Pith. sign in

REVIEW 4 major objections 5 minor 66 references

Hyperspectral Mamba for Hyperspectral Object Tracking

T0 review · 4 major / 5 minor · reviewed 2026-08-04 · deepseek-v4-flash

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

Pith's one-line read This paper claims HyMamba, a hyperspectral tracker using state space modules over raw spectral bands, achieves state-of-the-art results on seven hyperspectral tracking benchmarks.

desk verdict HyMamba is the first Mamba-based HS tracker and the internal ablation story is credible, but the headline SOTA margin on HOTC2020 is undercut by test-set hyperparameter selection and unverified baseline comparability. read the letter →

arxiv 2509.08265 v1 pith:Z3WCXCG4 submitted 2025-09-10 cs.CV

Long Gao , Yunhe Zhang , Yan Jiang , Weiying Xie , Yunsong Li This is my paper · ORCID
classification cs.CV
keywords hyperspectralobjecttrackingstatespacemodelMambaspectralfeaturemodelingvisualtemporalcontextpropagationfalse-colorcompressionbenchmarkevaluation
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

This paper proposes HyMamba, a hyperspectral object tracker whose central claim is that tracking should model the full spectral dimension with state space modules, not compress hyperspectral frames into false-color RGB images. It reports that the HyMamba network, built from Spectral State Integration modules with a three-directional Hyperspectral Mamba scan, achieves state-of-the-art AUC and precision on seven hyperspectral tracking benchmarks, including 73.0% AUC and 96.3% DP@20 on HOTC2020. The design propagates a spectral hidden state across transformer depths and across video frames, so spectral information from earlier frames and layers contributes to the current prediction. If true, this means hyperspectral tracking can use the raw spectral band structure rather than hand-designed band selection or regrouping, which the paper argues lose inter-band correlations.

What carries the argument

Spectral State Integration (SSI): a module inserted between transformer encoder layers that carries a three-part recurrent spectral hidden state (forward, backward, and spectral scans) across depths and frames, plus Joint Augment and Spectral Augment cross-attention mechanisms that enrich both the joint false-color-plus-spectral feature and the raw hyperspectral feature. Its engine is Hyperspectral Mamba (HSM), which adapts the Mamba sequence model—linear-time, hidden-state scanning—by adding a spectral-channel SSM to the usual forward/backward spatial scanning, and fuses the three directional streams with a joint branch through element-wise products in a Multi-Directional Fusion Module. Thi

What would settle it

Re-run the published baseline trackers on HOTC2020 with the same train/test splits, template sizes, and evaluation script; if SpectralTrack, SP-HST, or MCITrack reaches or exceeds HyMamba's 0.730 AUC / 0.963 DP@20 under identical conditions, the state-of-the-art claim is refuted. Also, remove the spectral SSM path from HSM; if AUC stays at 0.730 instead of dropping near the reported 1.0%, the spectral-scan mechanism is not causal.

Watch

Extended reading notes

Core claim

HyMamba's central claim is that unifying spectral, cross-depth, and temporal modeling in one state-space network is the right way to track in hyperspectral video. Existing hyperspectral trackers either select or regroup bands into false-color images, or fuse modalities, losing fine-grained inter-band correlations and leaving layer-wise and frame-wise spectral semantics isolated. HyMamba instead feeds the unconverted hyperspectral cube through the Spectral State Integration module: a Hyperspectral Mamba scans features along forward, backward, and spectral-channel directions, updating a recurrent hidden state that passes from transformer layer to layer and from frame to frame, while joint and

Load-bearing premise

The load-bearing premise is that the published baseline scores on the seven benchmarks were measured under the same conditions as the authors' own runs; the smallest claimed win—0.3% AUC over SpectralTrack on HOTC2020—would be the first to disappear if any baseline changes under a re-run.

Editorial extensions

If this is right

  • If the central claim is right, hyperspectral tracking no longer needs to trade away spectral fidelity: feeding raw spectral bands into a state-space tracker beats converting them to one or several false-color images.
  • The recurrent spectral hidden state means information from earlier frames and shallower layers propagates into current predictions, so tracking should stay more stable through occlusion, fast motion, and deformation than per-frame trackers.
  • The same architecture works across 15-, 16-, and 25-band datasets, indicating the design transfers across spectral resolutions without per-dataset band selection.
  • Because only the ASD module, SSI/HSM modules, and hyperspectral patch-embedding layers are trained while the backbone stays frozen, the method can adapt to new hyperspectral benchmarks with limited training data.
  • The spectral SSM scan is a load-bearing component: the ablation reports that removing it costs 1.0% AUC and 1.7% DP@20 on HOTC2020.

Reading between the lines

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

  • The temporal hidden state invites a stress test on longer sequences with extended occlusions: if spectral continuity is doing the work, tracking should degrade gracefully with occlusion length, and if it degrades abruptly, the temporal propagation claim is weaker than stated.
  • A band-subsampling experiment could separate two effects the paper confounds: the number of spectral channels fed in and the architecture's ability to use them. Running the same network on random 8-, 16-, 25-, and 32-band subsets of the same videos would show where the benefit saturates.
  • The same three-directional scanning and cross-depth state propagation idea could carry over to other hyperspectral video tasks such as detection or segmentation, which also often compress spectral data or process frames independently.
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

4 major / 5 minor

Summary. The paper proposes HyMamba, a Mamba/SSM-based tracker for hyperspectral video. It introduces a Spectral State Integration (SSI) module that propagates a recurrent spectral hidden state across transformer layers and frames, and a Hyperspectral Mamba (HSM) module with forward, backward, and spectral scanning SSMs. The network uses a frozen RGB-pretrained backbone and trains only the ASD module, SSI stack, and HS-specific patch embedding. Experiments on seven HS benchmarks report state-of-the-art results, e.g., 0.730 AUC and 0.963 DP@20 on HOTC2020, with ablations showing each proposed component contributes positively.

Significance. If verified, HyMamba would be a meaningful architectural contribution to hyperspectral tracking: it explicitly models cross-depth and temporal spectral information while avoiding the spectral-information loss of band selection/regrouping, and it demonstrates that Mamba-style state-space scanning can be adapted to the spectral dimension. Strengths include evaluation across seven benchmarks, consistent ablation results, an attribute-level analysis, and a commitment to release code. The central SOTA claim, however, currently rests on baseline numbers taken from heterogeneous publications and on hyperparameters selected on the same test set used for the headline results; the mathematical formulation of the SSM update also contains a dimension inconsistency. These issues are fixable but are load-bearing for the paper's main claim.

major comments (4)
  1. [III-C, Eqs. (6)-(7)] Equation (7) is dimensionally inconsistent as written. H_spec_i and H_spec_{i-1} are in R^n, F_HS_Linear is in R^{L×2D}, and B is declared as R^{n×1}; the term B F_HS_Linear cannot be computed, and even if broadcast, the result cannot be added to A H_{i-1} in R^n. The forward and backward SSM paths presumably have the same issue. Please provide the correct selective-scan formulation with explicit shapes for the input projection, the discretized A/B/C matrices, and the scan dimension. Without this, the central module is underspecified and not reproducible.
  2. [IV-B, Tables I-III] The state-of-the-art claim compares HyMamba with 16 RGB and 16 HS trackers using numbers tabulated from their source publications. The paper does not state that baselines were re-run under a unified protocol (same train/test splits, evaluation script, template-update policy, and hardware/software versions). The narrowest margin on the headline benchmark is 0.3% AUC over SpectralTrack (0.730 vs 0.727), which is within typical run-to-run variation for single-object tracking. Without controlled re-runs or clearly stated protocols, the 'state-of-the-art' claim is not robust. Please provide a unified comparison or explicitly qualify the claim.
  3. [IV-E, Table V, Fig. 8] The number of SSI layers (N=4) and the spectral hidden-state length (n=16) are selected by maximizing AUC/DP@20 on HOTC2020, the same benchmark used for the reported headline result of 0.730 AUC. No held-out validation split is described. The reported number is therefore an optimistic in-sample estimate, and the margin over the closest competitor may be exaggerated. Please select hyperparameters on a validation set, or demonstrate that the chosen values are stable across the other six datasets.
  4. [IV-B, IV-E] No experiment in the paper reports multiple runs, standard deviations, or significance tests. Given the small margins over second-place methods on HOTC2020 (+0.3% AUC, +0.9% DP@20), the reported improvements may not be statistically meaningful. Please report variance across repeated runs, or at least confirm the key margins (especially against SpectralTrack, SP-HST, and MCITrack) with multiple seeds.
minor comments (5)
  1. [Fig. 5] The legend lists 'MCCT', 'SiamRPN++', and 'fDSST', which do not appear in Table I; 'MCCT' likely refers to MCITrack. Please harmonize tracker names and curve colors between the table and figures.
  2. [Eq. (9)] The loss weights 5 and 2 in L_total are introduced without motivation or ablation. A brief justification or reference would help.
  3. [III-A and IV-A] The dynamic template update is said to be controlled by a fixed temporal interval and a confidence threshold, but neither the interval nor the threshold value is reported in Section IV-A. Please provide these implementation details.
  4. [III-C] The sentence 'for the HS feature, this step is only applied on F_HS_0' is ambiguous. It should be clarified whether the search filter is applied to the HS feature only in the first SSI layer, and why.
  5. [References [20], [21]] References [20] and [21] are given as URLs. If formal challenge papers or dataset descriptions exist, please cite them; otherwise, provide access dates and version information.

Circularity Check

1 steps flagged · score 4.0 of 10

HOTC2020 headline result is an in-sample hyperparameter selection; the architectural claim retains independent support from other benchmarks.

  1. fitted input called prediction [Section IV-E2 (Table V), Section IV-E3 (Fig. 8), Abstract]
    "The tracker employing 4 SSI modules yields the most favorable results. ... As the length increases, the AUC and DP@20 scores show a progressive improvement, culminating in peak performance at a dimension of 16. ... Thus, our investigation substantiates that the length of 16 offers the best performance. ... For instance, it achieves 73.0% of the AUC score and 96.3% of the DP@20 score on the HOTC2020 dataset."

    Tables V and Fig. 8 select N=4 and n=16 by maximizing AUC/DP@20 on the HOTC2020 test set; the paper then reports the resulting 0.730 AUC/0.963 DP@20 on the same test set as its headline SOTA evidence. The reported value is the objective of the hyperparameter selection, not an independent holdout estimate, so the HOTC2020 prediction is the fit itself. This is the fitted-input-called-prediction pattern. Results on HOTC2023/2024 were not used for this selection and provide independent support, so the circularity is partial.

full rationale

No equation-level circularity exists in the proposed architecture: HSM, SSI, and the hidden-state recurrence are defined independently of the benchmark numbers. The self-citations ([13], [61], [67]) are not load-bearing; the SOTA comparison includes many independent baselines, and the same-group baselines are not used to justify the design. The baseline scores copied from other papers create a comparability risk, but that is not circularity. The only substantive circularity is the HOTC2020 hyperparameter tuning: N=4 and n=16 are chosen by looking at the HOTC2020 test set, and the same set is used for the headline 73.0% AUC claim. Because six other benchmarks were not used for this tuning, the central architectural claim still has independent empirical content, so the overall score is 4 rather than higher.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The central claim rests primarily on empirical evaluation; the free parameters are architecture choices tuned by ablation on the test set, and the axioms are standard deep-learning assumptions plus domain assumptions about transferability of the frozen RGB backbone, usefulness of channel-wise SSM scanning, and comparability of published benchmark numbers. No new physical entities are introduced; the spectral hidden state is an internal network state.

free parameters (4)
  • spectral hidden state length n = 16
    Selected by ablation on HOTC2020 test (Fig. 8); the reported peak at 16 is used in all final results.
  • number of SSI transformer encoder layers N = 4
    Selected by ablation on HOTC2020 test (Table V); N=4 gives best AUC/DP and is used in the final model.
  • loss weights for bounding-box regression = 5 (L1), 2 (IoU)
    Chosen by hand in Eq. (9); no sensitivity analysis is provided.
  • dynamic template update interval / confidence threshold = not specified
    Section III-D says update is controlled by a fixed temporal interval and confidence threshold, but the values are never given; inference behavior is therefore not fully specified.
assumptions (5)
  • standard math RMS Norm, SiLU, and zero-order hold discretization of SSM matrices are computed as in Mamba [15] and produce stable gradients.
    Used throughout Eq. (3)-(7); standard deep learning toolbox.
  • domain assumption A frozen RGB-pretrained SUTrack backbone, with only the ASD, SSI, and HS patch-embedding modules trained, suffices as a feature extractor for HS data.
    Section III-D: only these modules are trained on HS data. If the frozen backbone does not transfer to HS, the reported improvements would not materialize.
  • domain assumption The spectral SSM's channel-wise scan captures long-range inter-band dependencies that are useful for tracking; the three-scan fusion in Eq. (8) is an effective aggregation.
    Section III-C and Table VIII; the paper provides no theoretical or statistical justification, only an ablation (+1.0% AUC for the spectral SSM).
  • domain assumption The evaluation protocols on HOTC2020/2023/2024, including the published baseline scores, are comparable across papers.
    Section IV; comparisons are taken from published results, but no unified re-run is described.
  • domain assumption Temporal continuity of the spectral hidden state is learnable from the training videos given the described 15-epoch training schedule.
    Section III-D and IV-A; no temporal sampling strategy is specified, so the inter-frame hidden-state learning is an assumed capability.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Hyperspectral Mamba for Hyperspectral Object Tracking." pith.science (2026). https://pith.science/paper/Z3WCXCG4

@misc{pith2026250908265,
  author       = {Pith},
  title        = {Pith review of: Hyperspectral Mamba for Hyperspectral Object Tracking},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/Z3WCXCG4}},
  note         = {Machine review of arXiv:2509.08265}
}
read the original abstract

Hyperspectral object tracking holds great promise due to the rich spectral information and fine-grained material distinctions in hyperspectral images, which are beneficial in challenging scenarios. While existing hyperspectral trackers have made progress by either transforming hyperspectral data into false-color images or incorporating modality fusion strategies, they often fail to capture the intrinsic spectral information, temporal dependencies, and cross-depth interactions. To address these limitations, a new hyperspectral object tracking network equipped with Mamba (HyMamba), is proposed. It unifies spectral, cross-depth, and temporal modeling through state space modules (SSMs). The core of HyMamba lies in the Spectral State Integration (SSI) module, which enables progressive refinement and propagation of spectral features with cross-depth and temporal spectral information. Embedded within each SSI, the Hyperspectral Mamba (HSM) module is introduced to learn spatial and spectral information synchronously via three directional scanning SSMs. Based on SSI and HSM, HyMamba constructs joint features from false-color and hyperspectral inputs, and enhances them through interaction with original spectral features extracted from raw hyperspectral images. Extensive experiments conducted on seven benchmark datasets demonstrate that HyMamba achieves state-of-the-art performance. For instance, it achieves 73.0\% of the AUC score and 96.3\% of the DP@20 score on the HOTC2020 dataset. The code will be released at https://github.com/lgao001/HyMamba.

Figures

Figures reproduced from arXiv: 2509.08265 by the authors.

Figure 1
Figure 1. Overall architecture of the proposed HyMamba framework, comprising ASD, a feature extraction network, a tracking head, and a dynamic template [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. (a) Overview of a single SSI transformer encoder layer, illustrating the interaction between joint and HS features. (b) Internal composition of the [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Detailed architecture of the Hyperspectral Mamba (HSM) module. [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Illustration of the three scanning strategies adopted in the HSM. [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Comparisons of HyMamba and RGB trackers on the corresponding RGB or false-color videos of HOTC2020 dataset. [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 6
Figure 6. Figure 6: Comparisons of HyMamba and other HS trackers on HOTC2023 dataset. [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: Comparisons of HyMamba and other trackers on HOTC2024 dataset. [PITH_FULL_IMAGE:figures/full_fig_p009_7.png]
Figure 8
Figure 8. Figure 8: Effect of varying spectral hidden state lengths on tracking perfor [PITH_FULL_IMAGE:figures/full_fig_p010_8.png]
Figure 9
Figure 9. Figure 9: Visualization results of the proposed HyMamba and other HS trackers on six representative video sequences (vis2023-card16, nir2023-basketball3, [PITH_FULL_IMAGE:figures/full_fig_p011_9.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

66 extracted references · 50 canonical work pages

  1. [1]

    Material based object tracking in hyperspectral videos,

    F. Xiong, J. Zhou, and Y . Qian, “Material based object tracking in hyperspectral videos,”IEEE Transactions on Image Processing, vol. 29, pp. 3719–3733, 2020

    2020

  2. [2]

    Mixformer: End-to-end tracking with iterative mixed attention,

    Y . Cui, C. Jiang, L. Wang, and G. Wu, “Mixformer: End-to-end tracking with iterative mixed attention,” inProceedings of the IEEE/CVF con- ference on computer vision and pattern recognition, 2022, pp. 13 608– 13 618

    work page 2022

  3. [3]

    High- performance transformer tracking,

    X. Chen, B. Yan, J. Zhu, H. Lu, X. Ruan, and D. Wang, “High- performance transformer tracking,”IEEE Transactions on Pattern Anal- ysis and Machine Intelligence, vol. 45, no. 7, pp. 8507–8523, 2022. IEEE TRANSACTIONS ON IMAGE PROCESSING 12

    work page 2022

  4. [4]

    Material- guided multiview fusion network for hyperspectral object tracking,

    Z. Li, F. Xiong, J. Zhou, J. Lu, Z. Zhao, and Y . Qian, “Material- guided multiview fusion network for hyperspectral object tracking,” IEEE Transactions on Geoscience and Remote Sensing, vol. 62, pp. 1–15, 2024

    2024

  5. [5]

    Tftn: A transformer-based fusion tracking framework of hyperspectral and rgb,

    C. Zhao, H. Liu, N. Su, and Y . Yan, “Tftn: A transformer-based fusion tracking framework of hyperspectral and rgb,”IEEE Transactions on Geoscience and Remote Sensing, vol. 60, pp. 1–15, 2022

    work page 2022

  6. [6]

    Band regrouping and response-level fusion for end-to-end hyperspectral object tracking,

    E. Ouyang, J. Wu, B. Li, L. Zhao, and W. Hu, “Band regrouping and response-level fusion for end-to-end hyperspectral object tracking,” IEEE Geoscience and Remote Sensing Letters, vol. 19, pp. 1–5, 2021

    work page 2021

  7. [7]

    Swintrack: A simple and strong baseline for transformer tracking,

    L. Lin, H. Fan, Z. Zhang, Y . Xu, and H. Ling, “Swintrack: A simple and strong baseline for transformer tracking,”Advances in Neural Information Processing Systems, vol. 35, pp. 16 743–16 754, 2022

    work page 2022

  8. [8]

    Spirit: Spectral awareness interaction network with dynamic template for hyperspectral object tracking,

    Y . Chen, Q. Yuan, Y . Tang, Y . Xiao, J. He, and L. Zhang, “Spirit: Spectral awareness interaction network with dynamic template for hyperspectral object tracking,”IEEE Transactions on Geoscience and Remote Sensing, 2023

    work page 2023

Show all 66 references
  1. [9]

    Siambag: Band attention grouping- based siamese object tracking network for hyperspectral videos,

    W. Li, Z. Hou, J. Zhou, and R. Tao, “Siambag: Band attention grouping- based siamese object tracking network for hyperspectral videos,”IEEE Transactions on Geoscience and Remote Sensing, vol. 61, pp. 1–12, 2023

    2023

  2. [10]

    Bs-siamrpn: Hyperspectral video tracking based on band selection and the siamese region proposal network,

    S. Wang, K. Qian, and P. Chen, “Bs-siamrpn: Hyperspectral video tracking based on band selection and the siamese region proposal network,” in2022 12th Workshop on Hyperspectral Imaging and Signal Processing: Evolution in Remote Sensing (WHISPERS). IEEE, 2022, pp. 1–8

    2022

  3. [11]

    Hy-tracker: A novel framework for enhancing efficiency and accuracy of object tracking in hyperspectral videos,

    M. A. Islam, W. Xing, J. Zhou, Y . Gao, and K. K. Paliwal, “Hy-tracker: A novel framework for enhancing efficiency and accuracy of object tracking in hyperspectral videos,”IEEE Transactions on Geoscience and Remote Sensing, 2024

    2024

  4. [12]

    A siamese network-based tracking framework for hyperspectral video,

    Y . Tang, H. Huang, Y . Liu, and Y . Li, “A siamese network-based tracking framework for hyperspectral video,”Neural Computing and Applications, vol. 35, no. 3, pp. 2381–2397, 2023

    2023

  5. [13]

    Cbff-net: A new framework for efficient and accurate hyperspectral object tracking,

    L. Gao, P. Liu, Y . Jiang, W. Xie, J. Lei, Y . Li, and Q. Du, “Cbff-net: A new framework for efficient and accurate hyperspectral object tracking,” IEEE Transactions on Geoscience and Remote Sensing, vol. 61, pp. 1– 14, 2023

    2023

  6. [14]

    Learning a deep ensemble network with band importance for hyperspectral object tracking,

    Z. Li, F. Xiong, J. Zhou, J. Lu, and Y . Qian, “Learning a deep ensemble network with band importance for hyperspectral object tracking,”IEEE Transactions on Image Processing, vol. 32, pp. 2901–2914, 2023

    2023

  7. [15]

    Mamba: Linear-time sequence modeling with selective state spaces,

    A. Gu and T. Dao, “Mamba: Linear-time sequence modeling with selective state spaces,”arXiv preprint arXiv:2312.00752, 2023

    2023 arXiv

  8. [16]

    Mamba-fetrack v2: Revisiting state space model for frame- event based visual object tracking,

    S. Wang, J. Huang, Q. Ma, J. Gao, C. Xu, X. Wang, L. Chen, and B. Jiang, “Mamba-fetrack v2: Revisiting state space model for frame- event based visual object tracking,”arXiv preprint arXiv:2506.23783, 2025

    2025 arXiv

  9. [17]

    Exploring enhanced contextual information for video-level object tracking,

    B. Kang, X. Chen, S. Lai, Y . Liu, Y . Liu, and D. Wang, “Exploring enhanced contextual information for video-level object tracking,” in Proceedings of the AAAI Conference on Artificial Intelligence, vol. 39, no. 4, 2025, pp. 4194–4202

    2025

  10. [18]

    Vision mamba: Efficient visual representation learning with bidirectional state space model,

    L. Zhu, B. Liao, Q. Zhang, X. Wang, W. Liu, and X. Wang, “Vision mamba: Efficient visual representation learning with bidirectional state space model,”arXiv preprint arXiv:2401.09417, 2024

    2024 arXiv

  11. [19]

    Vmamba: Visual state space model,

    Y . Liu, Y . Tian, Y . Zhao, H. Yu, L. Xie, Y . Wang, Q. Ye, J. Jiao, and Y . Liu, “Vmamba: Visual state space model,”Advances in neural information processing systems, vol. 37, pp. 103 031–103 063, 2024

    2024

  12. [20]

    The hyperspectral object tracking challenge (hot2023),

    J. Zhou, F. Xiong, H. Quangluong, M. Zahiri, W. Charle, Y . Zhong, P. Ghamisi, and J. Chanussot, “The hyperspectral object tracking challenge (hot2023),” [Online]. Available: https://www.hsitracking.com, 2023

    2023

  13. [21]

    The hyperspectral object tracking challenge (hot2024),

    F. Xiong, J. Zhou, R. Muszynski, W. Charle, Y . Zhong, P. Ghamisi, and J. Chanussot, “The hyperspectral object tracking challenge (hot2024),” [Online]. Available: https://www.hsitracking.com, 2024

    2024

  14. [22]

    Fully-convolutional siamese networks for object tracking,

    L. Bertinetto, J. Valmadre, J. F. Henriques, A. Vedaldi, and P. H. Torr, “Fully-convolutional siamese networks for object tracking,” inComputer Vision–ECCV 2016 Workshops: Amsterdam, The Netherlands, October 8-10 and 15-16, 2016, Proceedings, Part II 14. Springer, 2016, pp. 850–865

    2016

  15. [23]

    High performance visual tracking with siamese region proposal network,

    B. Li, J. Yan, W. Wu, Z. Zhu, and X. Hu, “High performance visual tracking with siamese region proposal network,” inProceedings of the IEEE conference on computer vision and pattern recognition, 2018, pp. 8971–8980

    2018

  16. [24]

    Siamrpn++: Evolution of siamese visual tracking with very deep networks,

    B. Li, W. Wu, Q. Wang, F. Zhang, J. Xing, and J. Yan, “Siamrpn++: Evolution of siamese visual tracking with very deep networks,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2019, pp. 4282–4291

    2019

  17. [26]

    Deformable siamese attention networks for visual object tracking,

    Y . Yu, Y . Xiong, W. Huang, and M. R. Scott, “Deformable siamese attention networks for visual object tracking,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2020, pp. 6728–6737

    2020

  18. [27]

    Transformer tracking,

    X. Chen, B. Yan, J. Zhu, D. Wang, X. Yang, and H. Lu, “Transformer tracking,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2021, pp. 8126–8135

    2021

  19. [28]

    Learning spatio-temporal transformer for visual tracking,

    B. Yan, H. Peng, J. Fu, D. Wang, and H. Lu, “Learning spatio-temporal transformer for visual tracking,” inProceedings of the IEEE/CVF international conference on computer vision, 2021, pp. 10 448–10 457

    2021

  20. [29]

    Seqtrack: Sequence to sequence learning for visual object tracking,

    X. Chen, H. Peng, D. Wang, H. Lu, and H. Hu, “Seqtrack: Sequence to sequence learning for visual object tracking,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2023, pp. 14 572–14 581

    2023

  21. [30]

    Autoregressive visual tracking,

    X. Wei, Y . Bai, Y . Zheng, D. Shi, and Y . Gong, “Autoregressive visual tracking,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 9697–9706

    2023

  22. [31]

    Hiptrack: Visual tracking with historical prompts,

    W. Cai, Q. Liu, and Y . Wang, “Hiptrack: Visual tracking with historical prompts,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 19 258–19 267

    2024

  23. [32]

    Tracking via object reflectance using a hyperspectral video camera,

    H. Van Nguyen, A. Banerjee, and R. Chellappa, “Tracking via object reflectance using a hyperspectral video camera,” in2010 IEEE Com- puter Society Conference on Computer Vision and Pattern Recognition- Workshops. IEEE, 2010, pp. 44–51

    2010

  24. [33]

    Spatial–spectral weighted and regularized tensor sparse correlation filter for object tracking in hyper- spectral videos,

    Z. Hou, W. Li, J. Zhou, and R. Tao, “Spatial–spectral weighted and regularized tensor sparse correlation filter for object tracking in hyper- spectral videos,”IEEE Transactions on Geoscience and Remote Sensing, vol. 60, pp. 1–12, 2022

    2022

  25. [34]

    Siamohot: A lightweight dual siamese network for onboard hyperspectral object tracking via joint spatial-spectral knowledge distillation,

    C. Sun, X. Wang, Z. Liu, Y . Wan, L. Zhang, and Y . Zhong, “Siamohot: A lightweight dual siamese network for onboard hyperspectral object tracking via joint spatial-spectral knowledge distillation,”IEEE Trans- actions on Geoscience and Remote Sensing, 2023

    2023

  26. [35]

    Spectral- spatial-aware transformer fusion network for hyperspectral object track- ing,

    Y . Wang, Y . Liu, G. Zhang, Y . Su, S. Zhang, and S. Mei, “Spectral- spatial-aware transformer fusion network for hyperspectral object track- ing,” in2022 12th Workshop on Hyperspectral Imaging and Signal Processing: Evolution in Remote Sensing (WHISPERS). IEEE, 2022, pp. 1–5

    2022

  27. [36]

    Spatial–spectral cross- correlation embedded dual-transfer network for object tracking using hyperspectral videos,

    J. Lei, P. Liu, W. Xie, L. Gao, Y . Li, and Q. Du, “Spatial–spectral cross- correlation embedded dual-transfer network for object tracking using hyperspectral videos,”Remote Sensing, vol. 14, no. 15, p. 3512, 2022

    2022

  28. [37]

    Bae-net: A band attention aware ensemble network for hyperspectral object tracking,

    Z. Li, F. Xiong, J. Zhou, J. Wang, J. Lu, and Y . Qian, “Bae-net: A band attention aware ensemble network for hyperspectral object tracking,” in 2020 IEEE international Conference on image processing (ICIP). IEEE, 2020, pp. 2106–2110

    2020

  29. [38]

    A spectral–spatial transformer fu- sion method for hyperspectral video tracking,

    Y . Wang, Y . Liu, M. Ma, and S. Mei, “A spectral–spatial transformer fu- sion method for hyperspectral video tracking,”Remote Sensing, vol. 15, no. 7, p. 1735, 2023

    2023

  30. [39]

    Rgbt tracking via all-layer multimodal interactions with progressive fusion mamba,

    A. Lu, W. Wang, C. Li, J. Tang, and B. Luo, “Rgbt tracking via all-layer multimodal interactions with progressive fusion mamba,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 39, no. 6, 2025, pp. 5793–5801

    2025

  31. [40]

    Smamba: Sparse mamba for event-based object detection,

    N. Yang, Y . Wang, Z. Liu, M. Li, Y . An, and X. Zhao, “Smamba: Sparse mamba for event-based object detection,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 39, no. 9, 2025, pp. 9229– 9237

    2025

  32. [41]

    Trackmamba: Mamba-transformer tracking,

    J. Zhang, C. Liang, Y . Cui, X. Shu, G. Wu, and L. Wang, “Trackmamba: Mamba-transformer tracking,” 2024. [Online]. Available: https://openreview.net/forum?id=V7QRVEZ0le

    2024

  33. [42]

    Mambaevt: Event stream based visual object tracking using state space model,

    X. Wang, C. Wang, S. Wang, X. Wang, Z. Zhao, L. Zhu, and B. Jiang, “Mambaevt: Event stream based visual object tracking using state space model,”IEEE Transactions on Circuits and Systems for Video Technology, 2025

    2025

  34. [43]

    Root mean square layer normalization,

    B. Zhang and R. Sennrich, “Root mean square layer normalization,” Advances in neural information processing systems, vol. 32, 2019

    2019

  35. [44]

    Sigmoid-weighted linear units for neural network function approximation in reinforcement learning,

    S. Elfwing, E. Uchibe, and K. Doya, “Sigmoid-weighted linear units for neural network function approximation in reinforcement learning,” Neural networks, vol. 107, pp. 3–11, 2018

    2018

  36. [45]

    Su- track: Towards simple and unified single object tracking,

    X. Chen, B. Kang, W. Geng, J. Zhu, Y . Liu, D. Wang, and H. Lu, “Su- track: Towards simple and unified single object tracking,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 39, no. 2, 2025, pp. 2239–2247. IEEE TRANSACTIONS ON IMAGE PROCESSING 13

    2025

  37. [46]

    Cornernet: Detecting objects as paired keypoints,

    H. Law and J. Deng, “Cornernet: Detecting objects as paired keypoints,” inProceedings of the European conference on computer vision (ECCV), 2018, pp. 734–750

    2018

  38. [47]

    Generalized intersection over union: A metric and a loss for bounding box regression,

    H. Rezatofighi, N. Tsoi, J. Gwak, A. Sadeghian, I. Reid, and S. Savarese, “Generalized intersection over union: A metric and a loss for bounding box regression,” inProceedings of the IEEE/CVF conference on com- puter vision and pattern recognition, 2019, pp. 658–666

    2019

  39. [48]

    Hivit: A simpler and more efficient design of hierarchical vision transformer,

    X. Zhang, Y . Tian, L. Xie, W. Huang, Q. Dai, Q. Ye, and Q. Tian, “Hivit: A simpler and more efficient design of hierarchical vision transformer,” inThe eleventh international conference on learning representations, 2023

    2023

  40. [49]

    Siamfc++: Towards robust and accurate visual tracking with target estimation guidelines,

    Y . Xu, Z. Wang, Z. Li, Y . Yuan, and G. Yu, “Siamfc++: Towards robust and accurate visual tracking with target estimation guidelines,” in Proceedings of the AAAI conference on artificial intelligence, vol. 34, no. 07, 2020, pp. 12 549–12 556

    2020

  41. [50]

    Siamcar: Siamese fully convolutional classification and regression for visual tracking,

    D. Guo, J. Wang, Y . Cui, Z. Wang, and S. Chen, “Siamcar: Siamese fully convolutional classification and regression for visual tracking,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2020, pp. 6269–6277

    2020

  42. [51]

    Joint feature learning and relation modeling for tracking: A one-stream framework,

    B. Ye, H. Chang, B. Ma, S. Shan, and X. Chen, “Joint feature learning and relation modeling for tracking: A one-stream framework,” inEuropean Conference on Computer Vision. Springer, 2022, pp. 341– 357

    2022

  43. [52]

    Towards real- world visual tracking with temporal contexts,

    Z. Cao, Z. Huang, L. Pan, S. Zhang, Z. Liu, and C. Fu, “Towards real- world visual tracking with temporal contexts,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 45, no. 12, pp. 15 834– 15 849, 2023

    2023

  44. [53]

    Separable self and mixed attention trans- formers for efficient object tracking,

    G. Y . Gopal and M. A. Amer, “Separable self and mixed attention trans- formers for efficient object tracking,” inProceedings of the IEEE/CVF winter conference on applications of computer vision, 2024, pp. 6708– 6717

    2024

  45. [54]

    Autoregressive queries for adaptive tracking with spatio-temporal trans- formers,

    J. Xie, B. Zhong, Z. Mo, S. Zhang, L. Shi, S. Song, and R. Ji, “Autoregressive queries for adaptive tracking with spatio-temporal trans- formers,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 19 300–19 309

    2024

  46. [55]

    Artrackv2: Prompting autore- gressive tracker where to look and how to describe,

    Y . Bai, Z. Zhao, Y . Gong, and X. Wei, “Artrackv2: Prompting autore- gressive tracker where to look and how to describe,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2024, pp. 19 048–19 057

    2024

  47. [56]

    Promptvt: Prompting for efficient and accurate visual tracking,

    M. Zhang, Q. Zhang, W. Song, D. Huang, and Q. He, “Promptvt: Prompting for efficient and accurate visual tracking,”IEEE Transactions on Circuits and Systems for Video Technology, 2024

    2024

  48. [57]

    Transformer- based band regrouping with feature refinement for hyperspectral object tracking,

    H. Wang, W. Li, X.-G. Xia, Q. Du, J. Tian, and Q. Shen, “Transformer- based band regrouping with feature refinement for hyperspectral object tracking,”IEEE Transactions on Geoscience and Remote Sensing, 2024

    2024

  49. [58]

    Phtrack: Prompting for hyperspectral video tracking,

    Y . Chen, Y . Tang, X. Su, J. Li, Y . Xiao, J. He, and Q. Yuan, “Phtrack: Prompting for hyperspectral video tracking,”IEEE Transactions on Geoscience and Remote Sensing, 2024

    2024

  50. [59]

    Sense: Hyperspectral video object tracker via fusing material and motion cues,

    Y . Chen, Q. Yuan, Y . Tang, Y . Xiao, J. He, and Z. Liu, “Sense: Hyperspectral video object tracker via fusing material and motion cues,” Information Fusion, vol. 109, p. 102395, 2024

    2024

  51. [60]

    Hyperspectral video tracking with spectral-spatial fusion and memory enhancement,

    Y . Chen, Q. Yuan, H. Xie, Y . Tang, Y . Xiao, J. He, R. Guan, X. Liu, and L. Zhang, “Hyperspectral video tracking with spectral-spatial fusion and memory enhancement,”IEEE Transactions on Image Processing, 2025

    2025

  52. [61]

    Hyperspectral object tracking with spectral information prompt,

    G. He, L. Gao, L. Chen, Y . Jiang, W. Xie, and Y . Li, “Hyperspectral object tracking with spectral information prompt,”IEEE Transactions on Circuits and Systems for Video Technology, 2025

    2025

  53. [62]

    Hotmoe: Exploring sparse mixture-of-experts for hyperspectral object tracking,

    W. Sun, Y . Tan, J. Li, S. Hou, X. Li, Y . Shao, Z. Wang, and B. Song, “Hotmoe: Exploring sparse mixture-of-experts for hyperspectral object tracking,”IEEE Transactions on Multimedia, 2025

    2025

  54. [63]

    Multi-domain universal representation learning for hyperspectral object tracking,

    Z. Li, F. Xiong, J. Lu, J. Wang, D. Chen, J. Zhou, and Y . Qian, “Multi-domain universal representation learning for hyperspectral object tracking,”Pattern Recognition, vol. 162, p. 111389, 2025

    2025

  55. [64]

    Ssttrack: A unified hyperspectral video tracking framework via modeling spectral-spatial-temporal conditions,

    Y . Chen, Q. Yuan, Y . Tang, Y . Xiao, J. He, T. Han, Z. Liu, and L. Zhang, “Ssttrack: A unified hyperspectral video tracking framework via modeling spectral-spatial-temporal conditions,”Information Fusion, vol. 114, p. 102658, 2025

    2025

  56. [65]

    Domain adaptation- aware transformer for hyperspectral object tracking,

    Y . Wu, L. Jiao, X. Liu, F. Liu, S. Yang, and L. Li, “Domain adaptation- aware transformer for hyperspectral object tracking,”IEEE Transactions on Circuits and Systems for Video Technology, vol. 34, no. 9, pp. 8041– 8052, 2024

    2024

  57. [66]

    Ubstrack: Unified band selection and multi-model ensemble for hyperspectral object tracking,

    M. A. Islam, J. Zhou, W. Xing, Y . Gao, and K. K. Paliwal, “Ubstrack: Unified band selection and multi-model ensemble for hyperspectral object tracking,”IEEE Transactions on Geoscience and Remote Sensing, 2025

    2025

  58. [67]

    A transformer- based network for hyperspectral object tracking,

    L. Gao, L. Chen, P. Liu, Y . Jiang, W. Xie, and Y . Li, “A transformer- based network for hyperspectral object tracking,”IEEE Transactions on Geoscience and Remote Sensing, 2023. Long Gao(Member, IEEE) received the B.S. degree and the M.S. degree in Control Theory and Control ...

    2023

Pith tools

Reviewed August 4, 2026 · model on record in the stance chip above.