address DG Re-ID by focusing on training domain-generalizable encoders but ignore the possible refinements in inference stage. In contrast, this work explores an alternative direction which improves inference re-ranking to enhance DG Re-ID. Conventional re-ranking methods typically rely on neighborhood-based distances to refine the initial ranking list, inherently depending on features produced by the Re-ID encoder. However, they deteriorate on target domains since the encoder lacks sufficient generalizability to produce reliable feature distances on unseen scenarios. Inspired by the remarkable generalization capabilities of recent Multimodal Large Language Models (MLLMs), we propose an MLLM-empowered distance metric to improve re-ranking in DG Re-ID. Specifically, we first adapt an MLLM to Re-ID data through supervised fine-tuning, which incorporates a domain-agnostic prompt and a query-candidate hard mining scheme. Then, the adapted MLLM is employed to compute a μ-distance during inference, which is robust to domain gap and significantly enhances subsequent re-ranking performance. Our approach is model-agnostic and can be seamlessly integrated into previous re-ranking frameworks. Extensive experiments demonstrate that our approach consistently yields substantial performance improvements across multiple DG Re-ID benchmarks. The code of this work will be released at https://github.com/RikoLi/MUSE soon.
Jiachen Li, Xiaojin Gong
College of Information Science and Electronic Engineering Zhejiang University