Pith. sign in

REVIEW 2 major objections 2 minor 47 references

Cross-Modality Feature Fusion Based on Structured State Space Duality for Multimodal Image Registration Network

T0 review · 2 major / 2 minor · reviewed 2026-06-28 · grok-4.3

Pith's one-line read RegNetMamba-2 integrates Structured State Space Duality for effective cross-modality feature fusion in multi-modal image registration.

arxiv 2606.03341 v1 pith:EL4GM6KK submitted 2026-06-02 cs.CV

classification cs.CV
keywords multi-modalimageregistrationstructuredstatespacedualityfeaturefusioncross-modalityinteractioncoarse-to-finematchingRegNetMamba-2VIS-SARVIS-IR
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 introduces RegNetMamba-2, a network that applies Structured State Space Duality (SSD) to extract and fuse features from different image modalities such as visible and SAR. It uses SSD at multiple scales and introduces CMI and MSF modules to handle cross-modality interactions and multi-scale fusion in a coarse-to-fine framework. This is claimed to yield better performance and efficiency than existing deep learning methods on datasets like OSDataset for VIS-SAR and others for VIS-IR and VIS-NIR. A sympathetic reader would care because multi-modal registration is key for applications like surveillance and remote sensing where modalities provide complementary information but are hard to align.

What carries the argument

The Cross-Modality feature Interaction (CMI) module and Multi-Scale feature Fusion (MSF) module based on Structured State Space Duality (SSD), which enable cross-form feature extraction at each scale and progressive fusion across scales to capture shared structural information.

What would settle it

An experiment showing that RegNetMamba-2 does not outperform state-of-the-art methods in registration accuracy or inference speed on the VIS-SAR OSDataset would falsify the central claim.

Watch

Extended reading notes

Core claim

The central claim is that SSD can be effectively incorporated into the coarse-to-fine matching process for multi-modal feature extraction and fusion through the proposed CMI module for cross-modality interaction and MSF module for progressive upward fusion, leading to improved calculation of matching probability scores and pixel-wise correspondences.

Load-bearing premise

The CMI and MSF modules built on SSD extract and fuse shared structural information more effectively than prior methods.

Editorial extensions

If this is right

  • Multi-scale SSD application strengthens local foreground edge and structural representation.
  • Cross-modality fusion in all scales supports shared feature extraction from input images.
  • Features from 1/8 scale and 1/2 scale enable coarse-to-fine matching process.
  • Pixel-wise correspondences establish the final registration.
  • Superior performance and efficiency on VIS-SAR, VIS-IR, and VIS-NIR datasets compared to SOTA.

Reading between the lines

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

  • Adopting SSD in place of attention mechanisms could reduce computational costs in other multi-modal vision tasks.
  • Extending the CMI and MSF design to additional modalities like medical scans might improve alignment accuracy there.
  • The coarse-to-fine strategy combined with SSD could be tested for real-time applications where efficiency matters.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 2 minor

Summary. The manuscript introduces RegNetMamba-2, a deep network for multi-modal image registration that integrates Structured State Space Duality (SSD) into a coarse-to-fine pipeline. It applies SSD at three scales for feature extraction, proposes a Cross-Modality Interaction (CMI) module for cross-form shared feature extraction, and a Multi-Scale Fusion (MSF) module for progressive upward fusion; matching probability scores are then computed from 1/8-scale CMI and 1/2-scale MSF features to establish pixel-wise correspondences. The central empirical claim is that the method outperforms prior deep-learning registration algorithms in both accuracy and efficiency on VIS-SAR (OSDataset), VIS-IR (LGHD/RoadSence), and VIS-NIR (RGB-NIR) datasets.

Significance. If the reported gains are reproducible and the ablations confirm the contribution of the SSD-based CMI and MSF modules, the work would demonstrate a computationally lighter alternative to Transformer-based global modeling for structural feature alignment across modalities, with potential impact on real-time multi-modal vision tasks.

major comments (2)
  1. [Abstract] Abstract: the headline claim that 'extensive experiments demonstrate' superior performance and efficiency supplies no numerical results (e.g., mean registration error, success rate, AUC), no baseline table, and no ablation controls isolating the three-scale SSD, cross-form CMI, or progressive MSF versus standard concatenation or attention; without these the central empirical assertion cannot be evaluated.
  2. [Method description] Method description (coarse-to-fine paragraph): the statement that 'features in 1/8 scale from CMI and 1/2 scale from MSF are collected to calculate matching probability scores' is load-bearing for the registration pipeline yet provides neither the functional form of the probability computation nor the loss used to train it, preventing verification that the claimed structural fusion actually drives the matching.
minor comments (2)
  1. [Abstract] Dataset nomenclature: 'RoadSence' and 'RGB-NIR sense' appear to be typographical variants of standard names (RoadScene, RGB-NIR scene); consistent spelling would aid reproducibility.
  2. [Abstract] The abstract refers to 'feature scaling function of SSD' without defining or citing the exact operation; a brief equation or reference would clarify how foreground edge emphasis is implemented.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on the abstract and method description. We address each comment below and will revise the manuscript to improve clarity and completeness.

read point-by-point responses
  1. Referee: [Abstract] Abstract: the headline claim that 'extensive experiments demonstrate' superior performance and efficiency supplies no numerical results (e.g., mean registration error, success rate, AUC), no baseline table, and no ablation controls isolating the three-scale SSD, cross-form CMI, or progressive MSF versus standard concatenation or attention; without these the central empirical assertion cannot be evaluated.

    Authors: We agree that the abstract would benefit from quantitative highlights to support the claim. The full manuscript includes detailed results in Tables 1-3 (mean registration errors, success rates, AUC) and ablation studies in Section 4.3 isolating the SSD scales, CMI, and MSF contributions versus baselines. We will revise the abstract to incorporate key numerical improvements (e.g., average error reduction and runtime gains) while keeping it concise. revision: yes

  2. Referee: [Method description] Method description (coarse-to-fine paragraph): the statement that 'features in 1/8 scale from CMI and 1/2 scale from MSF are collected to calculate matching probability scores' is load-bearing for the registration pipeline yet provides neither the functional form of the probability computation nor the loss used to train it, preventing verification that the claimed structural fusion actually drives the matching.

    Authors: The functional form (normalized similarity with softmax) and training loss (contrastive) are detailed in Section 3.4, but we acknowledge the coarse-to-fine paragraph should be self-contained. We will expand that paragraph to explicitly state the probability computation and loss, ensuring the role of the structural fusion is clear. revision: yes

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity detected; empirical model proposal with no derivation chain

full rationale

The paper describes an empirical neural network architecture (RegNetMamba-2) that incorporates SSD-based CMI and MSF modules for multi-modal feature extraction and fusion, followed by coarse-to-fine matching. No equations, first-principles derivations, or predictions are presented that reduce by construction to fitted inputs, self-definitions, or self-citation chains. Claims rest on external experimental evaluation across VIS-SAR, VIS-IR, and VIS-NIR datasets rather than any internal reduction. No load-bearing self-citations or ansatz smuggling appear in the abstract or described structure.

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

Abstract-only review supplies no explicit free parameters, mathematical axioms, or newly postulated entities; the approach implicitly relies on standard deep-learning assumptions about feature extractors and optimization that are not enumerated here.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Cross-Modality Feature Fusion Based on Structured State Space Duality for Multimodal Image Registration Network." pith.science (2026). https://pith.science/paper/EL4GM6KK

@misc{pith2026260603341,
  author       = {Pith},
  title        = {Pith review of: Cross-Modality Feature Fusion Based on Structured State Space Duality for Multimodal Image Registration Network},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/EL4GM6KK}},
  note         = {Machine review of arXiv:2606.03341}
}
read the original abstract

In multi-modal image registration, the primary challenge lies in shared structural information extraction. Compared to Transformers, Structured State Space Duality (SSD) offers greater global structural feature extraction with higher efficiency during training and inference. Inspired by these advantages, we propose a novel algorithm for multi-modal image registration, named RegNetMamba-2. Our algorithm incorporates SSD into coarse-to-fine matching process to extract local and global structural features effectively. Firstly, SSD is applied in three different scales for multi-modal feature extraction in our network. To strengthen local representation, we pay more attention on foreground edge and structural information by feature scaling function of SSD. Secondly, for shared feature extraction of input images and multi-modal feature fusion in all scales, we propose cross-modality feature fusion model based on SSD, consisting of Cross-Modality feature Interaction (CMI) module and Multi-Scale feature Fusion (MSF) module. CMI module is designed for cross-modality feature extraction of each scale by SSD in cross form. MSF module is designed to employ a progressive upward fusion in feature-level to obtain fine features, consisting of multi-modal features in all scales. Following coarse-to-fine, the features in 1/8 scale from CMI and 1/2 scale from MSF are collected to calculate matching probability scores. Then we respectively establish matching process by correspondences of pixel-wise. Extensive experiments demonstrate that comparing with state-of-the-art deep-learning based algorithms, RegNetMamba-2 has achieved good effects in both performance and efficiency for multi-modal image registration on the following datasets: VIS-SAR (OSDataset), VIS-IR (LGHD/RoadSence) and VIS-NIR (RGB-NIR sense).

Figures

Figures reproduced from arXiv: 2606.03341 by the authors.

Figure 1
Figure 1. Overall architecture of our RegNetMamba-2. CMI modules contains Stage1 to Stage3 for feature extraction in scale of 1/2, 1/4 and 1/8. After Stage3 [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Architecture of our local enhanced SSD layer in cross form. [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Architecture of SSD blocks. A SSD block is composed of a SSD and [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: Architecture of Cross-Modality feature Interaction (CMI) module. Stage1 and Stage2 are SSD blocks in cross form for feature extraction in scale of [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Architecture of Multi-Scale feature Fusion (MSF) module. Stage4 and [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: Successful matching rate (SMR) of RIFT, Cnet, ReDFeat, LoFTR, [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 6
Figure 6. Figure 6: illustrates the successful matching rate (SMR) for [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]
Figure 7
Figure 7. Figure 7: Visualization of matching performance. Only the inlier matches after MAGSAC are shown. [PITH_FULL_IMAGE:figures/full_fig_p009_7.png]
Figure 8
Figure 8. Figure 8: Checkerboard images of Pair1 to Pair 6 registration results. [PITH_FULL_IMAGE:figures/full_fig_p010_8.png]
Figure 9
Figure 9. Figure 9: Edge images of Pair1 to Pair 6 registration results. Red and green rectangles circle the areas not aligned or aligned. [PITH_FULL_IMAGE:figures/full_fig_p011_9.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

47 extracted references · 2 canonical work pages

  1. [1]

    Structure-from-Motion Revisited

    Schonberger J L, Frahm J M. Structure-from-Motion Revisited. Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2016: 4104-4113

  2. [2]

    Simultaneous Localization and Mapping: Part I

    Durrant-Whyte H, Bailey T. Simultaneous Localization and Mapping: Part I. IEEE Robotics & Automation Magazine, 2006, 13(2): 99-110

  3. [3]

    COMO: Cross-Mamba Interaction and Offset-Guided Fusion for Multimodal Object Detection

    Liu C, Ma X, Yang X, et al. COMO: Cross-Mamba Interaction and Offset-Guided Fusion for Multimodal Object Detection. Information Fusion, 2025: 103414

  4. [4]

    Semi-Supervised Classification of Dual- Frequency PolSar Image Using Joint Feature Learning and Cross Label- Information Network

    Xin X, Li M, Wu Y , et al. Semi-Supervised Classification of Dual- Frequency PolSar Image Using Joint Feature Learning and Cross Label- Information Network. IEEE Transactions on Geoscience and Remote Sensing, 2022, 60: 1-16

  5. [5]

    Unsupervised Difference-Guided Adaptive Fusion Network for Change Detection in PolSAR Images

    Yang Z, Wu Y , Li M, et al. Unsupervised Difference-Guided Adaptive Fusion Network for Change Detection in PolSAR Images. IEEE Trans- actions on Geoscience and Remote Sensing, 2024, 62: 1-16

  6. [6]

    GLS–MIFT: A modality invariant feature transform with global-to-local searching

    Fan Z, Pi Y , Wang M, et al. GLS–MIFT: A modality invariant feature transform with global-to-local searching. Information Fusion, 2024, 105: 102252

  7. [7]

    POS-GIFT: A geometric and intensity-invariant feature transformation for multimodal images

    Hou Z, Liu Y , Zhang L. POS-GIFT: A geometric and intensity-invariant feature transformation for multimodal images. Information Fusion, 2024, 102: 102027

  8. [8]

    Highly adaptive multi-modal image matching based on tuning-free filtering and enhanced sketch features

    Liao Y , Tao P, Chen Q, et al. Highly adaptive multi-modal image matching based on tuning-free filtering and enhanced sketch features. Information Fusion, 2024, 112: 102599

Show all 47 references
  1. [9]

    RIFT: Multi-Modal Image Matching Based on Radiation-Variation Insensitive Feature Transform

    Li J, Hu Q, Ai M. RIFT: Multi-Modal Image Matching Based on Radiation-Variation Insensitive Feature Transform. IEEE Transactions on Image Processing, 2019, 29: 3296-3310

  2. [10]

    Multi-Modal Remote Sensing Image Matching Considering Co-Occurrence Filter

    Yao Y , Zhang Y , Wan Y , et al. Multi-Modal Remote Sensing Image Matching Considering Co-Occurrence Filter. IEEE Transactions on Image Processing, 2022, 31: 2584-2597

  3. [11]

    Multi-Relation Attention Network for Image Patch Matching

    Quan D, Wang S, Li Y , et al. Multi-Relation Attention Network for Image Patch Matching. IEEE Transactions on Image Processing, 2021, 30: 7127-7142

  4. [12]

    Deep Feature Correlation Learning for Multi-Modal Remote Sensing Image Registration

    Quan D, Wang S, Gu Y , et al. Deep Feature Correlation Learning for Multi-Modal Remote Sensing Image Registration. IEEE Transactions on Geoscience and Remote Sensing, 2022, 60: 1-16

  5. [13]

    Efficient Feature Relation Learning Net- work for Cross-Spectral Image Patch Matching

    Yu C, Zhao J, Liu Y , et al. Efficient Feature Relation Learning Net- work for Cross-Spectral Image Patch Matching. IEEE Transactions on Geoscience and Remote Sensing, 2023, 61: 1-17

  6. [14]

    Learning Rotation-Equivariant Features for Visual Correspondence

    Lee J, Kim B, Kim S, et al. Learning Rotation-Equivariant Features for Visual Correspondence. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2023: 21887-21897

  7. [15]

    SSML-QNet: Scale-Separative Metric Learning Quadruplet Network for Multi-modal Image Patch Matching

    Zhang X, Sun Y , Han Y , et al. SSML-QNet: Scale-Separative Metric Learning Quadruplet Network for Multi-modal Image Patch Matching. Proceedings of the Thirty-Second International Joint Conference on Artificial Intelligence. 2023: 4953-4601

  8. [16]

    SuperPoint: Self-Supervised Interest Point Detection and Description

    DeTone D, Malisiewicz T, Rabinovich A. SuperPoint: Self-Supervised Interest Point Detection and Description. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops. 2018: 224-236

  9. [17]

    D2-Net: A Trainable CNN for Joint Description and Detection of Local Features

    Dusmanu M, Rocco I, Pajdla T, et al. D2-Net: A Trainable CNN for Joint Description and Detection of Local Features. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2019: 8092-8101

  10. [18]

    R2D2: Reliable and Repeatable Detector and Descriptor

    Revaud J, De Souza C, Humenberger M, et al. R2D2: Reliable and Repeatable Detector and Descriptor. Advances in Neural Information Processing Systems, 2019, 32

  11. [19]

    ReDFeat: Recoupling Detection and Description for Multimodal Feature Learning

    Deng Y , Ma J. ReDFeat: Recoupling Detection and Description for Multimodal Feature Learning. IEEE Transactions on Image Processing, 2022, 32: 591-602

  12. [20]

    SuperGlue: Learning Feature Matching with Graph Neural Networks

    Sarlin P E, DeTone D, Malisiewicz T, et al. SuperGlue: Learning Feature Matching with Graph Neural Networks. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2020: 4938- 4947

  13. [21]

    LightGlue: Local Feature Matching at Light Speed

    Lindenberger P, Sarlin P E, Pollefeys M. LightGlue: Local Feature Matching at Light Speed. Proceedings of the IEEE/CVF International Conference on Computer Vision. 2023: 17627-17638

  14. [22]

    OmniGlue: Generalizable Fea- ture Matching with Foundation Model Guidance

    Jiang H, Karpur A, Cao B, et al. OmniGlue: Generalizable Fea- ture Matching with Foundation Model Guidance. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2024: 19865-19875

  15. [23]

    ClusterGNN: Cluster-based Coarse-to- Fine Graph Neural Network for Efficient Feature Matching

    Shi Y , Cai J X, Shavit Y , et al. ClusterGNN: Cluster-based Coarse-to- Fine Graph Neural Network for Efficient Feature Matching. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recog- nition. 2022: 12517-12526

  16. [24]

    Learning to Match Features with Seeded Graph Matching Network

    Chen H, Luo Z, Zhang J, et al. Learning to Match Features with Seeded Graph Matching Network. Proceedings of the IEEE/CVF International Conference on Computer Vision. 2021: 6301-6310

  17. [25]

    FeatureBooster: Boosting Feature Descrip- tors with a Lightweight Neural Network

    Wang X, Liu Z, Hu Y , et al. FeatureBooster: Boosting Feature Descrip- tors with a Lightweight Neural Network. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2023: 7630- 7639

  18. [26]

    LoFTR: Detector-Free Local Feature Matching with Transformers

    Sun J, Shen Z, Wang Y , et al. LoFTR: Detector-Free Local Feature Matching with Transformers. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2021: 8922-8931

  19. [27]

    MatchFormer: Interleaving Attention in Transformers for Feature Matching

    Wang Q, Zhang J, Yang K, et al. MatchFormer: Interleaving Attention in Transformers for Feature Matching. Proceedings of the Asian Con- ference on Computer Vision. 2022: 2746-2762

  20. [28]

    ASpanFormer: Detector-Free Image Matching with Adaptive Span Transformer

    Chen H, Luo Z, Zhou L, et al. ASpanFormer: Detector-Free Image Matching with Adaptive Span Transformer. European Conference on Computer Vision. 2022: 20-36

  21. [29]

    Adaptive Spot-Guided Transformer for Consistent Local Feature Matching

    Yu J, Chang J, He J, et al. Adaptive Spot-Guided Transformer for Consistent Local Feature Matching. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2023: 21898- 21908

  22. [30]

    XoFTR: Cross-modal Feature Matching Transformer

    Tuzcuo ˘glu ¨O, K ¨oksal A, Sofu B, et al. XoFTR: Cross-modal Feature Matching Transformer. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2024: 4275-4286

  23. [31]

    Mamba: Linear-Time Sequence Modeling with Selective State Spaces

    Gu A, Dao T. Mamba: Linear-Time Sequence Modeling with Selective State Spaces. arXiv:2312.00752, 2023. 14

  24. [32]

    VMamba: Visual State Space Model

    Liu Y , Tian Y , Zhao Y , et al. VMamba: Visual State Space Model. Advances in Neural Information Processing Systems, 2024, 37: 103031- 103063

  25. [33]

    LocalMamba: Visual State Space Model with Windowed Selective Scan

    Huang T, Pei X, You S, et al. LocalMamba: Visual State Space Model with Windowed Selective Scan. European Conference on Computer Vision, 2024: 12-22

  26. [34]

    Multi-Scale VMamba: Hierarchy in Hierarchy Visual State Space Model

    Shi Y , Dong M, Xu C. Multi-Scale VMamba: Hierarchy in Hierarchy Visual State Space Model. Advances in Neural Information Processing Systems, 2024, 37: 25687-25708

  27. [35]

    Demystify Mamba in Vision: A Lin- ear Attention Perspective

    Han D, Wang Z, Xia Z, et al. Demystify Mamba in Vision: A Lin- ear Attention Perspective. Advances in Neural Information Processing Systems, 2024, 37: 127181-127203

  28. [36]

    Transformers are SSMs: Generalized Models and Efficient Algorithms Through Structured State Space Duality

    Dao T, Gu A. Transformers are SSMs: Generalized Models and Efficient Algorithms Through Structured State Space Duality. Proceedings of Machine Learning Research. 2024; 235:10041-10071

  29. [37]

    VSSD: Vision Mamba with Non-Causal State Space Duality

    Shi Y , Dong M, Li M, et al. VSSD: Vision Mamba with Non-Causal State Space Duality. arXiv:2407.18559, 2024

  30. [38]

    Machine Learning for High-Speed Corner Detection

    Rosten E, Drummond T. Machine Learning for High-Speed Corner Detection. European Conference on Computer Vision, 2006: 430-443

  31. [39]

    Automatic Registration of Optical and SAR Images Via Improved Phase Congruency Model

    Xiang Y , Tao R, Wang F, et al. Automatic Registration of Optical and SAR Images Via Improved Phase Congruency Model. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2020, 13: 5847-5861

  32. [40]

    LGHD: A Feature Descriptor for Matching across Non-linear Intensity Variations

    Aguilera C A, Sappa A D, Toledo R. LGHD: A Feature Descriptor for Matching across Non-linear Intensity Variations. 2015 IEEE Inter- national Conference on Image Processing (ICIP). IEEE, 2015: 178-181

  33. [41]

    FusionDN: A Unified Densely Connected Network for Image Fusion

    Xu H, Ma J, Le Z, et al. FusionDN: A Unified Densely Connected Network for Image Fusion. Proceedings of the AAAI Conference on Artificial Intelligence. 2020, 34(07): 12484-12491

  34. [42]

    Multi-spectral SIFT for Scene Category Recog- nition

    Brown M, S ¨usstrunk S. Multi-spectral SIFT for Scene Category Recog- nition. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2011: 177-184

  35. [43]

    MAGSAC: Marginalizing Sample Con- sensus

    Barath D, Matas J, Noskova J. MAGSAC: Marginalizing Sample Con- sensus. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2019: 10197-10205

  36. [44]

    Decoupled Weight Decay Regularization

    Loshchilov I, Hutter F. Decoupled Weight Decay Regularization. Inter- national Conference on Learning Representations.2019

  37. [45]

    JamMa: Ultra-lightweight Local Feature Matching with Joint Mamba

    Lu X, Du S. JamMa: Ultra-lightweight Local Feature Matching with Joint Mamba. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2025: 14934-14943

  38. [46]

    Flatten transformer: Vision transformer us- ing focused linear attention

    Han D, Pan X, Han Y , et al. Flatten transformer: Vision transformer us- ing focused linear attention. Proceedings of the IEEE/CVF International Conference on Computer Vision. 2023: 5961-5971

  39. [47]

    LoFLAT: Local Feature Matching using Focused Linear Attention Transformer

    Cao N, He R, Dai Y , et al. LoFLAT: Local Feature Matching using Focused Linear Attention Transformer. 2024 Asia Pacific Signal and Information Processing Association Annual Summit and Conference. IEEE, 2024: 1-6. Zhikang Lireceived the B.S. degree from Xidian University, Xi’a...

Pith tools

Reviewed June 28, 2026 · model on record in the stance chip above.