FM-ReID: Selective Competitive Token Routing for Object Re-Identification

Zhiqi Li; Xiaowei Zhou; Zeyuan Sun; Feng Gao; Junyu Dong

cs.CVSep 29, 2026

FM-ReID: Selective Competitive Token Routing for Object Re-Identification

Authors: Zhiqi Li, Xiaowei Zhou, Zeyuan Sun, Feng Gao, Junyu Dong

Organizations: Faculty of Information Science and Engineering, Ocean University of China, Qingdao, China · Sanya Oceanographic Institution, Ocean University of China, Sanya, China

Abstract

Object re-identification (ReID) faces a recurring challenge: different identities can share highly similar global appearances, while the cues that distinguish them are localized, heterogeneous, and visible only under particular viewpoints. This challenge arises in animal ReID through markings, contours, and scars, in person ReID through subtle clothing and accessory cues, and in vehicle ReID through localized appearance details. Although visual foundation models encode such information in dense tokens, a single holistic descriptor can obscure discriminative local signals. We propose FM-ReID, an end-to-end framework that formulates local representation learning as selective competitive token routing. Its Competitive Fine-grained Mining module uses multiple mining queries and a residual query to compete for dense DINOv3 tokens. Above-prior selection retains tokens preferentially allocated to each mining query, while the residual slot receives tokens excluded from the retrieval descriptors. The resulting multi-query descriptors are jointly trained with a holistic representation for retrieval, without fixed spatial partitions or equal-area constraints. FM-ReID achieves strong results on animal, person, and vehicle ReID benchmarks, supporting competitive token routing as an effective way to augment holistic foundation-model representations.

Figures & tables

Figure 1: (a) Object ReID often requires localized identity cues to complement holistic representations when different instances have similar global appearances. (b) FM-ReID selectively mines such cues from dense foundation-model tokens through competitive query routing, without fixed spatial partitions or equal-area constraints.

Figure 2: Overview of FM-ReID. DINOv3 provides a holistic CLS token and dense patch tokens. CFM selectively routes patch-token evidence through competing mining and residual queries, then retains above-prior assignments for the mining descriptors. During training, the holistic feature, holistic–fine-grained pairs, and their joint fusion receive ReID supervision. At inference, normalized holistic and fine-grained features form the retrieval embedding, while the residual output is discarded.

Figure 3: Competitive Fine-grained Mining (CFM). Mining queries and a residual query compete for each patch token through a softmax across queries. Above-prior mining assignments are retained and renormalized over tokens to form query-specific descriptors; their supports may overlap and vary in size. Query refinement produces the fine-grained features, whereas the residual output is excluded from the retrieval embedding.

Model	Input	mTop-1	mTop-5	BAKS
ConvNeXt-Base	$224^{2}$	81.1	90.3	78.5
EfficientNet-B3	$300^{2}$	77.8	88.5	75.1
ViT-Base	$224^{2}$	78.3	88.7	75.8
Swin-Base	$224^{2}$	81.5	90.4	79.1
TransReID	$256^{2}$	80.52	89.46	78.22
CLIP-ReID	$256^{2}$	76.49	87.61	74.21

Table 1: Comparison on (a) WildlifeReID-10k under the official closed-set protocol and (b) vehicle ReID benchmarks. The first four baselines in (a) are from Adam et al. (2025) ; TransReID and CLIP-ReID are from our runs. Bold indicates the best result in each metric.

Method	Reference	Backbone	MSMT17		Market-1501		DukeMTMC		Occ-Duke
Method	Reference	Backbone	mAP	R1	mAP	R1	mAP	R1	mAP	R1
PCB Sun et al. (2018)	ECCV18	CNN	40.4	68.2	81.6	93.8	69.2	83.3	-	-
OSNet Zhou et al. (2019)	ICCV19		52.9	78.7	84.9	94.8	73.5	88.6	-	-
Fast-ReID He et al. (2023)	MM23		59.9	83.3	-	-	78.9	89.6	-	-
TransReID He et al. (2021)	ICCV21	ViT	67.4	85.3	88.9	95.2	82.0	90.7	59.2	66.4
DCAL Zhu et al. (2022a)	CVPR22		64.0	83.1	87.5	94.7	80.1	89.0	-	-

Table 2: Comparison with selected published methods on MSMT17, Market-1501, DukeMTMC-reID, and Occ-Duke. Bold indicates the best result among the listed methods.

Source	IDs	Images
Self-collected	36	1,205
FS-48 subset	48	22,692
AAUZebraFish	5	281
WhaleSharkID	519	7,305
MFT25	116	14,149
Total	724	45,632

Table 3: FM-FISH: (a) dataset statistics and (b) comparison results. MFT25 identity labels denote annotated tracks. Bold indicates the best result in each metric.

Backbone	Holistic		CFM		WildlifeReID-10k			MSMT17
Backbone	CLS	GAP	Mining	Residual	mTop-1	mTop-5	BAKS	mAP	R1
DINOv3-B/16	$\checkmark$				83.29	91.44	81.46	76.1	89.6
	$\checkmark$	$\checkmark$			84.28	91.56	82.41	77.4	90.5
	$\checkmark$	$\checkmark$	$\checkmark$		84.54	92.01	82.90	79.6	91.4
	$\bm{\checkmark}$	$\bm{\checkmark}$	$\bm{\checkmark}$	$\bm{\checkmark}$	85.03	92.07	83.33	79.7	91.4

Table 4: Component ablation on WildlifeReID-10k and MSMT17. Mining and Residual denote the mining queries and auxiliary residual query. Within each benchmark, all variants use DINOv3-B/16, the same training recipe, and seed 1234; CFM variants retain the default fusion and diversity objectives. Bold indicates the best result in each metric.

Panel of Figure 4 — Figure 4: Sensitivity to the number of mining queries $K$ : (a) WildlifeReID-10k, with BAKS/mTop-1 on the left/right axes; (b) MSMT17, with mAP/Rank-1 on the left/right axes. Dotted lines mark $K=5$ ; the residual query is retained and excluded from $K$ . All runs use epoch-120 checkpoints and seed 1234. Panel (a) retains the full-model result at $K=5$ , while (b) uses independent sweep runs for every $K$ .

Method	Parameters (M)	GFLOPs	Peak memory (MiB)
TransReID	92.92	40.77	371.5
CLIP-ReID	86.14	22.78	346.1
Baseline	85.66	23.40	341.5
Baseline + GAP	86.84	23.40	346.0
FM-ReID	(+9.66%) 95.23	(+1.41%) 23.73	(+9.28%) 378.1

Table 5: Inference cost at

256\times 128

, FP32, and batch size 1 on an RTX 5080. Baseline uses DINOv3 with CLS pooling; parentheses indicate increases over Baseline + GAP.

Figure 5: Qualitative visualization on two panda identities. From left to right: input images, holistic responses, and five miner attention maps. Holistic responses use feature–patch cosine similarity, while miner maps show token aggregation weights. Each map is independently min–max normalized for display; colors indicate relative responses within each map.

Explore similar work

DiffReID: Discriminative Diffusion Model for Object Re-Identification
Sep 29, 2026Yingquan Wang, Pingping Zhang, Dong Wang +1Re-Identification Discriminative Models
School of Information and Communication Engineering, Dalian University of Technology · School of Future Technology, Dalian University of Technology
Thinking Before Matching: A Reinforcement Reasoning Paradigm Towards General Person Re-Identification
Apr 21, 2026Quan Zhang, Jingze Wu, Jialong Wang +3Re-Identification Scene Reasoning
Sun Yat-sen University · Alibaba Cloud Computing
Beyond Visual Cues: Semantic-Driven Token Filtering and Expert Routing for Anytime Person ReID
Apr 16, 2026Jiaxuan Li, Xin Wen, Zhihang LiRe-Identification Pedestrian Detection
Tsinghua University · Independent Researcher · University of Chinese Academy of Sciences

Sep 29, 2026, cs.CVJ/K move · Enter open · S save

DiffReID: Discriminative Diffusion Model for Object Re-Identification

Yingquan Wang, Pingping Zhang, Dong Wang, Huchuan Lu

School of Information and Communication Engineering, Dalian University of Technology · School of Future Technology, Dalian University of Technology

Re-Identification Discriminative Models

Open paper AskPDF arXiv

As a fundamental image processing task, object Re-Identification (ReID) aims to retrieve objects across non-overlapping cameras. Recently, with the development of deep learning, significant advancements have been made in object ReID. However, most existing methods suffer from generalization due to the limited size and diversity of ReID datasets. Meanwhile, current models tend to focus on extracting semantic patterns rather than learning identity-aware feature distributions. To address these issues, we propose a novel feature learning framework named \textbf{DiffReID} for object ReID. It leverages a discriminative diffusion model to gradually learn identity-aware distributions and generate identity-invariant features. More specifically, with the Contrastive Language-Image Pre-training (CLIP) model, we first obtain identity-aware text features by prompt tuning. Then, we propose a Vision-guided Noise Generator (VNG) to initialize probabilistic noises and gradually corrupt identity-aware text features. Afterwards, we take visual features as conditions and propose a Light Weight Denoiser (LWD) to denoise the corrupted text features step-by-step for identity-aware distribution learning. To obtain discriminative features, we further generate identity-invariant guided features from randomly sampling visual-guided noises. Finally, we propose a Mutual Enhancement Constraint (MEC) to facilitate mutual learning between visual features and guided features to enhance the representation robustness and discrimination. Extensive experiments on five object ReID benchmarks demonstrate that our method shows better results than most state-of-the-art methods. The source code is available at https://github.com/AWangYQ/DiffReID.

Abstract

Explore similar work

DiffReID: Discriminative Diffusion Model for Object Re-Identification

Thinking Before Matching: A Reinforcement Reasoning Paradigm Towards General Person Re-Identification

Beyond Visual Cues: Semantic-Driven Token Filtering and Expert Routing for Anytime Person ReID