Pith. sign in

REVIEW 3 major objections 4 minor 111 references

HAMoBE: Hierarchical and Adaptive Mixture of Biometric Experts for Video-based Person ReID

T0 review · 3 major / 4 minor · reviewed 2026-08-05 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read HAMoBE claims that pair-adaptive fusion of long-term, short-term, and temporal biometric features via a two-level mixture of experts achieves state-of-the-art video person re-identification.

desk verdict The architecture is worth watching, but the headline MEVID gains currently rest on an unstated inference-time percentile selection (q) and a misattributed baseline comparison—two fixable, load-bearing problems. read the letter →

arxiv 2508.05038 v2 pith:QGAGEEOH submitted 2025-08-07 cs.CV

Yiyang Su , Yunping Shi , Feng Liu , Xiaoming Liu This is my paper
classification cs.CV
keywords videopersonre-identificationmixtureofexpertsbiometricfeaturesdual-inputgatingcloth-changingtemporal/gaitCLIPvisualencoder
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper sets out to establish that video-based person re-identification improves when a model deliberately separates three biometric cues—stable body shape, appearance, and gait—and then adaptively reweights them for each query-gallery pair instead of collapsing a track into one fixed feature. It proposes HAMoBE, a two-level mixture-of-experts built on frozen CLIP features, and reports state-of-the-art accuracy on MARS, LS-VID, CCVID, and MEVID, including a +13.0% Rank-1 gain over prior art on MEVID. If the claim holds, it would mean context-aware feature fusion matters more than finding a single robust representation, especially when clothing changes or motion cues dominate.

What carries the argument

The load-bearing mechanism is the two-level gating hierarchy: the first-layer gating network assigns per-token, per-expert weights over the eight low-level MLP experts, while the second-layer gating network produces a softmax weight vector over the long-term, short-term, and temporal experts. In dual-input mode, query and gallery features are concatenated, processed by multi-head self-attention, and fed through separate MLPs to generate pair-specific gate weights. This pair-adaptive weighting is what lets the model emphasize gait when clothing changes and appearance when clothing is stable.

What would settle it

Recompute Rank-1 and mAP on MEVID and CCVID using every query-gallery pair (q = 100) under the same training setup; if the accuracy drops toward the single-input numbers or the +13.0 margin disappears, the gain is an artifact of pair selection. A second check is to vary the definition of the central percentile and compare gate weights: if the final feature changes substantially with the selection rule, the protocol is influencing the result.

Watch

Extended reading notes

Core claim

The central claim is that a hierarchical mixture of biometric experts can disentangle long-term features ($f_L$), short-term features ($f_S$), and temporal/gait features ($f_T$) from frozen CLIP multi-layer representations, and then fuse them dynamically based on the paired inputs. The first level uses eight MLP experts to distill low-level cues from concatenated CLIP layers; the second level has three specialized experts for the three biometric attributes. A dual-input gating network, built from multi-head self-attention over the concatenated query and gallery features followed by separate MLPs, produces query- and gallery-specific weights. The final feature is a weighted sum of the three e

Load-bearing premise

The reported gains rest on an inference protocol that applies dual-input gating to only the central q% of query-gallery pairs; q and the percentile-ranking criterion are not specified, so unless those values were fixed in advance the headline margin could come from test-time subset selection rather than from the learned features.

Editorial extensions

If this is right

  • On MARS, HAMoBE reports mAP of 91.1 and top-1 of 94.6, outperforming the previous CLIP-based method TF-CLIP by roughly 1.6–1.7 points, and on LS-VID it reports top-1 of 92.1, a gain of 1.7 points.
  • On CCVID, HAMoBE reports 85.8 mAP and 87.6 top-1 under the different-clothes protocol, and the gap between same-clothes and different-clothes accuracy is smaller than for prior models, supporting the clothing-invariance argument.
  • On MEVID, HAMoBE reports top-1 of 74.0% in the same-clothes protocol and 20.1% in the different-clothes protocol, beating the previous best by 13.1 and 8.9 points respectively.
  • Ablations show the temporal/gait expert is the strongest individual expert, and the learned gating weights shift by dataset: gait-dominated on CCVID, appearance-dominated on MARS, and balanced between appearance and gait on MEVID.
  • A backbone ablation using ViT-B, matched to TF-CLIP's capacity, still beats TF-CLIP on MARS and LS-VID, suggesting the gains come from the framework design rather than from a larger frozen backbone.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A direct extension would apply HAMoBE to every query-gallery pair rather than only the central q% described in the inference protocol; if the margin persists at q=100, the adaptive-gating story is robust to pair selection.
  • The same two-level gating idea could transfer to multimodal biometric recognition, where per-pair weights decide which modality—face, body shape, or gait—should dominate, a setting the paper does not address.
  • Because the temporal expert alone nearly matches the full model on CCVID, an explicit pose or silhouette input could push different-clothes accuracy further, but that fusion is not explored in this work.
  • The reported consistency losses, which enforce long-term stability within a video and alignment across paired videos, appear to be a key training signal; their removal sharply reduces different-clothes accuracy, so they deserve closer scrutiny as the real source of some gains.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 4 minor

Summary. The paper proposes HAMoBE, a video-based person ReID framework built on a frozen CLIP visual encoder. Multi-layer CLIP features are passed through a two-level mixture of biometric experts: eight first-layer MLP experts distill low-level cues, and three second-layer experts produce long-term (body shape), short-term (appearance), and temporal (gait) features. A dual-input gating network, which can also operate in single-input mode, adaptively fuses the expert outputs per query-gallery pair. Training combines cross-entropy, consistency, temporal, and contrastive losses. The authors report state-of-the-art results on MARS, LS-VID, CCVID, and MEVID, including a large gain on the cloth-changing MEVID benchmark, plus ablations over experts, dual-input mode, and losses.

Significance. If the reported results hold, HAMoBE would be a meaningful empirical contribution: it demonstrates that pair-adaptive fusion of explicitly disentangled long-term, short-term, and temporal features can improve video-based ReID, especially under clothing change. The use of a frozen CLIP backbone, the explicit decomposition into three biometric experts, and the dual-input gating idea are all reasonable and potentially useful design choices. The paper also includes ablations (Tables 4–5, appendix) and acknowledges limitations. However, the central experimental claim depends on an underspecified inference protocol—the 'central q%' of query-gallery pairs—and on a reported gain number that is not correctly attributed in the text. Because this is an empirical architecture paper, these issues gate the credibility of the headline results and must be resolved before the contribution can be assessed.

major comments (3)
  1. [§3.3, 'Dual-Input Decision Mechanism in Inference'] The inference protocol is not specified. The text states that applying the dual-input decision mechanism to 'the central q% of query-gallery pairs—ranging from the (50−q/2)th to the (50+q/2)th percentile—maximizes the adaptive capabilities,' but q is never given, nor is the score used to order query-gallery pairs for the percentile. If q<100, all reported results in Tables 1–4 are computed on a subset of pairs selected at test time, and if the ordering is by the model's own similarity or by a criterion tuned on the test set, the reported gains, including the +13.0% Rank-1 MEVID claim, could reflect test-time subset selection rather than learned feature quality. Please specify q, the ordering criterion, the value used in each table, and report results on the full query-gallery evaluation set. A comparison between single-input, full-pair dual-input, and the q% protocol is essential for aud
  2. [§4.2, 'MEVID Dataset'] The text claims HAMoBE achieves 'improvements of 14.7% in mAP and 13.1% in top-1 over the SoTA CAL model.' According to Table 3, the differences between HAMoBE and CAL are +15.3 mAP (54.3 vs 39.0) and +17.4 top-1 (74.0 vs 56.6). The stated numbers match the differences vs CLIP3DReID: +14.7 mAP (54.3 vs 39.6) and +13.1 top-1 (74.0 vs 60.9). Please correct the attribution and tie the abstract's '+13.0% Rank-1' to the exact baseline and table row.
  3. [§3.4, Eqs. (8)–(9)] The consistency losses are not well-defined as written. Equations (8) and (9) use '1/t2' with a sum over t1≠t2, but t2 is not defined as a variable (the frame count is T elsewhere), and the normalization of the pairwise sum is unclear. Since these losses are part of the training objective, this is a reproducibility issue. Please provide the exact normalization, e.g., 1/(T(T−1)) or 1/T^2, and define all indices.
minor comments (4)
  1. [§3.3, Eq. (5)] The text says the multi-layer raw features are concatenated 'along the spatial dimension' but gives the result as [Ggallery, Gquery] ∈ R^{T×257×8d}. Concatenating along the token (spatial) dimension would yield T×514×4d, while concatenating along the feature dimension yields T×257×8d. Please clarify which dimension is used.
  2. [Table 5] The header 'MoE Con. Loss' is unclear. The text describes 'consistency losses,' so please rename the column to something like 'Consistency Loss' to avoid confusion.
  3. [Appendix, 'Ablation on Backbone'] The reported comparison 'achieving 90.5% (-0.6%) vs. 89.4% on MARS' is ambiguous: is 90.5 the HAMoBE score with ViT-B, and is -0.6 the gap to the main model? Please state the numbers explicitly and consistently.
  4. [General] The paper cites a 'Project Link' but no URL or code release is provided. Given the reproducibility concerns above, please include the link in the final version and consider releasing code.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: HAMoBE's claims are empirical comparisons against external benchmarks; no equation reduces to its inputs.

full rationale

The paper's derivation chain is an architecture proposal validated by benchmark accuracy, not a formal derivation from assumptions. The central claim (higher mAP/Rank-1 on MARS, LS-VID, CCVID, MEVID) is measured against external datasets and prior methods, so there is no self-definitional loop or fitted parameter renamed as a prediction. The only load-bearing methodological borrowings (multi-layer CLIP feature concatenation and the Transformer decoder in Eq. 3) are attributed to [57], an external prior work, not to the authors' own results. Self-citations in the references (e.g., [58], [60], [110]) are contextual related-work citations and are not used to justify the framework's core mechanism. The underspecified dual-input inference percentile q in Sec. 3.3 is a legitimate auditability and potential test-time-selection concern, but it is not a circularity: the paper never states q, so there is no exhibited reduction of a reported number to a fitted q value. Under the stated criteria, no circular step can be quoted, and the score is 0.

Assumptions & free parameters 7 free parameters · 4 assumptions · 2 invented entities

The central claim rests on: (i) frozen CLIP ViT-L/14 providing usable multi-layer features without fine-tuning; (ii) the assumption that long-term, short-term, and temporal features are separable and that their adaptive weighting is the right inductive bias for video ReID; (iii) standard supervised training with the specified losses; (iv) the chosen hyperparameters, including the undefined inference percentile q. No code or data are shipped, so the evaluation protocol itself is an unverifiable axiom.

free parameters (7)
  • Loss weights alpha, beta = alpha=0.5, beta=1
    Chosen by hand (Sec. 3.4); no sensitivity analysis on these values is reported.
  • Number of first-layer experts n1 = 8
    Ablated in Appendix A: 4 gives 82.6% Rank-1, 8 gives 85.8%, 16 gives 84.5% on CCVID DC. Selected on benchmark outcome.
  • Number of second-layer experts n2 = 3
    Fixed by the three hand-defined features (long, short, temporal); not ablated.
  • Transformer decoder blocks M = 4
    Reported in implementation details without ablation.
  • Contrastive margin m = 4
    Set in implementation details; no sensitivity analysis.
  • Number of frames T = 16
    Input length chosen in implementation details; no ablation.
  • Inference percentile q = unspecified
    The dual-input inference mechanism applies to the 'central q%' of query-gallery pairs (Sec. 3.3); q and the ordering criterion are never given, so this is a free parameter that directly affects the reported retrieval scores.
assumptions (4)
  • domain assumption Frozen CLIP ViT-L/14 features contain discriminative multi-layer visual semantics for person ReID without fine-tuning
    The entire pipeline rests on features G1..G4 from a frozen CLIP visual encoder (Sec. 3.1); if CLIP features are not suited to fine-grained identity matching, the framework has no basis.
  • domain assumption Long-term, short-term, and temporal features are separable and their adaptive weighting is the correct inductive bias for video ReID
    The core design in Sec. 3.2 assumes that body shape, appearance, and gait are the three key discriminative signals and that scenario-dependent weighting improves matching (Fig. 1).
  • standard math Standard supervised training with cross-entropy, consistency, and contrastive losses yields a model that generalizes to the test sets
    The learning objective in Eq. (7) follows standard practice for ReID; the paper does not prove any generalization bound and relies on benchmark evaluation as evidence.
  • domain assumption The dual-input gating benefits only from paired context, not from test-set membership information
    The inference protocol in Sec. 3.3 applies the gating to a subset of query-gallery pairs. For the reported results to be valid, the selection of that subset must not use ground-truth identity information.
invented entities (2)
  • The three 'biometric experts' (long-term, short-term, temporal) as architectural modules
    purpose: They disentangle the feature space into interpretable biometric channels that the gating networks combine for matching
    These are internal network modules, not independently measurable physical entities. Their named biometric semantics are an interpretation of the learned representations, supported only by the paper's own ablations and visualizations.
  • The dual-input gating network's per-pair weights
    purpose: Dynamically adjust expert contributions based on query-gallery context
    These weights are learned functions, not externally observable quantities; their claimed adaptivity is demonstrated only through the paper's benchmark numbers and weight distribution plots.

how reviews work

0 comments
Cite this review

Pith. "Pith review of HAMoBE: Hierarchical and Adaptive Mixture of Biometric Experts for Video-based Person ReID." pith.science (2026). https://pith.science/paper/QGAGEEOH

@misc{pith2026250805038,
  author       = {Pith},
  title        = {Pith review of: HAMoBE: Hierarchical and Adaptive Mixture of Biometric Experts for Video-based Person ReID},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QGAGEEOH}},
  note         = {Machine review of arXiv:2508.05038}
}
read the original abstract

Recently, research interest in person re-identification (ReID) has increasingly focused on video-based scenarios, which are essential for robust surveillance and security in varied and dynamic environments. However, existing video-based ReID methods often overlook the necessity of identifying and selecting the most discriminative features from both videos in a query-gallery pair for effective matching. To address this issue, we propose a novel Hierarchical and Adaptive Mixture of Biometric Experts (HAMoBE) framework, which leverages multi-layer features from a pre-trained large model (e.g., CLIP) and is designed to mimic human perceptual mechanisms by independently modeling key biometric features--appearance, static body shape, and dynamic gait--and adaptively integrating them. Specifically, HAMoBE includes two levels: the first level extracts low-level features from multi-layer representations provided by the frozen large model, while the second level consists of specialized experts focusing on long-term, short-term, and temporal features. To ensure robust matching, we introduce a new dual-input decision gating network that dynamically adjusts the contributions of each expert based on their relevance to the input scenarios. Extensive evaluations on benchmarks like MEVID demonstrate that our approach yields significant performance improvements (e.g., +13.0% Rank-1 accuracy).

Figures

Figures reproduced from arXiv: 2508.05038 by the authors.

Figure 1
Figure 1. Comparison of video-based person ReID methods. Existing solutions typically process videos to extract the most overlapping and robust features, such as long-term features (e.g., static body shape), which may not adequately address all iden￾tification scenarios, especially when dynamic attributes like gait are absent. Our HAMoBE approach dynamically adjusts the pro￾cessing of features based on the specific context of… view at source ↗
Figure 2
Figure 2. Overview of the proposed HAMoBE. Starting with the frozen CLIP Visual Encoder, which processes input videos into multi￾layer feature maps (g1, g2, g3, g4). These raw maps are then concatenated and passed through a hierarchical system of biometric experts. The first layer of experts processes raw features into more distinct features, which are then dynamically weighted by a gating network. This preparatory output (F1… view at source ↗
Figure 3
Figure 3. The proposed dual-input gating network in HAMoBE. This setup enables the network to adaptively synthe￾size information from both inputs, enhancing identification ac￾curacy by leveraging contextual differences between the paired videos. Note that [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Heatmaps of \protect \mathbf {W}^{*1} illustrating the influence of eight first-layer experts on the long-term, short-term, and temporal features. Each row reflects an expert’s contribution. The CLS token is removed; remaining tokens are reshaped to 16 × 16 to match im…
Figure 5
Figure 5. Figure 5: Distribution of adaptive weights {w}^{*2} . from the second-layer gating network across two datasets: CCVID and MEVID. For CCVID, the emphasis is predom￾inantly on gait (0.56), whereas MEVID balances between appearance (0.51) and gait (0.37), reflecting the dataset’s d…
Figure 6
Figure 6. Figure 6: (a) Average \protect \mathbf {W}^{*2} distribution on the MARS dataset. One example from (b) MARS, (c) CCVID, and (d) MEVID. A. Additional Experiments Number of First-Layer Experts. As suggested, we ab￾late the number of first-layer experts (4, 8, 16). Rank-1 accuracy …
Figure 7
Figure 7. Figure 7: An illutration of the full model in dual-input mode. [PITH_FULL_IMAGE:figures/full_fig_p014_7.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

111 extracted references · 75 canonical work pages

  1. [1]

    Biased mixtures of experts: Enabling computer vision inference under data transfer limitations

    Alhabib Abbas and Yiannis Andreopoulos. Biased mixtures of experts: Enabling computer vision inference under data transfer limitations. TIP, 2020. 3

    2020

  2. [2]

    Network of experts for large-scale image categoriza- tion

    Karim Ahmed, Mohammad Haris Baig, and Lorenzo Torre- sani. Network of experts for large-scale image categoriza- tion. In ECCV, 2016. 3

    2016

  3. [3]

    Salient-to-broad transition for video person re- identification

    Shutao Bai, Bingpeng Ma, Hong Chang, Rui Huang, and Xilin Chen. Salient-to-broad transition for video person re- identification. In CVPR, 2022. 6

    2022

  4. [4]

    Quo vadis, action recognition? a new model and the kinetics dataset

    Joao Carreira and Andrew Zisserman. Quo vadis, action recognition? a new model and the kinetics dataset. In CVPR, 2017. 6

    2017

  5. [5]

    Lagrange motion analysis and view em- beddings for improved gait recognition

    Tianrui Chai, Annan Li, Shaoxiong Zhang, Zilong Li, and Yunhong Wang. Lagrange motion analysis and view em- beddings for improved gait recognition. In CVPR, 2022. 3

    2022

  6. [6]

    Gaitset: Regarding gait as a set for cross-view gait recognition

    Hanqing Chao, Yiwei He, Junping Zhang, and Jianfeng Feng. Gaitset: Regarding gait as a set for cross-view gait recognition. In AAAI, 2019. 6

    2019

  7. [7]

    Gaitset: Cross-view gait recognition through utilizing gait as a deep set

    Hanqing Chao, Kun Wang, Yiwei He, Junping Zhang, and Jianfeng Feng. Gaitset: Cross-view gait recognition through utilizing gait as a deep set. TPAMI, 2021. 3

    2021

  8. [8]

    Learn- ing 3D shape feature for texture-insensitive person re- identification

    Jiaxing Chen, Xinyang Jiang, Fudong Wang, Jun Zhang, Feng Zheng, Xing Sun, and Wei-Shi Zheng. Learn- ing 3D shape feature for texture-insensitive person re- identification. In CVPR, 2021. 2

    2021

Show all 111 references
  1. [9]

    Eve: Efficient vision-language pre-training with masked prediction and modality-aware moe

    Junyi Chen, Longteng Guo, Jia Sun, Shuai Shao, Zehuan Yuan, Liang Lin, and Dongyu Zhang. Eve: Efficient vision-language pre-training with masked prediction and modality-aware moe. In AAAI, 2024. 3

    2024

  2. [10]

    Adamv-moe: Adaptive multi-task vision mixture-of-experts

    Tianlong Chen, Xuxi Chen, Xianzhi Du, Abdullah Rash- wan, Fan Yang, Huizhong Chen, Zhangyang Wang, and Yeqing Li. Adamv-moe: Adaptive multi-task vision mixture-of-experts. In ICCV, 2023. 3

    2023

  3. [11]

    Video person re-identification by temporal residual learning

    Ju Dai, Pingping Zhang, Dong Wang, Huchuan Lu, and Hongyu Wang. Video person re-identification by temporal residual learning. TIP, 2018. 2

    2018

  4. [12]

    Generalizable person re-identification with relevance-aware mixture of experts

    Yongxing Dai, Xiaotong Li, Jun Liu, Zekun Tong, and Ling-Yu Duan. Generalizable person re-identification with relevance-aware mixture of experts. In CVPR, 2021. 3

    2021

  5. [13]

    MEVID: Multi-view extended videos with identities for video person re-identification

    Daniel Davila, Dawei Du, Bryon Lewis, Christopher Funk, Joseph Van Pelt, Roderic Collins, Kellie Corona, Matt Brown, Scott McCloskey, Anthony Hoogs, et al. MEVID: Multi-view extended videos with identities for video person re-identification. In WACV, 2023. 6, 7, 8

    2023

  6. [14]

    Metagait: Learning to learn an omni sample adaptive representation for gait recognition

    Huanzhang Dou, Pengyi Zhang, Wei Su, Yunlong Yu, and Xi Li. Metagait: Learning to learn an omni sample adaptive representation for gait recognition. In ECCV, 2022. 3

    2022

  7. [15]

    Gaitgci: Generative counterfactual in- tervention for gait recognition

    Huanzhang Dou, Pengyi Zhang, Wei Su, Yunlong Yu, Yin- ing Lin, and Xi Li. Gaitgci: Generative counterfactual in- tervention for gait recognition. In CVPR, 2023. 3

    2023

  8. [16]

    Learning factored representations in a deep mixture of ex- perts

    David Eigen, Marc’Aurelio Ranzato, and Ilya Sutskever. Learning factored representations in a deep mixture of ex- perts. arXiv preprint arXiv:1312.4314, 2013. 3

    2013 arXiv

  9. [17]

    Video-based person re-identification with spatial and tem- poral memory networks

    Chanho Eom, Geon Lee, Junghyup Lee, and Bumsub Ham. Video-based person re-identification with spatial and tem- poral memory networks. In ICCV, 2021. 7

    2021

  10. [18]

    Gaitpart: Temporal part-based model for gait recognition

    Chao Fan, Yunjie Peng, Chunshui Cao, Xu Liu, Saihui Hou, Jiannan Chi, Yongzhen Huang, Qing Li, and Zhiqiang He. Gaitpart: Temporal part-based model for gait recognition. In CVPR, 2020. 3

    2020

  11. [19]

    Opengait: Revisiting gait recognition towards better practicality

    Chao Fan, Junhao Liang, Chuanfu Shen, Saihui Hou, Yongzhen Huang, and Shiqi Yu. Opengait: Revisiting gait recognition towards better practicality. In CVPR, 2023. 3

    2023

  12. [20]

    Switch transformers: Scaling to trillion parameter models with simple and efficient sparsity

    William Fedus, Barret Zoph, and Noam Shazeer. Switch transformers: Scaling to trillion parameter models with simple and efficient sparsity. Journal of Machine Learning Research, 2022. 3

    2022

  13. [21]

    Sta: Spatial-temporal attention for large-scale video-based person re-identification

    Yang Fu, Xiaoyang Wang, Yunchao Wei, and Thomas Huang. Sta: Spatial-temporal attention for large-scale video-based person re-identification. In AAAI, 2019. 1

    2019

  14. [22]

    Mutual mean-teaching: Pseudo label refinery for unsupervised do- main adaptation on person re-identification

    Yixiao Ge, Dapeng Chen, and Hongsheng Li. Mutual mean-teaching: Pseudo label refinery for unsupervised do- main adaptation on person re-identification. In ICLR, 2020. 2

    2020

  15. [23]

    Self- paced contrastive learning with hybrid memory for domain adaptive object re-id

    Yixiao Ge, Feng Zhu, Dapeng Chen, Rui Zhao, et al. Self- paced contrastive learning with hybrid memory for domain adaptive object re-id. In NeurIPS, 2020. 2

    2020

  16. [24]

    Hard mixtures of experts for large scale weakly supervised vision

    Sam Gross, Marc’Aurelio Ranzato, and Arthur Szlam. Hard mixtures of experts for large scale weakly supervised vision. In CVPR, 2017. 3

    2017

  17. [25]

    Appearance-preserving 3D convolution for video-based person re-identification

    Xinqian Gu, Hong Chang, Bingpeng Ma, Hongkai Zhang, and Xilin Chen. Appearance-preserving 3D convolution for video-based person re-identification. In ECCV, 2020. 2, 6, 7

    2020

  18. [26]

    Clothes-changing person re-identification with RGB modality only

    Xinqian Gu, Hong Chang, Bingpeng Ma, Shutao Bai, Shiguang Shan, and Xilin Chen. Clothes-changing person re-identification with RGB modality only. In CVPR, 2022. 2, 6, 7

    2022

  19. [27]

    Motion feature aggregation for video-based person re-identification

    Xinqian Gu, Hong Chang, Bingpeng Ma, and Shiguang Shan. Motion feature aggregation for video-based person re-identification. TIP, 2022. 6

    2022

  20. [28]

    Dense interaction learning for video-based person re-identification

    Tianyu He, Xin Jin, Xu Shen, Jianqiang Huang, Zhibo Chen, and Xian-Sheng Hua. Dense interaction learning for video-based person re-identification. In ICCV, 2021. 6

    2021

  21. [29]

    In defense of the triplet loss for person re-identification

    Alexander Hermans, Lucas Beyer, and Bastian Leibe. In defense of the triplet loss for person re-identification. arXiv preprint arXiv:1703.07737, 2017. 6

    2017 arXiv

  22. [30]

    Temporal complementary learning for video person re-identification

    Ruibing Hou, Hong Chang, Bingpeng Ma, Shiguang Shan, and Xilin Chen. Temporal complementary learning for video person re-identification. In ECCV, 2020. 6, 7

    2020

  23. [31]

    Bicnet-tks: Learning efficient spatial- temporal representation for video person re-identification

    Ruibing Hou, Hong Chang, Bingpeng Ma, Rui Huang, and Shiguang Shan. Bicnet-tks: Learning efficient spatial- temporal representation for video person re-identification. In CVPR, 2021. 7

    2021

  24. [32]

    Gait lateral network: Learning discriminative and compact representations for gait recognition

    Saihui Hou, Chunshui Cao, Xu Liu, and Yongzhen Huang. Gait lateral network: Learning discriminative and compact representations for gait recognition. In ECCV, 2020. 3

    2020

  25. [33]

    Celebrities-ReID: A benchmark for clothes variation in long-term person re-identification

    Yan Huang, Qiang Wu, Jingsong Xu, and Yi Zhong. Celebrities-ReID: A benchmark for clothes variation in long-term person re-identification. In IJCNN, 2019. 2

    2019

  26. [34]

    Beyond scalar neuron: Adopting vector-neuron capsules for long-term person re- identification

    Yan Huang, Jingsong Xu, Qiang Wu, Yi Zhong, Peng Zhang, and Zhaoxiang Zhang. Beyond scalar neuron: Adopting vector-neuron capsules for long-term person re- identification. TCSVT, 2019. 2

    2019

  27. [35]

    Clothing status awareness for long-term person re-identification

    Yan Huang, Qiang Wu, JingSong Xu, Yi Zhong, and ZhaoXiang Zhang. Clothing status awareness for long-term person re-identification. In ICCV, 2021. 2

    2021

  28. [36]

    3d local con- volutional neural networks for gait recognition

    Zhen Huang, Dixiu Xue, Xu Shen, Xinmei Tian, Houqiang Li, Jianqiang Huang, and Xian-Sheng Hua. 3d local con- volutional neural networks for gait recognition. In CVPR,

  29. [37]

    Adaptive mixtures of local experts

    Robert A Jacobs, Michael I Jordan, Steven J Nowlan, and Geoffrey E Hinton. Adaptive mixtures of local experts. Neural computation, 1991. 2, 3

    1991

  30. [38]

    Prox- yfusion: Face feature aggregation through sparse experts

    Bhavin Jawade, Alexander Stone, Deen Dayal Mohan, Xiao Wang, Srirangaraj Setlur, and Venu Govindaraju. Prox- yfusion: Face feature aggregation through sparse experts. NeurIPS, 2024. 3

    2024

  31. [39]

    Mixtral of experts

    Albert Q Jiang, Alexandre Sablayrolles, Antoine Roux, Arthur Mensch, Blanche Savary, Chris Bamford, Deven- dra Singh Chaplot, Diego de las Casas, Emma Bou Hanna, Florian Bressand, et al. Mixtral of experts. arXiv preprint arXiv:2401.04088, 2024. 3

    2024 arXiv

  32. [40]

    Cloth-changing person re- identification from a single image with gait prediction and regularization

    Xin Jin, Tianyu He, Kecheng Zheng, Zhiheng Yin, Xu Shen, Zhen Huang, Ruoyu Feng, Jianqiang Huang, Zhibo Chen, and Xian-Sheng Hua. Cloth-changing person re- identification from a single image with gait prediction and regularization. In CVPR, 2022. 2

    2022

  33. [41]

    Hierarchical mix- tures of experts and the em algorithm

    Michael I Jordan and Robert A Jacobs. Hierarchical mix- tures of experts and the em algorithm. Neural computation,

  34. [42]

    Feature disentanglement learning with switching and aggregation for video-based person re-identification

    Minjung Kim, MyeongAh Cho, and Sangyoun Lee. Feature disentanglement learning with switching and aggregation for video-based person re-identification. In WACV, 2023. 2

    2023

  35. [43]

    Keypoint relative position encoding for face recog- nition

    Minchul Kim, Yiyang Su, Feng Liu, Anil Jain, and Xiaom- ing Liu. Keypoint relative position encoding for face recog- nition. In CVPR, 2024. 3

    2024

  36. [44]

    Sapiensid: Foundation for human recognition

    Minchul Kim, Dingqiang Ye, Yiyang Su, Feng Liu, and Xi- aoming Liu. Sapiensid: Foundation for human recognition. In CVPR, 2025. 2

    2025

  37. [45]

    Sparse upcy- cling: Training mixture-of-experts from dense checkpoints

    Aran Komatsuzaki, Joan Puigcerver, James Lee-Thorp, Carlos Riquelme Ruiz, Basil Mustafa, Joshua Ainslie, Yi Tay, Mostafa Dehghani, and Neil Houlsby. Sparse upcy- cling: Training mixture-of-experts from dense checkpoints. In ICLR, 2023. 3

    2023

  38. [46]

    Gshard: Scaling giant models with conditional computation and automatic sharding

    Dmitry Lepikhin, HyoukJoong Lee, Yuanzhong Xu, Dehao Chen, Orhan Firat, Yanping Huang, Maxim Krikun, Noam Shazeer, and Zhifeng Chen. Gshard: Scaling giant models with conditional computation and automatic sharding. In ICLR, 2021. 3

    2021

  39. [47]

    Global-local temporal representations for video per- son re-identification

    Jianing Li, Jingdong Wang, Qi Tian, Wen Gao, and Shiliang Zhang. Global-local temporal representations for video per- son re-identification. In ICCV, 2019. 6

    2019

  40. [48]

    Multi- scale 3D convolution network for video based person re- identification

    Jianing Li, Shiliang Zhang, and Tiejun Huang. Multi- scale 3D convolution network for video based person re- identification. In AAAi, 2019. 2

    2019

  41. [49]

    Unsuper- vised person re-identification by deep learning tracklet as- sociation

    Minxian Li, Xiatian Zhu, and Shaogang Gong. Unsuper- vised person re-identification by deep learning tracklet as- sociation. In ECCV, 2018. 2

    2018

  42. [50]

    Unsuper- vised tracklet person re-identification

    Minxian Li, Xiatian Zhu, and Shaogang Gong. Unsuper- vised tracklet person re-identification. TPAMI, 2019. 2

    2019

  43. [51]

    Merge, then compress: Demystify efficient SMoe with hints from its routing policy

    Pingzhi Li, Zhenyu Zhang, Prateek Yadav, Yi-Lin Sung, Yu Cheng, Mohit Bansal, and Tianlong Chen. Merge, then compress: Demystify efficient SMoe with hints from its routing policy. In ICLR, 2024. 3

    2024

  44. [52]

    Learn- ing shape representations for person re-identification under clothing change

    Yu-Jhe Li, Xinshuo Weng, and Kris M Kitani. Learn- ing shape representations for person re-identification under clothing change. In WACV, 2021. 2

    2021

  45. [53]

    Gaitedge: Beyond plain end-to-end gait recognition for better practicality

    Junhao Liang, Chao Fan, Saihui Hou, Chuanfu Shen, Yongzhen Huang, and Shiqi Yu. Gaitedge: Beyond plain end-to-end gait recognition for better practicality. InECCV,

  46. [54]

    Gaitmask: Mask- based model for gait recognition

    Beibei Lin, Yu Liu, and Shunli Zhang. Gaitmask: Mask- based model for gait recognition. In BMVC, 2021. 3

    2021

  47. [55]

    Gait recognition via effective global-local feature representation and local temporal aggregation

    Beibei Lin, Shunli Zhang, and Xin Yu. Gait recognition via effective global-local feature representation and local temporal aggregation. In ICCV, 2021. 3

    2021

  48. [56]

    A bottom-up clustering approach to unsupervised person re-identification

    Yutian Lin, Xuanyi Dong, Liang Zheng, Yan Yan, and Yi Yang. A bottom-up clustering approach to unsupervised person re-identification. In AAAI, 2019. 2

    2019

  49. [57]

    Frozen CLIP models are efficient video learners

    Ziyi Lin, Shijie Geng, Renrui Zhang, Peng Gao, Gerard De Melo, Xiaogang Wang, Jifeng Dai, Yu Qiao, and Hong- sheng Li. Frozen CLIP models are efficient video learners. In ECCV, 2022. 4

    2022

  50. [58]

    Learning clothing and pose invariant 3D shape rep- resentation for long-term person re-identification

    Feng Liu, Minchul Kim, ZiAng Gu, Anil Jain, and Xiaom- ing Liu. Learning clothing and pose invariant 3D shape rep- resentation for long-term person re-identification. In ICCV,

  51. [59]

    Farsight: A physics-driven whole-body biometric system at large dis- tance and altitude

    Feng Liu, Ryan Ashbaugh, Nicholas Chimitt, Najmul Has- san, Ali Hassani, Ajay Jaiswal, Minchul Kim, Zhiyuan Mao, Christopher Perry, Zhiyuan Ren, et al. Farsight: A physics-driven whole-body biometric system at large dis- tance and altitude. In WACV, 2024. 1

    2024

  52. [60]

    Distilling clip with dual guidance for learning discrimina- tive human body shape representation

    Feng Liu, Minchul Kim, Zhiyuan Ren, and Xiaoming Liu. Distilling clip with dual guidance for learning discrimina- tive human body shape representation. In CVPR, 2024. 2, 6, 7

    2024

  53. [61]

    Person recogni- tion at altitude and range: Fusion of face, body shape and gait

    Feng Liu, Nicholas Chimitt, Lanqing Guo, Jitesh Jain, Aditya Kane, Minchul Kim, Wes Robbins, Yiyang Su, Dingqiang Ye, Xingguang Zhang, et al. Person recogni- tion at altitude and range: Fusion of face, body shape and gait. arXiv preprint arXiv:2505.04616, 2025. 1

    2025 arXiv

  54. [62]

    Video-based person re-identification with accumulative motion context

    Hao Liu, Zequn Jie, Karlekar Jayashree, Meibin Qi, Jian- guo Jiang, Shuicheng Yan, and Jiashi Feng. Video-based person re-identification with accumulative motion context. TCSVT, 2017. 1

    2017

  55. [63]

    Watching you: Global-guided recip- rocal learning for video-based person re-identification

    Xuehu Liu, Pingping Zhang, Chenyang Yu, Huchuan Lu, and Xiaoyun Yang. Watching you: Global-guided recip- rocal learning for video-based person re-identification. In CVPR, 2021. 2

    2021

  56. [64]

    Deeply coupled convolution–transformer with spatial– temporal complementary learning for video-based person re-identification

    Xuehu Liu, Chenyang Yu, Pingping Zhang, and Huchuan Lu. Deeply coupled convolution–transformer with spatial– temporal complementary learning for video-based person re-identification. TNNLS, 2023. 6

    2023

  57. [65]

    Video-based person re-identification with long short-term representation learning

    Xuehu Liu, Pingping Zhang, and Huchuan Lu. Video-based person re-identification with long short-term representation learning. In International Conference on Image and Graph- ics. Springer, 2023. 6

    2023

  58. [66]

    A video is worth three views: Trigeminal transformers for video-based per- son re-identification

    Xuehu Liu, Pingping Zhang, Chenyang Yu, Xuesheng Qian, Xiaoyun Yang, and Huchuan Lu. A video is worth three views: Trigeminal transformers for video-based per- son re-identification. TITS, 2024. 6

    2024

  59. [67]

    Spatial and temporal mutual promotion for video-based person re-identification

    Yiheng Liu, Zhenxun Yuan, Wengang Zhou, and Houqiang Li. Spatial and temporal mutual promotion for video-based person re-identification. In AAAI, 2019. 1

    2019

  60. [68]

    Recurrent convolutional network for video-based person re-identification

    Niall McLaughlin, Jesus Martinez Del Rincon, and Paul Miller. Recurrent convolutional network for video-based person re-identification. In CVPR, 2016. 1, 2

    2016

  61. [69]

    Boult, Anderson Rocha, Haidong Zhu, Zhao- heng Zheng, Ram Nevatia, Zaigham Randhawa, Sinan Sabri, and Gianfranco Doretto

    Kien Nguyen, Clinton Fookes, Sridha Sridharan, Feng Liu, Xiaoming Liu, Arun Ross, Dana Michalski, Huy Nguyen, Debayan Deb, Mahak Kothari, Manisha Saini, Dawei Du, Scott McCloskey, Gabriel Bertocco, Fernanda Andal ´o, Terrance E. Boult, Anderson Rocha, Haidong Zhu, Zhao- heng Z...

    2023

  62. [70]

    Temporal 3D shape modeling for video-based cloth- changing person re-identification

    Vuong D Nguyen, Pranav Mantini, and Shishir K Shah. Temporal 3D shape modeling for video-based cloth- changing person re-identification. In WACV, 2024. 2

    2024

  63. [71]

    Wheeler, and Xi- aoming Liu

    Necmiye Ozay, Yan Tong, Frederick W. Wheeler, and Xi- aoming Liu. Improving face recognition with a quality- based probabilistic framework. In CVPRW, 2009. 3

    2009

  64. [72]

    Video person re-id: Fantastic techniques and where to find them (student abstract)

    Priyank Pathak, Amir Erfan Eshratifar, and Michael Gormish. Video person re-id: Fantastic techniques and where to find them (student abstract). In AAAI, 2020. 7

    2020

  65. [73]

    Using mixture of expert models to gain insights into semantic segmentation

    Svetlana Pavlitskaya, Christian Hubschneider, Michael We- ber, Ruby Moritz, Fabian Huger, Peter Schlicht, and Marius Zollner. Using mixture of expert models to gain insights into semantic segmentation. In CVPRW, 2020. 3

    2020

  66. [74]

    Learn- ing transferable visual models from natural language super- vision

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learn- ing transferable visual models from natural language super- vision. In ICML, 2021. 2, 3

    2021

  67. [75]

    Scaling vision with sparse mix- ture of experts

    Carlos Riquelme, Joan Puigcerver, Basil Mustafa, Maxim Neumann, Rodolphe Jenatton, Andr´e Susano Pinto, Daniel Keysers, and Neil Houlsby. Scaling vision with sparse mix- ture of experts. In NeurIPS, 2021. 3

    2021

  68. [76]

    On the exploration of joint attribute learning for person re-identification

    Joseph Roth and Xiaoming Liu. On the exploration of joint attribute learning for person re-identification. In ACCV,

  69. [77]

    Outrageously large neural networks: The sparsely-gated mixture-of-experts layer

    Noam Shazeer, Azalia Mirhoseini, Krzysztof Maziarz, Andy Davis, Quoc Le, Geoffrey Hinton, and Jeff Dean. Outrageously large neural networks: The sparsely-gated mixture-of-experts layer. In ICLR, 2017. 3

    2017

  70. [78]

    Lidargait: Benchmarking 3d gait recognition with point clouds

    Chuanfu Shen, Chao Fan, Wei Wu, Rui Wang, George Q Huang, and Shiqi Yu. Lidargait: Benchmarking 3d gait recognition with point clouds. In CVPR, 2023. 3

    2023

  71. [79]

    Large-scale spatio- temporal person re-identification: Algorithms and bench- mark

    Xiujun Shu, Xiao Wang, Xianghao Zang, Shiliang Zhang, Yuanqi Chen, Ge Li, and Qi Tian. Large-scale spatio- temporal person re-identification: Algorithms and bench- mark. TCSVT, 2021. 2

    2021

  72. [80]

    Open-set biometrics: Beyond good closed-set models

    Yiyang Su, Minchul Kim, Feng Liu, Anil Jain, and Xi- aoming Liu. Open-set biometrics: Beyond good closed-set models. In ECCV, 2024. 2

    2024

  73. [81]

    Wheeler, and Xiaoming Liu

    Yan Tong, Frederick W. Wheeler, and Xiaoming Liu. Im- proving biometric identification through quality- based face and fingerprint biometric fusion. In CVPRW, 2010. 3

    2010

  74. [82]

    When person re-identification meets changing clothes

    Fangbin Wan, Yang Wu, Xuelin Qian, Yixiong Chen, and Yanwei Fu. When person re-identification meets changing clothes. In CVPRW, 2020. 2

    2020

  75. [83]

    Transferable joint attribute-identity deep learning for unsu- pervised person re-identification

    Jingya Wang, Xiatian Zhu, Shaogang Gong, and Wei Li. Transferable joint attribute-identity deep learning for unsu- pervised person re-identification. In CVPR, 2018. 2

    2018

  76. [84]

    Dygait: Ex- ploiting dynamic representations for high-performance gait recognition

    Ming Wang, Xianda Guo, Beibei Lin, Tian Yang, Zheng Zhu, Lincheng Li, Shunli Zhang, and Xin Yu. Dygait: Ex- ploiting dynamic representations for high-performance gait recognition. In ICCV, 2023. 3

    2023

  77. [85]

    Non-local neural networks

    Xiaolong Wang, Ross Girshick, Abhinav Gupta, and Kaim- ing He. Non-local neural networks. In CVPR, 2018. 6

    2018

  78. [86]

    Deep mixture of experts via shallow embedding

    Xin Wang, Fisher Yu, Lisa Dunlap, Yi-An Ma, Ruth Wang, Azalia Mirhoseini, Trevor Darrell, and Joseph E Gonzalez. Deep mixture of experts via shallow embedding. In Uncer- tainty in artificial intelligence, 2020. 3

    2020

  79. [87]

    Multi-expert adversarial attack detection in person re-identification using context inconsis- tency

    Xueping Wang, Shasha Li, Min Liu, Yaonan Wang, and Amit K Roy-Chowdhury. Multi-expert adversarial attack detection in person re-identification using context inconsis- tency. In ICCV, 2021. 3

    2021

  80. [88]

    Pyramid spatial-temporal aggre- gation for video-based person re-identification

    Yingquan Wang, Pingping Zhang, Shang Gao, Xia Geng, Hu Lu, and Dong Wang. Pyramid spatial-temporal aggre- gation for video-based person re-identification. In ICCV,

  81. [89]

    Decoupled feature-based mixture of experts for multi- modal object re-identification

    Yuhao Wang, Yang Liu, Aihua Zheng, and Pingping Zhang. Decoupled feature-based mixture of experts for multi- modal object re-identification. In AAAI, 2025. 3

    2025

  82. [90]

    Cavit: Contextual alignment vision transformer for video object re-identification

    Jinlin Wu, Lingxiao He, Wu Liu, Yang Yang, Zhen Lei, Tao Mei, and Stan Z Li. Cavit: Contextual alignment vision transformer for video object re-identification. In ECCV,

  83. [91]

    Adaptive graph representation learn- ing for video person re-identification

    Yiming Wu, Omar El Farouk Bourahla, Xi Li, Fei Wu, Qi Tian, and Xue Zhou. Adaptive graph representation learn- ing for video person re-identification. TIP, 2020. 7

    2020

  84. [92]

    DeepChange: A large long-term person re-identification benchmark with clothes change

    Peng Xu and Xiatian Zhu. DeepChange: A large long-term person re-identification benchmark with clothes change. In ICCV, 2023. 2

    2023

  85. [93]

    Condconv: Conditionally parameterized convolu- tions for efficient inference

    Brandon Yang, Gabriel Bender, Quoc V Le, and Jiquan Ngiam. Condconv: Conditionally parameterized convolu- tions for efficient inference. In NeurIPS, 2019. 3

    2019

  86. [94]

    Spatial-temporal graph convolutional network for video-based person re-identification

    Jinrui Yang, Wei-Shi Zheng, Qize Yang, Ying-Cong Chen, and Qi Tian. Spatial-temporal graph convolutional network for video-based person re-identification. In CVPR, 2020. 1

    2020

  87. [95]

    Person re- identification by contour sketch under moderate clothing change

    Qize Yang, Ancong Wu, and Wei-Shi Zheng. Person re- identification by contour sketch under moderate clothing change. TPAMI, 2019. 2

    2019

  88. [96]

    BigGait: Learning gait representation you want by large vision models

    Dingqiang Ye, Chao Fan, Jingzhe Ma, Xiaoming Liu, and Shiqi Yu. BigGait: Learning gait representation you want by large vision models. In CVPR, 2024. 3

    2024

  89. [97]

    Shiftaddvit: Mixture of multiplication primitives towards efficient vision transformer

    Haoran You, Huihong Shi, Yipin Guo, and Yingyan Lin. Shiftaddvit: Mixture of multiplication primitives towards efficient vision transformer. In NeurIPS, 2024. 3

    2024

  90. [98]

    Top- push video-based person re-identification

    Jinjie You, Ancong Wu, Xiang Li, and Wei-Shi Zheng. Top- push video-based person re-identification. In CVPR, 2016. 1

    2016

  91. [99]

    TF-CLIP: Learning text-free CLIP for video-based person re-identification

    Chenyang Yu, Xuehu Liu, Yingquan Wang, Pingping Zhang, and Huchuan Lu. TF-CLIP: Learning text-free CLIP for video-based person re-identification. In AAAI,

  92. [100]

    Unsupervised per- son re-identification by soft multilabel learning

    Hong-Xing Yu, Wei-Shi Zheng, Ancong Wu, Xiaowei Guo, Shaogang Gong, and Jian-Huang Lai. Unsupervised per- son re-identification by soft multilabel learning. In CVPR,

  93. [101]

    COCAS: A large-scale clothes changing per- son dataset for re-identification

    Shijie Yu, Shihua Li, Dapeng Chen, Rui Zhao, Junjie Yan, and Yu Qiao. COCAS: A large-scale clothes changing per- son dataset for re-identification. In CVPR, 2020. 2

    2020

  94. [102]

    Multidirection and multiscale pyramid in transformer for video-based pedes- trian retrieval

    Xianghao Zang, Ge Li, and Wei Gao. Multidirection and multiscale pyramid in transformer for video-based pedes- trian retrieval. IEEE Transactions on Industrial Informat- ics, 2022. 7

    2022

  95. [103]

    Ad-cluster: Aug- mented discriminative clustering for domain adaptive per- son re-identification

    Yunpeng Zhai, Shijian Lu, Qixiang Ye, Xuebo Shan, Jie Chen, Rongrong Ji, and Yonghong Tian. Ad-cluster: Aug- mented discriminative clustering for domain adaptive per- son re-identification. In CVPR, 2020. 2

    2020

  96. [104]

    Spatiotemporal trans- former for video-based person re-identification

    Tianyu Zhang, Longhui Wei, Lingxi Xie, Zijie Zhuang, Yongfei Zhang, Bo Li, and Qi Tian. Spatiotemporal trans- former for video-based person re-identification. arXiv preprint arXiv:2103.16469, 2021. 6

    2021 arXiv

  97. [105]

    Gait recognition via disentangled representation learning

    Ziyuan Zhang, Luan Tran, Xi Yin, Yousef Atoum, Xiaom- ing Liu, Jian Wan, and Nanxin Wang. Gait recognition via disentangled representation learning. In CVPR, 2019. 3

    2019

  98. [106]

    On learning disentangled representations for gait recognition

    Ziyuan Zhang, Luan Tran, Feng Liu, and Xiaoming Liu. On learning disentangled representations for gait recognition. TPAMI, 2022. 3, 6

    2022

  99. [107]

    Mars: A video benchmark for large-scale person re-identification

    Liang Zheng, Zhi Bie, Yifan Sun, Jingdong Wang, Chi Su, Shengjin Wang, and Qi Tian. Mars: A video benchmark for large-scale person re-identification. In ECCV, 2016. 6

    2016

  100. [108]

    Joint discriminative and generative learning for person re-identification

    Zhedong Zheng, Xiaodong Yang, Zhiding Yu, Liang Zheng, Yi Yang, and Jan Kautz. Joint discriminative and generative learning for person re-identification. In CVPR,

  101. [109]

    Parameter- efficient person re-identification in the 3D space

    Zhedong Zheng, Nenggan Zheng, and Yi Yang. Parameter- efficient person re-identification in the 3D space. TNNLS,

  102. [110]

    A quality-guided mixture of score-fusion experts framework for human recognition

    Jie Zhu, Yiyang Su, Minchul Kim, Anil Jain, and Xiaom- ing Liu. A quality-guided mixture of score-fusion experts framework for human recognition. In ICCV, 2025. 3

    2025

  103. [111]

    St- moe: Designing stable and transferable sparse expert mod- els

    Barret Zoph, Irwan Bello, Sameer Kumar, Nan Du, Yanping Huang, Jeff Dean, Noam Shazeer, and William Fedus. St- moe: Designing stable and transferable sparse expert mod- els. arXiv preprint arXiv:2202.08906, 2022. 3 0.39 0.52 0.09 0.34 0.54 0.12 0.35 0.15 0.50 0.49 0.10 0.41 (a...

    2022 arXiv

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.