Pith. sign in

REVIEW 3 major objections 6 minor 70 references

A Survey on Mamba Architecture for Vision Applications

T0 review · 3 major / 6 minor · reviewed 2026-08-08 · deepseek-v4-flash

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

Pith's one-line read The paper surveys Mamba-based state-space models for vision and argues that, with bidirectional and multi-directional scanning, they offer a linear-complexity alternative to Transformers competitive on classification, segmentation…

desk verdict A useful architectural survey whose performance comparison and recommendations are undermined by non-comparable benchmark data; fixable but needs a major revision. read the letter →

arxiv 2502.07161 v1 pith:7MNVDWSE submitted 2025-02-11 cs.CV cs.AI

Fady Ibrahim , Guangjun Liu , Guanghui Wang This is my paper · ORCID
classification cs.CVcs.AI
keywords MambaStateSpaceModelsVisionVideoSelectiveScanningBidirectionalSequenceModelingVisualSurvey
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 surveys the emerging family of Mamba-based architectures for computer vision, aiming to establish that state-space models (SSMs) with linear sequence complexity can serve as a practical alternative to Transformers, which suffer from quadratic attention cost. It centers on Vision Mamba (ViM) and VideoMamba as core models, and organizes recent architectural innovations—bidirectional scanning, cross-scan modules, position embeddings, hierarchical designs, token fusion, and non-causal state space duality—into a coherent design space. The survey's comparative tables show that top Mamba variants achieve competitive accuracy on ImageNet classification, ADE20K segmentation, MS-COCO detection, and video action recognition, with structure-aware state fusion and non-causal SSD leading the rankings. The paper concludes that these advancements position Mamba as a promising backbone for vision research and applications.

What carries the argument

The load-bearing mechanism is the selective state-space model block, inherited from the original Mamba (S6), in which the state-transition parameters B, C, and Δ are made input-dependent through linear projections, so the model can selectively retain or forget information along the sequence. For vision, the survey identifies the critical extension as the scanning mechanism: how the 2D image or 3D video is flattened into a 1D token sequence and scanned (e.g., bidirectional sweeping, four-direction cross-scan, windowed local scan, zigzag continuous scan, or spatiotemporal scans). This scanning choice, together with structural add-ons such as position embeddings, register tokens, token fusion, and non-causal state space duality, determines how well the model balances global context, local detail, and efficiency.

What would settle it

Run Spatial-Mamba-S, Vmamba-S, VSSD-S, and a comparable ViT backbone under identical training recipe, input resolution, and evaluation protocol on ImageNet-1k, ADE20K, and MS-COCO: if the Mamba models no longer lead, the survey's practical recommendations are an artifact of pooling numbers. A cheaper check: measure wall-clock latency at sequence lengths 1024, 4096, and 16384; if the Mamba models do not show the predicted linear-time advantage over an optimized attention implementation, the core efficiency motivation weakens.

Watch

Extended reading notes

Core claim

The central claim is that Mamba, originally a 1D selective state-space model for language, can be adapted to 2D images and 3D video by replacing unidirectional causal scanning with bidirectional or multi-directional scanning and by enriching the SSM with spatial structure. The paper argues that these adaptations—exemplified by ViM's forward-backward SSMs with shared parameters, VMamba's cross-scan module that flattens the image along four directions, and VideoMamba's spatiotemporal scans—let Mamba capture both local and global visual context while keeping memory and compute linear in sequence length. In the surveyed benchmarks, the best accuracy comes from models that fuse state variables across spatial neighborhoods (Spatial-Mamba's Structure-Aware State Fusion) or remove the causal mask entirely (VSSD's non-causal state space duality), suggesting that the scanning strategy, not the SSM core, is the main design lever for vision.

Load-bearing premise

The survey's performance rankings assume that benchmark numbers drawn from different papers are directly comparable, even though training schedules, input resolutions, downstream head design, and even datasets (in one case CIFAR-100 vs ImageNet) are not standardized across entries.

Editorial extensions

If this is right

  • If Mamba backbones hold their reported accuracy, vision models can process higher-resolution images and longer video clips without the quadratic memory growth of attention, enabling deployment on memory-limited hardware.
  • The identified design axes—scanning strategy, state fusion, and causal vs. non-causal dynamics—give a concrete recipe for building new vision backbones, not just tuning existing ones.
  • Efficiency-oriented variants such as EfficientViM indicate that Mamba can be compressed to roughly 20 million parameters with moderate accuracy loss, a plausible path to on-device vision.
  • Combining Mamba with CNN or Transformer components, as in the surveyed hybrid models, may close the remaining performance gap on video benchmarks while preserving linear scaling.

Reading between the lines

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

  • The benchmark tables are not controlled experiments: models are drawn from different papers with different training schedules and datasets (e.g., Famba-V is evaluated on CIFAR-100, not ImageNet), so a fair head-to-head could reorder the recommended choices.
  • The taxonomy suggests that the field is converging on a small set of scanning patterns; a native 2D SSM that avoids flattening altogether (as hinted by non-causal SSD) could render scan design obsolete.
  • Token fusion and register tokens, borrowed from ViT efficiency tricks, may be largely orthogonal to scanning, so stacking them on top of Spatial-Mamba or VSSD could yield further gains.
  • The video results still trail transformer-based methods on some benchmarks, so the 'promising' verdict rests more on scaling outlook than on demonstrated parity at the time of the survey.
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 / 6 minor

Summary. This paper surveys state-space-model (SSM) architectures for computer vision, centered on Vision Mamba (ViM) and VideoMamba. It introduces the Mamba block and selective-scanning formulation, compares ViM with VideoMamba, and catalogs recent architectural variants: VMamba, LocalMamba, PlainMamba, SpatialMamba, Famba-V, VSSD, EfficientViM, Mamba-R, Hi-Mamba, and VideoMambaPro. It then tabulates reported accuracy and efficiency numbers on ImageNet-1k, ADE20K, MS-COCO, and video action-recognition benchmarks and offers practical recommendations for model selection. The paper concludes with challenges and future research directions for Mamba in vision.

Significance. If the benchmark comparisons were reliable, the survey would provide a useful task-specific selection guide for Mamba-based vision backbones. The paper's main strengths are organizational: the taxonomy of scanning mechanisms and the compact visual summaries of recent blocks make the literature more accessible, and the SSM background in Section II is standard. The central claim that Mamba is a promising vision architecture is modest and consistent with the cited literature. However, the advertised contribution, namely practical recommendations for selecting the best architectures, depends on the comparative analysis in Section V, and that analysis is not a controlled comparison. As detailed below, the tables mix different datasets and model sizes and report non-comparable FLOPS, so the recommendations are not currently evidence-based. The paper offers no new experimental results or derivations, which is normal for a survey, but it means the value of the paper must come from accurate synthesis rather than novel contributions.

major comments (3)
  1. [Section V-A, Table I] The row for 'Famba-V applied to ViM-S' reports 75.2% Top-1 accuracy on CIFAR-100 in a table that is introduced as an ImageNet-1k comparison. The text explicitly uses this row to conclude that 'bidirectional scanning with token fusion is less effective for image classification.' Because accuracies from different datasets are not comparable, this row cannot support that conclusion. The table should be restricted to ImageNet-1k, or the row should be removed or clearly separated with a dataset-specific caveat.
  2. [Section V-C, Table III] The EfficientViM-M4 row reports 4.1G FLOPS for object detection with Mask R-CNN, the same value as the classification FLOPS in Table I. This is implausible for a detection pipeline: the other detection backbones in the same table are listed at 315-542G FLOPS. The text's claim that EfficientViM-M4 'despite its efficiency (4.1G FLOPS), sacrifices accuracy' therefore relies on a mislabeled or non-comparable number. The detection FLOPS should be corrected, or the efficiency comparison should be removed.
  3. [Section V and Section V-E] The comparison is introduced as including 'only those of a similar size,' but the tables mix model families and sizes: Table I includes EfficientViM-M4 (21.3M) and PlainMamba-L2 (25.7M) alongside S-size models, and Table II includes VSSD-T (tiny). Reported FLOPS are not normalized for input resolution, training schedule, or framework, so cross-model efficiency rankings are unreliable. Because Section V-E directly uses these numbers to recommend Spatial-Mamba-S and Vmamba-S as the best classification and segmentation choices and VSSD-S and LocalMamba as the best balance across tasks, the recommendations are not supported by the evidence as presented. The authors should either match training setups and model scales or explicitly label the tables as non-comparable literature values and remove or soften the rankings.
minor comments (6)
  1. [Section II-B] Equations (4)-(5) are hard to parse because 'SC' and 'S_delta' appear without subscripts and the symbol B is used both for the batch size and for the SSM input matrix; please write the input-dependent parameters explicitly as, for example, B_t = S_B(x_t), C_t = S_C(x_t), and delta_t = S_delta(x_t).
  2. [Section III-B] The statement that VideoMambaPro 'establish[es] it as a strong competitor to transformer models' is not supported by Table IV, which contains no transformer baselines; please add baselines or soften the wording.
  3. [Section V-C] The metric label 'APbox75' is nonstandard; use 'AP75' or 'AP at IoU=0.75' and define it in the text.
  4. [Figure 4] Subfigures (d)-(g) are very small and their internal labels are difficult to read; enlarging them or splitting them across multiple figures would improve usability.
  5. [References] Reference [4] should include the full author list, and references [1] and [2] are unusual citations for convolutional networks in the introduction; standard CNN references would be more appropriate.
  6. [Table III] The row for Vision Mamba (ViM-Ti) has empty FLOPS and AP entries; if these numbers are unavailable in the original paper, it may be cleaner to omit the row.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the survey's claims rest on externally published results, and the few coauthor self-citations are background references, not load-bearing premises.

full rationale

This is a survey paper, not a derivation, so the pattern-based circularity tests mostly do not apply. Its central assertion that Mamba is a promising vision architecture is supported by externally published results from ViM, VideoMamba, Vmamba, and other cited works; the survey introduces no fitted parameters, no equations whose outputs are defined by their inputs, and no prediction generated from its own data. The only self-citations are background references in the introduction (e.g., refs. [3], [7], [8], and [9] with coauthor G. Wang), and none supplies a load-bearing premise for the survey's taxonomy or recommendations. The performance-comparison tables do have a serious comparability problem: Table I places a CIFAR-100 result (Famba-V, 75.2%) within the ImageNet-1k discussion, and the 'similar size' selection criterion is applied loosely across tables. However, that is a correctness and evidence-quality concern, not circularity, because the numbers are transcribed from independent external papers rather than manufactured by the survey itself. No circular step can be exhibited, so the appropriate finding is no significant circularity.

Assumptions & free parameters 0 free parameters · 2 assumptions · 0 invented entities

Because the paper is a review, it introduces no free parameters and no invented entities. Its conclusions rest on the trustworthiness of cited benchmark results, the correctness of the reproduced SSM formulas, and the comparability of heterogeneous evaluation setups. The comparability assumption is the most fragile because it is visibly violated in at least one table.

assumptions (2)
  • domain assumption Reported benchmark numbers from cited papers are accurate, use comparable training and evaluation protocols, and can be ranked directly.
    The comparative conclusions in Section V rank models using numbers extracted from many papers. The survey does not rerun experiments, normalize training budgets, or adjust for dataset differences, yet it treats the numbers as directly comparable.
  • standard math The state-space discretization formulas in Section II correctly describe Mamba and its variants.
    Equations (1) through (6) are presented as background from Mamba and Vision Mamba. The survey does not prove them, and the notation in Equation (5) is sloppy, but the survey relies on these formulas when describing the architectures.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Survey on Mamba Architecture for Vision Applications." pith.science (2026). https://pith.science/paper/7MNVDWSE

@misc{pith2026250207161,
  author       = {Pith},
  title        = {Pith review of: A Survey on Mamba Architecture for Vision Applications},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/7MNVDWSE}},
  note         = {Machine review of arXiv:2502.07161}
}
read the original abstract

Transformers have become foundational for visual tasks such as object detection, semantic segmentation, and video understanding, but their quadratic complexity in attention mechanisms presents scalability challenges. To address these limitations, the Mamba architecture utilizes state-space models (SSMs) for linear scalability, efficient processing, and improved contextual awareness. This paper investigates Mamba architecture for visual domain applications and its recent advancements, including Vision Mamba (ViM) and VideoMamba, which introduce bidirectional scanning, selective scanning mechanisms, and spatiotemporal processing to enhance image and video understanding. Architectural innovations like position embeddings, cross-scan modules, and hierarchical designs further optimize the Mamba framework for global and local feature extraction. These advancements position Mamba as a promising architecture in computer vision research and applications.

Figures

Figures reproduced from arXiv: 2502.07161 by the authors.

Figure 2
Figure 2. Single direction scanning mechanism on patch embed [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. 3D Spatiotemporal Bidirectional Scanning. Video [PITH_FULL_IMAGE:figures/full_fig_p002_3.png] view at source ↗
Figure 4
Figure 4. Comparison of Mamba architectures for 2D image applications; (a) Vision Mamba, (b) Vmamba, (c) LocalMamba [PITH_FULL_IMAGE:figures/full_fig_p003_4.png] view at source ↗
Figures from the paper (1 more)
Figure 6
Figure 6. Figure 6: Scanning Mechanisms; (a) and (b) are applied on [PITH_FULL_IMAGE:figures/full_fig_p005_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

70 extracted references · 23 canonical work pages

  1. [1]

    What makes con- volutional models great on long sequence modeling?

    Y . Li, T. Cai, Y . Zhang, D. Chen, and D. Dey, “What makes con- volutional models great on long sequence modeling?” arXiv preprint arXiv:2210.09298, 2022

    arXiv 2022

  2. [2]

    Time-aware large kernel convolutions,

    V . Lioutas and Y . Guo, “Time-aware large kernel convolutions,” in International Conference on Machine Learning . PMLR, 2020, pp. 6172–6183

    work page 2020

  3. [3]

    Depth-wise convolutions in vision transformers for efficient training on small datasets,

    T. Zhang, W. Xu, B. Luo, and G. Wang, “Depth-wise convolutions in vision transformers for efficient training on small datasets,” Neurocom- puting, vol. 617, p. 128998, 2025

    work page 2025

  4. [4]

    Attention is all you need,

    A. Vaswani, “Attention is all you need,” Advances in Neural Information Processing Systems, 2017

    2017

  5. [5]

    An image is worth 16x16 words: Transformers for image recognition at scale,

    A. Dosovitskiy, “An image is worth 16x16 words: Transformers for image recognition at scale,” arXiv preprint arXiv:2010.11929 , 2020

    arXiv 2010

  6. [6]

    Predicting therapeutic response to hypoglossal nerve stimulation using deep learning,

    R. Alapati, B. Renslo, L. Jackson, H. Moradi, J. R. Oliver, M. Chowd- hury, T. Vyas, A. Bon Nieves, A. Lawrence, S. F. Wagoner et al. , “Predicting therapeutic response to hypoglossal nerve stimulation using deep learning,” The Laryngoscope , vol. 134, no. 12, pp. 5210–5216, 2024

    work page 2024

  7. [7]

    Aphid cluster recognition and detection in the wild using deep learning models,

    T. Zhang, K. Li, X. Chen, C. Zhong, B. Luo, I. Grijalva, B. McCornack, D. Flippo, A. Sharda, and G. Wang, “Aphid cluster recognition and detection in the wild using deep learning models,” Scientific Reports , vol. 13, no. 1, p. 13410, 2023

    work page 2023

  8. [8]

    A new dataset and comparative study for aphid cluster detection and segmentation in sorghum fields,

    R. Rahman, C. Indris, G. Bramesfeld, T. Zhang, K. Li, X. Chen, I. Grijalva, B. McCornack, D. Flippo, A. Sharda et al., “A new dataset and comparative study for aphid cluster detection and segmentation in sorghum fields,” Journal of Imaging , vol. 10, no. 5, p. 114, 2024

    work page 2024

Show all 70 references
  1. [9]

    Edge-aware multi-task network for integrating quantification segmentation and uncertainty prediction of liver tumor on multi-modality non-contrast mri,

    X. Xiao, Q. V . Hu, and G. Wang, “Edge-aware multi-task network for integrating quantification segmentation and uncertainty prediction of liver tumor on multi-modality non-contrast mri,” inInternational Confer- ence on Medical Image Computing and Computer-Assisted Intervention...

    2023

  2. [10]

    Deep residual learning for image recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” in Proceedings of the IEEE conference on computer vision and pattern recognition , 2016, pp. 770–778

    2016

  3. [11]

    Is space-time attention all you need for video understanding?

    G. Bertasius, H. Wang, and L. Torresani, “Is space-time attention all you need for video understanding?” in ICML, vol. 2, no. 3, 2021, p. 4

    2021

  4. [12]

    Generating long sequences with sparse transformers,

    R. Child, S. Gray, A. Radford, and I. Sutskever, “Generating long sequences with sparse transformers,” arXiv preprint arXiv:1904.10509 , 2019

    1904 arXiv

  5. [13]

    Xcit: Cross- covariance image transformers,

    A. Ali, H. Touvron, M. Caron, P. Bojanowski, M. Douze, A. Joulin, I. Laptev, N. Neverova, G. Synnaeve, J. Verbeek et al. , “Xcit: Cross- covariance image transformers,” Advances in neural information pro- cessing systems, vol. 34, pp. 20 014–20 027, 2021

    2021

  6. [14]

    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

  7. [15]

    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

  8. [16]

    Videomamba: State space model for efficient video understanding,

    K. Li, X. Li, Y . Wang, Y . He, Y . Wang, L. Wang, and Y . Qiao, “Videomamba: State space model for efficient video understanding,” in European Conference on Computer Vision . Springer, 2025, pp. 237– 255

    2025

  9. [17]

    A survey of mamba,

    H. Qu, L. Ning, R. An, W. Fan, T. Derr, H. Liu, X. Xu, and Q. Li, “A survey of mamba,” arXiv preprint arXiv:2408.01129 , 2024

    2024 arXiv

  10. [18]

    Mamba-360: Survey of state space models as transformer alternative for long sequence modelling: Methods, applications, and challenges,

    B. N. Patro and V . S. Agneeswaran, “Mamba-360: Survey of state space models as transformer alternative for long sequence modelling: Methods, applications, and challenges,” arXiv preprint arXiv:2404.16112 , 2024

    2024 arXiv

  11. [19]

    Medmamba: Vision mamba for medical image classification,

    Y . Yue and Z. Li, “Medmamba: Vision mamba for medical image classification,” arXiv preprint arXiv:2403.03849 , 2024

    2024 arXiv

  12. [20]

    A comprehensive survey of mamba architectures for medical image analysis: Classification, segmentation, restoration and beyond,

    S. Bansal, S. Madisetty, M. Z. U. Rehman, C. S. Raghaw, G. Duggal, N. Kumar et al., “A comprehensive survey of mamba architectures for medical image analysis: Classification, segmentation, restoration and beyond,” arXiv preprint arXiv:2410.02362 , 2024

  13. [21]

    A survey on vision mamba: Models, applications and challenges,

    R. Xu, S. Yang, Y . Wang, B. Du, and H. Chen, “A survey on vision mamba: Models, applications and challenges,” arXiv preprint arXiv:2404.18861, 2024

    2024 arXiv

  14. [22]

    A survey on visual mamba,

    H. Zhang, Y . Zhu, D. Wang, L. Zhang, T. Chen, Z. Wang, and Z. Ye, “A survey on visual mamba,” Applied Sciences, vol. 14, no. 13, p. 5683, 2024

    2024

  15. [23]

    Comparison between first-order hold with zero-order hold in discretization of input-delay nonlinear systems,

    Z. Zhang and K. T. Chong, “Comparison between first-order hold with zero-order hold in discretization of input-delay nonlinear systems,” in 2007 International Conference on Control, Automation and Systems . IEEE, 2007, pp. 2892–2896

    2007

  16. [24]

    Vmamba: Visual state space model,

    Y . Liu, Y . Tian, Y . Zhao, H. Yu, L. Xie, Y . Wang, Q. Ye, and Y . Liu, “Vmamba: Visual state space model,” arXiv preprint arXiv:2401.10166, 2024

    2024 arXiv

  17. [25]

    Mamba-r: Vision mamba also needs registers,

    F. Wang, J. Wang, S. Ren, G. Wei, J. Mei, W. Shao, Y . Zhou, A. Yuille, and C. Xie, “Mamba-r: Vision mamba also needs registers,” arXiv preprint arXiv:2405.14858, 2024

    2024 arXiv

  18. [26]

    Vssd: Vision mamba with non- causal state space duality,

    Y . Shi, M. Dong, M. Li, and C. Xu, “Vssd: Vision mamba with non- causal state space duality,” arXiv preprint arXiv:2407.18559 , 2024

    2024 arXiv

  19. [27]

    Localmamba: Visual state space model with windowed selective scan,

    T. Huang, X. Pei, S. You, F. Wang, C. Qian, and C. Xu, “Localmamba: Visual state space model with windowed selective scan,” arXiv preprint arXiv:2403.09338, 2024

    2024 arXiv

  20. [28]

    Mamba2d: A natively multi-dimensional state-space model for vision tasks,

    E. Baty, A. H. D ´ıaz, C. Bridges, R. Davidson, S. Eckersley, and S. Hadfield, “Mamba2d: A natively multi-dimensional state-space model for vision tasks,” arXiv preprint arXiv:2412.16146 , 2024

    2024

  21. [29]

    Efficientvim: Efficient vision mamba with hidden state mixer based state space duality,

    S. Lee, J. Choi, and H. J. Kim, “Efficientvim: Efficient vision mamba with hidden state mixer based state space duality,” arXiv preprint arXiv:2411.15241, 2024

    2024 arXiv

  22. [30]

    Transformers are ssms: Generalized models and ef- ficient algorithms through structured state space duality,

    T. Dao and A. Gu, “Transformers are ssms: Generalized models and ef- ficient algorithms through structured state space duality,” arXiv preprint arXiv:2405.21060, 2024

    2024 arXiv

  23. [31]

    Vivit: A video vision transformer,

    A. Arnab, M. Dehghani, G. Heigold, C. Sun, M. Lu ˇci´c, and C. Schmid, “Vivit: A video vision transformer,” in Proceedings of the IEEE/CVF international conference on computer vision , 2021, pp. 6836–6846

    2021

  24. [32]

    Videomambapro: A leap forward for mamba in video understanding,

    H. Lu, A. A. Salah, and R. Poppe, “Videomambapro: A leap forward for mamba in video understanding,” arXiv preprint arXiv:2406.19006 , 2024

    2024 arXiv

  25. [33]

    Video mamba suite: State space model as a versatile alternative for video understanding,

    G. Chen, Y . Huang, J. Xu, B. Pei, Z. Chen, Z. Li, J. Wang, K. Li, T. Lu, and L. Wang, “Video mamba suite: State space model as a versatile alternative for video understanding,” arXiv preprint arXiv:2403.09626 , 2024

    2024 arXiv

  26. [34]

    Vivim: a video vision mamba for medical video object segmentation,

    Y . Yang, Z. Xing, and L. Zhu, “Vivim: a video vision mamba for medical video object segmentation,” arXiv preprint arXiv:2401.14168 , 2024

    2024 arXiv

  27. [35]

    Imagenet classification with deep convolutional neural networks,

    A. Krizhevsky, I. Sutskever, and G. E. Hinton, “Imagenet classification with deep convolutional neural networks,” Advances in neural informa- tion processing systems , vol. 25, 2012

    2012

  28. [36]

    Object detection with deep learning: A review,

    Z.-Q. Zhao, P. Zheng, S.-t. Xu, and X. Wu, “Object detection with deep learning: A review,” IEEE transactions on neural networks and learning systems, vol. 30, no. 11, pp. 3212–3232, 2019

    2019

  29. [37]

    Unified perceptual parsing for scene understanding,

    T. Xiao, Y . Liu, B. Zhou, Y . Jiang, and J. Sun, “Unified perceptual parsing for scene understanding,” in Proceedings of the European conference on computer vision (ECCV) , 2018, pp. 418–434

    2018

  30. [38]

    Meth- ods and datasets on semantic segmentation: A review,

    H. Yu, Z. Yang, L. Tan, Y . Wang, W. Sun, M. Sun, and Y . Tang, “Meth- ods and datasets on semantic segmentation: A review,” Neurocomputing, vol. 304, pp. 82–103, 2018

    2018

  31. [39]

    A brief survey on semantic segmentation with deep learning,

    S. Hao, Y . Zhou, and Y . Guo, “A brief survey on semantic segmentation with deep learning,” Neurocomputing, vol. 406, pp. 302–321, 2020

    2020

  32. [40]

    Neural architecture search based global-local vision mamba for palm-vein recognition,

    H. Qin, Y . Fu, J. Chen, M. A. El-Yacoubi, X. Gao, and F. Xi, “Neural architecture search based global-local vision mamba for palm-vein recognition,” arXiv preprint arXiv:2408.05743 , 2024

    2024 arXiv

  33. [41]

    Venturing into uncharted waters: The navigation compass from transformer to mamba,

    Y . Zou, Y . Chen, Z. Li, L. Zhang, and H. Zhao, “Venturing into uncharted waters: The navigation compass from transformer to mamba,” arXiv preprint arXiv:2406.16722, 2024

    2024 arXiv

  34. [42]

    Mambabev: An efficient 3d detection model with mamba2,

    Z. You, H. Wang, Q. Zhao, and J. Wang, “Mambabev: An efficient 3d detection model with mamba2,” arXiv preprint arXiv:2410.12673, 2024

    2024 arXiv

  35. [43]

    Plainmamba: Improving non-hierarchical mamba in visual recognition,

    C. Yang, Z. Chen, M. Espinosa, L. Ericsson, Z. Wang, J. Liu, and E. J. Crowley, “Plainmamba: Improving non-hierarchical mamba in visual recognition,” arXiv preprint arXiv:2403.17695 , 2024

    2024 arXiv

  36. [44]

    Spatial-mamba: Effective visual state space models via structure-aware state fusion,

    C. Xiao, M. Li, Z. Zhang, D. Meng, and L. Zhang, “Spatial-mamba: Effective visual state space models via structure-aware state fusion,” arXiv preprint arXiv:2410.15091 , 2024

    2024 arXiv

  37. [45]

    Famba-v: Fast vision mamba with cross-layer token fusion,

    H. Shen, Z. Wan, X. Wang, and M. Zhang, “Famba-v: Fast vision mamba with cross-layer token fusion,” arXiv preprint arXiv:2409.09808 , 2024

    2024 arXiv

  38. [46]

    Towards accurate post-training quantization for vision transformer,

    Y . Ding, H. Qin, Q. Yan, Z. Chai, J. Liu, X. Wei, and X. Liu, “Towards accurate post-training quantization for vision transformer,” in Proceedings of the 30th ACM international conference on multimedia , 2022, pp. 5380–5388

    2022

  39. [47]

    Token merging: Your vit but faster,

    D. Bolya, C.-Y . Fu, X. Dai, P. Zhang, C. Feichtenhofer, and J. Hoffman, “Token merging: Your vit but faster,” arXiv preprint arXiv:2210.09461, 2022

    2022 arXiv

  40. [48]

    Pumer: Pruning and merg- ing tokens for efficient vision language models,

    Q. Cao, B. Paranjape, and H. Hajishirzi, “Pumer: Pruning and merg- ing tokens for efficient vision language models,” arXiv preprint arXiv:2305.17530, 2023. 9

    2023 arXiv

  41. [49]

    Hi-mamba: Hierarchical mamba for efficient image super-resolution,

    J. Qiao, J. Liao, W. Li, Y . Zhang, Y . Guo, Y . Wen, Z. Qiu, J. Xie, J. Hu, and S. Lin, “Hi-mamba: Hierarchical mamba for efficient image super-resolution,” arXiv preprint arXiv:2410.10140 , 2024

    2024 arXiv

  42. [50]

    Efficiently modeling long sequences with structured state spaces,

    A. Gu, K. Goel, and C. R ´e, “Efficiently modeling long sequences with structured state spaces,” arXiv preprint arXiv:2111.00396 , 2021

    2021 arXiv

  43. [51]

    Mask r-cnn,

    K. He, G. Gkioxari, P. Doll ´ar, and R. Girshick, “Mask r-cnn,” in Proceedings of the IEEE international conference on computer vision , 2017, pp. 2961–2969

    2017

  44. [52]

    Imagenet: A large-scale hierarchical image database,

    J. Deng, W. Dong, R. Socher, L.-J. Li, K. Li, and L. Fei-Fei, “Imagenet: A large-scale hierarchical image database,” in 2009 IEEE conference on computer vision and pattern recognition . Ieee, 2009, pp. 248–255

    2009

  45. [53]

    Semantic understanding of scenes through the ade20k dataset,

    B. Zhou, H. Zhao, X. Puig, T. Xiao, S. Fidler, A. Barriuso, and A. Torralba, “Semantic understanding of scenes through the ade20k dataset,” International Journal of Computer Vision , vol. 127, pp. 302– 321, 2019

    2019

  46. [54]

    Microsoft coco: Common objects in context,

    T.-Y . Lin, M. Maire, S. Belongie, J. Hays, P. Perona, D. Ramanan, P. Doll ´ar, and C. L. Zitnick, “Microsoft coco: Common objects in context,” in Computer Vision–ECCV 2014: 13th European Conference, Zurich, Switzerland, September 6-12, 2014, Proceedings, Part V 13 . Springer,...

    2014

  47. [55]

    The kinetics human action video dataset,

    W. Kay, J. Carreira, K. Simonyan, B. Zhang, C. Hillier, S. Vijaya- narasimhan, F. Viola, T. Green, T. Back, P. Natsev et al., “The kinetics human action video dataset,” arXiv preprint arXiv:1705.06950 , 2017

    2017 arXiv

  48. [56]

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

    J. Carreira and A. Zisserman, “Quo vadis, action recognition? a new model and the kinetics dataset,” in proceedings of the IEEE Conference on Computer Vision and Pattern Recognition , 2017, pp. 6299–6308

    2017

  49. [57]

    The” something something

    R. Goyal, S. Ebrahimi Kahou, V . Michalski, J. Materzynska, S. West- phal, H. Kim, V . Haenel, I. Fruend, P. Yianilos, M. Mueller-Freitaget al., “The” something something” video database for learning and evaluat- ing visual common sense,” in Proceedings of the IEEE internation...

    2017

  50. [58]

    Masked-attention mask transformer for universal image segmentation,

    B. Cheng, I. Misra, A. G. Schwing, A. Kirillov, and R. Girdhar, “Masked-attention mask transformer for universal image segmentation,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2022, pp. 1290–1299

    2022

  51. [59]

    Flowmamba: Learning point cloud scene flow with global motion propagation,

    M. Lin, G. Xu, Y . Wang, X. Wang, and X. Yang, “Flowmamba: Learning point cloud scene flow with global motion propagation,” arXiv preprint arXiv:2412.17366, 2024

    2024 arXiv

  52. [60]

    Pv-ssm: Exploring pure visual state space model for high-dimensional medical data analysis,

    C. Wang, X. Liu, C. Li, Y . Liu, and Y . Yuan, “Pv-ssm: Exploring pure visual state space model for high-dimensional medical data analysis,” in 2024 IEEE International Conference on Bioinformatics and Biomedicine (BIBM). IEEE, 2024, pp. 2542–2549

    2024

  53. [61]

    Mambau-lite: A lightweight model based on mamba and integrated channel-spatial attention for skin lesion segmentation,

    T.-N.-Q. Nguyen, Q.-H. Ho, D.-T. Nguyen, H.-M.-Q. Le, V .-T. Pham, and T.-T. Tran, “Mambau-lite: A lightweight model based on mamba and integrated channel-spatial attention for skin lesion segmentation,” arXiv preprint arXiv:2412.01405 , 2024

    2024 arXiv

  54. [62]

    Shuffle mamba: State space models with random shuffle for multi-modal image fusion,

    K. Cao, X. He, T. Hu, C. Xie, J. Zhang, M. Zhou, and D. Hong, “Shuffle mamba: State space models with random shuffle for multi-modal image fusion,” arXiv preprint arXiv:2409.01728 , 2024

    2024

  55. [63]

    Mambasod: Dual mamba-driven cross-modal fusion network for rgb-d salient object detection,

    Y . Zhan, Z. Zeng, H. Liu, X. Tan, and Y . Tian, “Mambasod: Dual mamba-driven cross-modal fusion network for rgb-d salient object detection,” arXiv preprint arXiv:2410.15015 , 2024

    2024 arXiv

  56. [64]

    Nimba: Towards robust and principled processing of point clouds with ssms,

    N. K ¨opr¨uc¨u, D. Okpekpe, and A. Orvieto, “Nimba: Towards robust and principled processing of point clouds with ssms,” arXiv preprint arXiv:2411.00151, 2024

    2024 arXiv

  57. [65]

    Serialized point mamba: A serialized point cloud mamba segmentation model,

    T. Wang, W. Wen, J. Zhai, K. Xu, and H. Luo, “Serialized point mamba: A serialized point cloud mamba segmentation model,” arXiv preprint arXiv:2407.12319, 2024

    2024 arXiv

  58. [66]

    Mamba yolo: Ssms-based yolo for object detection,

    Z. Wang, C. Li, H. Xu, and X. Zhu, “Mamba yolo: Ssms-based yolo for object detection,” arXiv preprint arXiv:2406.05835 , 2024

    2024 arXiv

  59. [67]

    Pillarmamba: A lightweight mamba-based model for 3d object detection,

    Y . Liu, Y . Ge, M. Li, G. Zheng, B. Sun, and F.-Y . Wang, “Pillarmamba: A lightweight mamba-based model for 3d object detection,” in 2024 IEEE 4th International Conference on Digital Twins and Parallel Intelligence (DTPI). IEEE, 2024, pp. 652–655

    2024

  60. [68]

    Mambadetr: Query-based temporal modeling using state space model for multi-view 3d object detection,

    T. Ning, K. Lu, X. Jiang, and J. Xue, “Mambadetr: Query-based temporal modeling using state space model for multi-view 3d object detection,” arXiv preprint arXiv:2411.13628 , 2024

    2024 arXiv

  61. [69]

    Mambatron: Efficient cross-modal point cloud enhancement using aggregate selective state space modeling,

    S. T. Inaganti and G. Petrenko, “Mambatron: Efficient cross-modal point cloud enhancement using aggregate selective state space modeling,” arXiv preprint arXiv:2501.16384 , 2025

    2025 arXiv

  62. [70]

    Ms-temba: Multi- scale temporal mamba for efficient temporal action detection,

    A. Sinha, M. S. Raj, P. Wang, A. Helmy, and S. Das, “Ms-temba: Multi- scale temporal mamba for efficient temporal action detection,” arXiv preprint arXiv:2501.06138, 2025. 10

    2025

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.