FedARKS: Federated Aggregation via Robust and Discriminative Knowledge Selection and Integration for Person Re-identification

Xin Xu; Binchang Ma; Zhixi Yu; Wei Liu

cs.CVMar 6, 2026

FedARKS: Federated Aggregation via Robust and Discriminative Knowledge Selection and Integration for Person Re-identification

Authors: Xin Xu, Binchang Ma, Zhixi Yu, Wei Liu

Organizations: School of Computer Science and Technology, Wuhan University of Science and Technology, Wuhan 430065 China

Abstract

The application of federated domain generalization in person re-identification (FedDG-ReID) aims to enhance the model's generalization ability in unseen domains while protecting client data privacy. However, existing mainstream methods typically rely on global feature representations and simple averaging operations for model aggregation, leading to two limitations in domain generalization: (1) Using only global features makes it difficult to capture subtle, domain-invariant local details (such as accessories or textures); (2) Uniform parameter averaging treats all clients as equivalent, ignoring their differences in robust feature extraction capabilities, thereby diluting the contributions of high quality clients. To address these issues, we propose a novel federated learning framework, Federated Aggregation via Robust and Discriminative Knowledge Selection and Integration (FedARKS), comprising two mechanisms: RK (Robust Knowledge) and KS (Knowledge Selection).

Figures & tables

Figure 1: Core challenges in FedDG-ReID: (a) Ignoring key local details that are invariant across domains; (b) Aggregation methods based on averaging dilute the ability of clients that are good at extracting cross-domain invariant features, thereby suppressing overall generalization ability.

Figure 2: Architectural diagram of FedARKS. Our method employs local branch training, in which each client fuses features in a dual-branch network. This figure illustrates the detailed process of local feature fusion and parameter aggregation, using client 1 as an example (other clients are shown in the background). After local training is complete, the server dynamically calibrates the aggregation weights of each client based on feature difference vectors and updates the global model in the next round of communication to improve cross-domain generalization capabilities.

Category	Method	Venue	MS+C2+C3→M		M+C2+C3→MS		MS+C2+M→C3		SAvg
Category	Method	Venue	mAP	R1	mAP	R1	mAP	R1	mAP	R1
Backbone: ResNet50
Federated Learning	SCAFFOLD	ICML’20	26.0	50.5	5.3	15.8	22.9	26.0	18.1	30.8
	MOON	CVPR’21	26.8	51.1	4.8	14.5	20.9	22.5	17.5	29.4
	FedProx	MLSys’20	29.3	53.8	5.8	17.4	19.1	17.7	18.1	29.6
Domain Generalization	SNR	CVPR’20	32.7	59.4	5.1	15.3	28.5	30.0	22.1	34.9

Table 1: Comparison of methods with different backbones and target unseen domains. M:Market1501, C2:CUHK02, C3:CUHK03, MS:MSMT17,R1: Rank-1. SAvg columns report the average mAP and average R1 across the three transfer tasks,highlighting model generalization ability.

Method	Market1501		CUHK02		CUHK03
	mAP	R1	mAP	R1	mAP	R1
FedProx	61.0	80.4	66.8	65.5	24.2	23.9
FedPav	53.9	76.0	59.7	56.3	19.6	19.6
FedReID	71.8	87.6	82.9	82.8	44.0	44.9
DACS	72.1	88.2	84.5	83.4	47.4	50.1
FedARKS	73.5	89.4	86.8	86.5	54.5	56.8

Table 2: Source domain performance comparison. Training domains: Market1501 + CUHK02 + CUHK03; Test domains: Market1501, CUHK02, and CUHK03.

	MS+C2	C3+C2	MS+C2
Attributes	+C3→M	+M→MS	+M→C3
RK KS	mAP R1	mAP R1	mAP R1
$\times$ $\times$	40.3 67.5	12.4 33.1	33.1 35.6
✓ $\times$	40.8 67.9	12.5 33.4	33.6 35.8
$\times$ ✓	41.0 68.0	12.5 33.9	34.0 35.8
✓ ✓	41.5 68.2	12.6 33.7	34.8 35.9

Table 3: Ablation study on Robust Knowledge(RK) and Knowledge Selection(KS).

Figure 3: Dynamic Attention Distribution Heatmap Comparison Across Diverse Scenarios.

Figure 4: Dynamic client weight distribution across 40 training epochs in FedARKS.

Federated Domain Generalization

See all 10 papers

FedBPrompt: Federated Domain Generalization Person Re-Identification via Body Distribution Aware Visual Prompts
Mar 13, 2026Xin Xu, Weilong Li, Wei Liu +5Federated Domain Generalization Person Re-Identification
School of Computer Science and Technology, Wuhan University of Science and Technology, China · Nanyang Technological University, Singapore · NERC for Multimedia Software, School of Computer Science, Wuhan University, China +3
CO-EVO: Co-evolving Semantic Anchoring and Style Diversification for Federated DG-ReID
Apr 29, 2026Fengchun Zhang, Qiang Ma, Liuyu Xiang +3Cross-Domain Generalization Federated Domain Generalization
School of Information and Software Engineering, University of Electronic Science and Technology of China · 2QiYuan Lab · School of Artificial Intelligence, Beijing University of Posts and Telecommunications +1
Domain-Skewed Federated Learning with Feature Decoupling and Calibration
Mar 15, 2026Huan Wang, Jun Shen, Jun Yan +1Federated Learning Heterogeneous Federated Learning
School of Computing and Information Technology, University of Wollongong, Australia · School of Computing and Information Systems, Singapore Management University, Singapore

Mar 13, 2026J/K move · Enter open · S save

FedBPrompt: Federated Domain Generalization Person Re-Identification via Body Distribution Aware Visual Prompts

Xin Xu, Weilong Li, Wei Liu, Wenke Huang+4

School of Computer Science and Technology, Wuhan University of Science and Technology, China · Nanyang Technological University, Singapore · NERC for Multimedia Software, School of Computer Science, Wuhan University, China+3

Federated Domain Generalization Person Re-Identification Federated Learning Visual Prompt Learning Federated Fine-Tuning

Open paper AskPDF arXiv

Federated Domain Generalization for Person Re-Identification (FedDG-ReID) learns domain-invariant representations from decentralized data. While Vision Transformer (ViT) is widely adopted, its global attention often fails to distinguish pedestrians from high similarity backgrounds or diverse viewpoints -- a challenge amplified by cross-client distribution shifts in FedDG-ReID. To address this, we propose Federated Body Distribution Aware Visual Prompt (FedBPrompt), introducing learnable visual prompts to guide Transformer attention toward pedestrian-centric regions. FedBPrompt employs a Body Distribution Aware Visual Prompts Mechanism (BAPM) comprising: Holistic Full Body Prompts to suppress cross-client background noise, and Body Part Alignment Prompts to capture fine-grained details robust to pose and viewpoint variations. To mitigate high communication costs, we design a Prompt-based Fine-Tuning Strategy (PFTS) that freezes the ViT backbone and updates only lightweight prompts, significantly reducing communication overhead while maintaining adaptability. Extensive experiments demonstrate that BAPM effectively enhances feature discrimination and cross-domain generalization, while PFTS achieves notable performance gains within only a few aggregation rounds. Moreover, both BAPM and PFTS can be easily integrated into existing ViT-based FedDG-ReID frameworks, making FedBPrompt a flexible and effective solution for federated person re-identification. The code is available at https://github.com/leavlong/FedBPrompt.

Abstract

Federated Domain Generalization

Explore similar work

FedBPrompt: Federated Domain Generalization Person Re-Identification via Body Distribution Aware Visual Prompts

CO-EVO: Co-evolving Semantic Anchoring and Style Diversification for Federated DG-ReID

Domain-Skewed Federated Learning with Feature Decoupling and Calibration