Pith. sign in

REVIEW 5 major objections 3 minor 71 references

STGBD-Net: Spatio-temporal Gradient Basis Decomposition Network for Infrared Small Target Detection

T0 review · 5 major / 3 minor · reviewed 2026-08-03 · deepseek-v4-flash

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

Pith's one-line read Dynamic projection onto difference bases gives a 0.27M-parameter network state-of-the-art infrared small-target detection.

desk verdict A plausible lightweight IRSTD module with dynamic difference features, but the SOTA claim rests on test-set tuning and an empty property-evaluation section. read the letter →

arxiv 2512.03470 v5 pith:ALI6CMVR submitted 2025-12-03 cs.CV

Chen Hu , Mingyu Zhou , Shuai Yuan , Hongbo Hu , Zhenming Peng , Tian Pu , Xiying Li This is my paper · ORCID
classification cs.CV
keywords infraredsmalltargetdetectionbasisdecompositiondifferenceconvolutionfeaturefusionlightweightU-Netmulti-frametemporaldynamicprojection
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 argues that the standard practice of fusing multi-directional gradient features with fixed weights is the main weakness of difference-based methods in infrared small target detection (IRSTD). The authors propose treating normalized difference features as basis vectors and reconstructing the input feature map by projecting it onto them, with coefficients computed per-sample and per-location. They integrate this Basis Decomposition Module into a lightweight three-stage U-Net (SD2Net) and a temporal variant (STD2Net) for multi-frame video. On public benchmarks the resulting networks reach mIoU of 95.13% (NUDT-SIRST), 69.40% (IRSTD-1K), and 87.68% (NUDT-MIRSDT) while using 0.27M parameters and running at 217 FPS, outperforming much heavier published models. The pith is that a simple input-dependent projection onto difference bases can replace learned fusion and deliver a better accuracy–efficiency trade-off.

What carries the argument

The Basis Decomposition Module (BDM) and its IRSTD specializations: the Spatial Difference Decomposition Module (SD2M) and Temporal Difference Decomposition Module (TD2M). The core identity is O = (T·P^T)·P, where P is the L2-normalized tensor of basis features drawn from eight directional difference kernels at multiple dilations (spatial) or frame differences plus the current frame (temporal). The dot-product S = T·P^T computes per-location projection coefficients; multiplying back is a linear reconstruction. This is a low-rank, input-dependent projection that replaces static coefficient fusions; it is not an attention mechanism because the coefficients come from geometric alignment with fi

What would settle it

Train SD2Net on IRSTD-1K with the eight fixed difference kernels replaced by frozen L2-normalized random kernels of the same size. If mIoU drops more than a few points, the specific kernels carry the performance; if it stays comparable, the projection mechanism itself is what matters. A second check is a test set with targets rotated by 30–90 degrees, which would directly test the claimed rotation invariance since Section IV.B of the paper leaves it unmeasured.

Watch

Extended reading notes

Core claim

The central claim is that feature fusion for IRSTD can be reformulated as basis decomposition: given an input feature T and a set of normalized basis features P derived from difference kernels (or frame differences), the output O = (T·P^T)·P reconstructs the input using only the task-relevant high-frequency basis features, thereby enhancing dim targets while suppressing low-frequency clutter. The coefficients S = T·P^T are not learned but computed dynamically from the input, which the authors argue makes the module robust to intensity rescaling and rotations. By relaxing orthogonality while retaining L2 normalization, the module stays lightweight and interpretable. Integrated into a three-st

Load-bearing premise

The load-bearing premise is that the fixed set of eight L2-normalized directional difference kernels (at dilations 1 and 3) forms an adequate task-relevant basis for representing infrared small targets in every condition; because orthogonality is relaxed, the module is a low-rank linear map, so if these hand-picked bases miss a target's structure the projection cannot recover it.

Editorial extensions

If this is right

  • Single-frame detection accuracy improves over established networks on NUDT-SIRST and IRSTD-1K with far fewer parameters and higher FPS.
  • The same module generalizes to temporal domain: STD2Net outperforms DTUM/RFR-based multi-frame methods on NUDT-MIRSDT.
  • Removing orthogonality but keeping normalization is not only cheaper but also empirically better than forcing orthogonality via SVD in the ablation study.
  • The design principle of a task-oriented basis pool can be transferred to other tasks by swapping the Basis Feature Extraction Module.
  • Dilation ratios [1,3] suit both small and larger targets, though large targets remain a recognized limitation.

Reading between the lines

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

  • Mathematically, relaxation of orthogonality means O = (T·P^T)·P is not an exact projection but a low-rank linear map; a testable consequence is that its power may come from dynamic high-frequency gating rather than the specific choice of the eight kernels.
  • The claimed rotation invariance applies to the projection step only; the full network includes rotation-sensitive downsampling and pooling, so the empty Section IV.B leaves an open question that a rotated test set could settle.
  • Since the basis is hand-picked and fixed, learning a lightweight transform of the difference features could extend the method to large or irregular targets while preserving the dynamic-projection property.
  • The same BDM principle could transfer to other low-contrast detection tasks such as salient object detection or moving-object detection in visible video, where difference features define the object.
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

5 major / 3 minor

Summary. The paper proposes the Basis Decomposition Module (BDM), which computes coefficients S = T·Pᵀ between an input feature T and a set of L2-normalized basis features P, then reconstructs O = (T·Pᵀ)·P. For infrared small target detection, difference features from eight fixed spatial kernels (with multiple dilations) or from frame differences are used as the basis. Two networks are built on a three-stage U-Module backbone: SD2Net for single-frame and STD2Net for multi-frame detection. On NUDT-SIRST, IRSTD-1K, and NUDT-MIRSDT the authors report mIoU of 95.13, 69.40, and 87.68 with 0.27M parameters, and claim state-of-the-art accuracy-efficiency trade-offs, dynamic-weight robustness, and invariance to affine/rotational transformations.

Significance. An accuracy-efficiency trade-off is important for IRSTD, and a 0.27M-parameter model with 1.51 GFLOPs reaching top mIoU would be a useful contribution. The paper evaluates many public benchmarks and baselines and promises code release; the multi-frame extension is also a positive. However, the theoretical framing as basis decomposition is impaired by the non-orthogonality of the chosen basis, the claimed invariances are not tested, and the empirical SOTA claim is weakened by test-set-based hyperparameter selection and the absence of variance estimates. If the experimental protocol is corrected and the theory is reworded, the contribution can still be valuable.

major comments (5)
  1. [§IV.A.1, §IV.C.2, §IV.C.3, Tables II–III and VI] No validation set is created; the 80/20 split in Sec IV.A.1 leaves only a test set. Tables II and III choose dilation [1,3] and channel [8,16,32,64] by maximizing mIoU on that test set, and Table VI then reports the result on the same test set. This is test-set model selection: the +0.86/+0.9 mIoU margins over DNANet can be inflated by selection bias. Please use a validation split or nested cross-validation and report performance of the selected configuration on an untouched test set.
  2. [§IV.A.2] The authors state that each algorithm is run three times and the average is reported, but no standard deviations or per-run values are given. Given the small margins (e.g., 95.13 vs 94.27 on NUDT-SIRST), the claimed SOTA is not statistically substantiated. Report variance and, if feasible, significance tests.
  3. [§III.A.1–2, Eq. (9)] The theory in Eqs. (1)–(3) relies on an orthonormal basis, but the implemented P in Eq. (9) uses non-orthogonal difference features. For non-orthogonal bases, S = T·Pᵀ is not the decomposition coefficient vector; the correct least-squares coefficients require the inverse Gram matrix (PᵀP)⁻¹. Thus O = (T·Pᵀ)·P is a linear filter, not a basis decomposition. The relaxation in Sec III.A.2 is acknowledged, but the paper should either prove which properties survive the relaxation or re-describe the module as a dynamic difference-feature transform.
  4. [§III.A.3, §IV.B, Fig. 3b] Rotation invariance is not established. The eight fixed kernels in Fig. 8 are closed under 45° rotations, not arbitrary rotations, and Fig. 3b does not provide a proof for the general claim. The entire Property Evaluation section IV.B is empty (subsections 1–3 contain no experiments or text). Either add the experiments/analysis or remove/qualify the invariance claims.
  5. [§IV.A.2, Table VI] All models, including baselines, are trained under the authors' experimental configuration (400 epochs, Adam, LR 5e-4, etc.). Published baselines may have been tuned to different schedules or losses; this can unfairly disadvantage them. Provide baseline training details, or cite original published numbers and clearly state any re-implementation differences.
minor comments (3)
  1. [Title/Abstract] The full-text title and abstract refer to 'Difference Decomposition Networks' and SD²Net/STD²Net, while the arXiv metadata says 'STGBD-Net: Spatio-temporal Gradient Basis Decomposition Network'. This mismatch should be reconciled.
  2. [Table VI] The caption of Table VI says 'Ablation studies of SD2Net' but the table is a comparison with state-of-the-art methods; the caption should be corrected.
  3. [Tables III and VII] There are formatting errors in the tables: Table III shows '0.07217.41' for the row [4,8,16,32], and Table VII shows '18. 16' for DATransNet. All numeric entries should be checked for similar garbled values.

Circularity Check

1 steps flagged · score 4.0 of 10

NUDT-SIRST headline mIoU is selected on the test set; core BDM projection is otherwise self-contained.

  1. fitted input called prediction [Sec. IV.A.1 (dataset split), Sec. IV.C.2-IV.C.3 (Tables II-III), Sec. IV.D.2 (Table VI)]
    "To determine the optimal setting, we performed an ablation study on dilation ratios, systematically evaluating the performance of various configurations, as shown in Table II. The experiment results indicated that the [1, 3] configuration significantly outperformed the others. Based on these findings, we adopted [1, 3] as our dilation configuration."

    The NUDT-SIRST test split (20% 'reserved as the test set for the final performance assessment') is used to select the dilation configuration [1,3] and the channel configuration [8,16,32,64] by their mIoU values, and then the same test split produces the headline 95.13 mIoU in Table VI. The configuration is therefore a fitted input: the reported performance is the in-sample best of the tested configurations, not an independent prediction, which inflates the claimed 0.86-point margin over DNANet. This is a fitted-value-called-prediction step rather than a definitional identity, leaving the IRSTD-1K and NUDT-MIRSDT results partially independent.

full rationale

The derivation chain for the module is self-contained: Eq. 9 defines O=(T*P^T)*P as a dynamic projection of FEM features onto L2-normalized BFEM features; no term in this definition is the benchmark metric, and the coefficients are computed per input rather than learned from ground truth. Comparisons are against external published methods on public datasets, and no load-bearing premise is imported from the authors' own prior papers (self-overlapping refs [28], [42], [62] appear only as baselines). The main circularity-adjacent problem is evaluation: Sec. IV.A.1 reserves 20% of NUDT-SIRST as test, and Tables II-III pick the dilation/channel configs by their NUDT-SIRST mIoU, so the same test split is reused for the 95.13 mIoU headline in Table VI. That number is an in-sample best-of-k selection, not an unbiased prediction, which biases the claimed +0.86 margin over DNANet. This is partial: IRSTD-1K and NUDT-MIRSDT results use the same config but were not the selection set, and the architecture itself is not defined by the result. Also flagged as missing support (not circularity): Sec. IV.B, meant to test the claimed affine/rotation invariances, is empty, and no variance is reported despite averaging three runs. Score 4 reflects one headline number that is partly fitted, while the core decomposition formulation remains independent.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central claim rests on standard linear algebra, a domain assumption about difference features, and two ad hoc assumptions: that non-orthogonal bases still behave like a decomposition, and that the fixed kernel set is invariant under rotations. No new physical entities are introduced.

free parameters (3)
  • dilation configuration = [1,3]
    Chosen via ablation on NUDT-SIRST (Table II), not on a held-out validation set; directly affects the receptive field of the basis features.
  • channel configuration = [8,16,32,64]
    Chosen via ablation on NUDT-SIRST (Table III) to balance mIoU and FPS; this is hyperparameter fitting on the evaluation benchmark.
  • temporal window size N = 5
    Set in Implementation Details with no ablation or justification; influences multi-frame STD²Net performance.
assumptions (4)
  • standard math A vector can be recovered as a linear combination of c_ful orthonormal basis vectors (Eqs. 1-3).
    Standard linear algebra; used to motivate the module design.
  • domain assumption Normalized difference features (from fixed kernels/frame differences) capture target-relevant high-frequency structure while low-frequency features are background clutter.
    The whole method depends on this; it is the reason the projection highlights targets and suppresses clutter.
  • ad hoc to paper Relaxing orthogonality while keeping L2-normalized bases still yields a meaningful 'basis decomposition' that enhances targets (Sec III.A.2, Table I).
    No proof is given that non-orthogonal P preserves the decomposition interpretation; the ablation only compares normalization on/off, not the semantic claim.
  • ad hoc to paper The fixed 8-direction kernel set is closed under rotations relevant to the data, giving rotation invariance (Sec III.A.3 and Fig 3b).
    No experiment or proof; arbitrary rotations do not permute the kernel set, so the claimed invariance is not established.

how reviews work

0 comments
Cite this review

Pith. "Pith review of STGBD-Net: Spatio-temporal Gradient Basis Decomposition Network for Infrared Small Target Detection." pith.science (2026). https://pith.science/paper/ALI6CMVR

@misc{pith2026251203470,
  author       = {Pith},
  title        = {Pith review of: STGBD-Net: Spatio-temporal Gradient Basis Decomposition Network for Infrared Small Target Detection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ALI6CMVR}},
  note         = {Machine review of arXiv:2512.03470}
}
read the original abstract

A key challenge in infrared small target detection (IRSTD) is that weak target signal responses are easily obscured by strong background clutter, frequently resulting in missed detections. While traditional gradient-based methods attempt to capture fine details, their robustness is limited by the static fusion of multi-directional gradient features. In this paper, we rethink feature fusion from the perspective of Basis Decomposition Theory and propose a novel framework that reformulates the process into an explicit and adaptive decomposition-and-reconstruction paradigm. Specifically, we introduce the Basis Decomposition Module (BDM) and its specialized variant, the Gradient Decomposition Module (GDM) for IRSTD. GDMs treat the normalized gradient features as basis vectors to reconstruct a new feature, thereby maintaining detailed structures and highlighting infrared small targets. By integrating GDMs into a lightweight three-stage U-Net, we develop two unified architectures: the Spatial Gradient Basis Decomposition Network for single-frame detection and the Spatio-temporal Gradient Basis Decomposition Network for multi-frame scenarios. Extensive experiments demonstrate that our networks achieve state-of-the-art (SOTA) performance across multiple benchmarks, offering a superior balance between detection accuracy and computational efficiency. Our codes will be made public at: https://github.com/greekinRoma/IRSTD_HC_Platform.

Figures

Figures reproduced from arXiv: 2512.03470 by the authors.

Figure 1
Figure 1. The standard CDC and frame-difference operations, [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. (a) The procedure of basis decomposition, where [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Robustness of our methods. (a) Invariant to global affine [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (8 more)
Figure 5
Figure 5. Figure 5: Detailed structure of the SD2M and SD3M used in Figures 7 and 9 respectively, are constructed based on the BDM. SD2M is utilized for the feature extraction without C C C C C C C C C U-Module Fine Stage C 2C 4C 8C 16C 8C 4C 2C C Main U-Module Refine Stage C C C C C C C …
Figure 6
Figure 6. Figure 6: Our backbone adopts a three-stage U-Module struc￾ture, where each U-Module is an encoder–decoder archi￾tecture. The first U-Module extracts coarse information, the second U-Module performs as the main feature-extraction stage to obtain the refined information, and the …
Figure 7
Figure 7. Figure 7: The overall architecture of SD2Net, which incorporates the SD2M and SD3M modules. P2 d1 P3 d1 P4 d1 P5 d1 P6 d1 P1 d1 P1 d1 Basis Features K1 0 0 0 0 -1 0 1 0 0 K2 0 0 0 1 -1 0 0 0 0 K3 1 0 0 0 -1 0 0 0 0 K4 0 0 0 0 -1 0 0 1 0 K5 0 1 0 0 -1 0 0 0 0 K6 0 0 0 0 -1 0 0 0 …
Figure 8
Figure 8. Figure 8: Architecture of the DFEM, which employs eight [PITH_FULL_IMAGE:figures/full_fig_p006_8.png]
Figure 10
Figure 10. Figure 10: The structure of STD2Net comprises two main components: the TFEM is employed to extract difference basis features in the temporal domain. At the same time, the FEM is utilized to fuse the reference frames with the key frame to obtain the features with temporal informa…
Figure 11
Figure 11. Figure 11: Visual results from varied data-driven methods. Green, red, and yellow boxes indicate detected targets, missed targets, [PITH_FULL_IMAGE:figures/full_fig_p008_11.png]
Figure 12
Figure 12. Figure 12: Comparative Analysis of mIoU, Parametric Efficiency, [PITH_FULL_IMAGE:figures/full_fig_p010_12.png]
Figure 13
Figure 13. Figure 13: ROC and zoomed-in ROC curves on NUDT-SIRST [PITH_FULL_IMAGE:figures/full_fig_p011_13.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

71 extracted references · 1 linked inside Pith

  1. [1]

    IRPruneDet: Efficient infrared small target detection via wavelet structure-regularized soft channel pruning,

    M. Zhang, H. Yang, J. Guo, Y . Li, X. Gao, and J. Zhang, “IRPruneDet: Efficient infrared small target detection via wavelet structure-regularized soft channel pruning,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 38, no. 7, Mar. 2024, pp. 7224–7232

    2024

  2. [2]

    Toward robust infrared small target detection via frequency and spatial feature fusion,

    Y . Zhu, Y . Ma, F. Fan, J. Huang, Y . Yao, X. Zhou, and R. Huang, “Toward robust infrared small target detection via frequency and spatial feature fusion,”IEEE Transactions on Geoscience and Remote Sensing, vol. 63, pp. 1–15, 2025. JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015 12

    2025

  3. [3]

    Language-Driven Motion Prior Knowledge Learning for Moving Infrared Small Target Detection,

    S. Chen, L. Ji, S. Peng, S. Zhu, M. Ye, and Y . Sang, “Language-Driven Motion Prior Knowledge Learning for Moving Infrared Small Target Detection,”IEEE Transactions on Geoscience and Remote Sensing, vol. 63, pp. 1–14, 2025

    2025

  4. [4]

    MICPL: Motion-Inspired Cross- Pattern Learning for Small-Object Detection in Satellite Videos,

    S. Chen, L. Ji, S. Zhu, and M. Ye, “MICPL: Motion-Inspired Cross- Pattern Learning for Small-Object Detection in Satellite Videos,”IEEE Transactions on Neural Networks and Learning Systems, vol. 36, no. 4, pp. 6437–6450, 2025

    2025

  5. [5]

    ASCNet: Asymmetric sampling correction network for infrared image destriping,

    S. Yuan, H. Qin, X. Yan, S. Yang, S. Yang, N. Akhtar, and H. Zhou, “ASCNet: Asymmetric sampling correction network for infrared image destriping,”IEEE Transactions on Geoscience and Remote Sensing, vol. 63, pp. 1–15, 2025

    2025

  6. [6]

    Toward accurate infrared small target detection via edge-aware gated transformer,

    Y . Zhu, Y . Ma, F. Fan, J. Huang, K. Wu, and G. Wang, “Toward accurate infrared small target detection via edge-aware gated transformer,”IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 17, pp. 8779–8793, 2024

    2024

  7. [7]

    Pick of the bunch: Detecting infrared small targets beyond hit-miss trade-offs via selective rank-aware attention,

    Y . Dai, P. Pan, Y . Qian, Y . Li, X. Li, J. Yang, and H. Wang, “Pick of the bunch: Detecting infrared small targets beyond hit-miss trade-offs via selective rank-aware attention,”IEEE Transactions on Geoscience and Remote Sensing, vol. 62, pp. 1–15, 2024

    2024

  8. [8]

    Daaf: Degradation-aware adaptive fusion framework for robust infrared and visible images fusion,

    T. Zhang, J. Zhao, Y . Zhu, G. Cui, Y . Jing, and Y . Lyu, “Daaf: Degradation-aware adaptive fusion framework for robust infrared and visible images fusion,”arXiv preprint arXiv:2504.10871, 2025

    arXiv 2025

Show all 71 references
  1. [9]

    Coupled degradation mod- eling and fusion: A vlm-guided degradation-coupled network for degradation-aware infrared and visible image fusion,

    T. Zhang, J. Zhao, Y . Zhu, and G. Cui, “Coupled degradation mod- eling and fusion: A vlm-guided degradation-coupled network for degradation-aware infrared and visible image fusion,”arXiv preprint arXiv:2510.11456, 2025

    2025

  2. [10]

    Beyond full labels: Energy-double-guided single-point prompt for infrared small target label generation,

    S. Yuan, H. Qin, R. Kou, X. Yan, Z. Li, C. Peng, D. Wu, and H. Zhou, “Beyond full labels: Energy-double-guided single-point prompt for infrared small target label generation,”IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 18, pp. 8125–8137, 2025

    2025

  3. [11]

    Infrared small target detection via nonconvex tensor fibered rank approximation,

    X. Kong, C. Yang, S. Cao, C. Li, and Z. Peng, “Infrared small target detection via nonconvex tensor fibered rank approximation,”IEEE Transactions on Geoscience and Remote Sensing, vol. 60, pp. 1–21, 2022

    2022

  4. [12]

    Infrared dim target detection via mode-k1k2 extension tensor tubal rank under complex ocean environment,

    Z. Cao, X. Kong, Q. Zhu, S. Cao, and Z. Peng, “Infrared dim target detection via mode-k1k2 extension tensor tubal rank under complex ocean environment,”ISPRS Journal of Photogrammetry and Remote Sensing, vol. 181, pp. 167–190, 2021

    2021

  5. [13]

    Reweighted infrared patch-tensor model with both nonlocal and local priors for single-frame small target detection,

    Y . Dai and Y . Wu, “Reweighted infrared patch-tensor model with both nonlocal and local priors for single-frame small target detection,”IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 10, no. 8, pp. 3752–3767, 2017

    2017

  6. [14]

    Miss detection vs. false alarm: Adversarial learning for small object segmentation in infrared images,

    H. Wang, L. Zhou, and L. Wang, “Miss detection vs. false alarm: Adversarial learning for small object segmentation in infrared images,” inIEEE International Conference on Computer Vision, 2019, pp. 8509– 8518

    2019

  7. [15]

    IRMamba: Pixel difference mamba with layer restoration for infrared small target detection,

    M. Zhang, X. Li, F. Gao, and J. Guo, “IRMamba: Pixel difference mamba with layer restoration for infrared small target detection,” in Proceedings of the AAAI Conference on Artificial Intelligence, vol. 39, no. 9, Apr. 2025, pp. 10 003–10 011

    2025

  8. [16]

    HSTNet: A hybrid spatial-channel sparse transformer network for infrared small target detection,

    K. Li, Y . Wang, F. Han, H. Wang, Z. Xiong, and Y . Tian, “HSTNet: A hybrid spatial-channel sparse transformer network for infrared small target detection,”IEEE Transactions on Geoscience and Remote Sensing, 2025

    2025

  9. [17]

    MMIDNet: A multi-level mutual information disentanglement network for cross- domain infrared small target detection,

    K. Li, F. Han, Y . Wang, K. Huang, B. Yang, and Y . Tian, “MMIDNet: A multi-level mutual information disentanglement network for cross- domain infrared small target detection,”IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2025

    2025

  10. [18]

    Shifting neighbors within temporal contexts for slow-moving infrared small target detec- tion,

    Y . Zhu, Y . Ma, F. Fan, J. Huang, and G. Wang, “Shifting neighbors within temporal contexts for slow-moving infrared small target detec- tion,”IEEE Signal Processing Letters, 2025

    2025

  11. [19]

    Object guided external memory network for video object detection,

    H. Deng, Y . Hua, T. Song, Z. Zhang, Z. Xue, R. Ma, N. Robertson, and H. Guan, “Object guided external memory network for video object detection,” inProceedings of IEEE International Conference on Computer Vision, 2019, pp. 6677–6686

    2019

  12. [20]

    Deep feature flow for video recognition,

    X. Zhu, Y . Xiong, J. Dai, L. Yuan, and Y . Wei, “Deep feature flow for video recognition,” inIEEE Conference on Computer Vision and Pattern Recognition, 2017, pp. 2349–2358

    2017

  13. [21]

    Sequence level semantics aggregation for video object detection,

    H. Wu, Y . Chen, N. Wang, and Z. Zhang, “Sequence level semantics aggregation for video object detection,” inProceedings of the IEEE International Conference on Computer Vision, 2019, pp. 9217–9225

    2019

  14. [22]

    TransVOD: End-to-end video object detection with spatial-temporal transformers,

    Q. Zhou, X. Li, L. He, Y . Yang, G. Cheng, Y . Tong, L. Ma, and D. Tao, “TransVOD: End-to-end video object detection with spatial-temporal transformers,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 45, no. 6, pp. 7853–7869, 2022

    2022

  15. [23]

    Ptseformer: Progressive temporal-spatial enhanced transformer towards video object detection,

    H. Wang, J. Tang, X. Liu, S. Guan, R. Xie, and L. Song, “Ptseformer: Progressive temporal-spatial enhanced transformer towards video object detection,” inProceedings of the European Conference on Computer Vision. Springer, 2022, pp. 732–747

    2022

  16. [24]

    Triple-domain feature learning with frequency-aware memory enhancement for moving in- frared small target detection,

    W. Duan, L. Ji, S. Chen, S. Zhu, and M. Ye, “Triple-domain feature learning with frequency-aware memory enhancement for moving in- frared small target detection,”IEEE Transactions on Geoscience and Remote Sensing, vol. 62, pp. 1–14, 2024

    2024

  17. [25]

    FSATFusion: Frequency- Spatial attention transformer for infrared and visible image fusion,

    T. Zhang, J. Zhao, Y . Zhu, G. Cui, and Y . Lyu, “FSATFusion: Frequency- Spatial attention transformer for infrared and visible image fusion,” Computer Vision and Image Understanding, 2026

    2026

  18. [26]

    WIFE-Fusion: Wavelet-aware intra-inter frequency enhancement for multi-model image fusion,

    T. Zhang, J. Zhao, Y . Zhu, and G. Cui, “WIFE-Fusion: Wavelet-aware intra-inter frequency enhancement for multi-model image fusion,”arXiv preprint arXiv:2506.03555, 2025

    2025 arXiv

  19. [27]

    Direction- coded temporal u-shape module for multiframe infrared small target de- tection,

    R. Li, W. An, C. Xiao, B. Li, Y . Wang, M. Li, and Y . Guo, “Direction- coded temporal u-shape module for multiframe infrared small target de- tection,”IEEE Transactions on Neural Networks and Learning Systems, vol. 36, no. 1, pp. 555–568, 2025

    2025

  20. [28]

    Saliency at the helm: steering infrared small target detection with learnable kernels,

    F. Wu, A. Liu, T. Zhang, L. Zhang, J. Luo, and Z. Peng, “Saliency at the helm: steering infrared small target detection with learnable kernels,” IEEE Transactions on Geoscience and Remote Sensing, vol. 63, pp. 1– 14, 2025

    2025

  21. [29]

    STDMANet: Spatio-temporal differential multiscale attention network for small mov- ing infrared target detection,

    P. Yan, R. Hou, X. Duan, C. Yue, X. Wang, and X. Cao, “STDMANet: Spatio-temporal differential multiscale attention network for small mov- ing infrared target detection,”IEEE Transactions on Geoscience and Remote Sensing, vol. 61, pp. 1–16, 2023

    2023

  22. [30]

    Edge and corner awareness-based spatial–temporal tensor model for infrared small-target detection,

    P. Zhang, L. Zhang, X. Wang, F. Shen, T. Pu, and C. Fei, “Edge and corner awareness-based spatial–temporal tensor model for infrared small-target detection,”IEEE Transactions on Geoscience and Remote Sensing, vol. 59, no. 12, pp. 10 708–10 724, 2021

    2021

  23. [31]

    Infrared small target detection based on facet kernel and random walker,

    Y . Qin, L. Bruzzone, C. Gao, and B. Li, “Infrared small target detection based on facet kernel and random walker,”IEEE Transactions on Geoscience and Remote Sensing, vol. 57, no. 9, pp. 7104–7118, 2019

    2019

  24. [32]

    Multiscale patch-based contrast measure for small infrared target detection,

    Y . Wei, X. You, and H. Li, “Multiscale patch-based contrast measure for small infrared target detection,”Pattern Recognition, vol. 58, pp. 216–226, 2016

    2016

  25. [33]

    Infrared small target detection based on the weighted strengthened local contrast measure,

    J. Han, S. Moradi, I. Faramarzi, H. Zhang, Q. Zhao, X. Zhang, and N. Li, “Infrared small target detection based on the weighted strengthened local contrast measure,”IEEE Geoscience and Remote Sensing Letters, vol. 18, no. 9, pp. 1670–1674, 2021

    2021

  26. [34]

    Infrared patch-image model for small target detection in a single image,

    C. Gao, D. Meng, Y . Yang, Y . Wang, X. Zhou, and A. G. Hauptmann, “Infrared patch-image model for small target detection in a single image,”IEEE transactions on Image Processing, vol. 22, no. 12, pp. 4996–5009, 2013

    2013

  27. [35]

    Infrared small target detection based on partial sum of the tensor nuclear norm,

    L. Zhang and Z. Peng, “Infrared small target detection based on partial sum of the tensor nuclear norm,”Remote Sensing, vol. 11, no. 4, 2019

    2019

  28. [36]

    Asymmetric contextual modulation for infrared small target detection,

    Y . Dai, Y . Wu, F. Zhou, and K. Barnard, “Asymmetric contextual modulation for infrared small target detection,” inProceedings of the IEEE Winter Conference on Applications of Computer Vision, January 2021, pp. 950–959

    2021

  29. [37]

    Attentional local contrast networks for infrared small target detection,

    Y . Dai, Y . Wu, and K. Barnard, “Attentional local contrast networks for infrared small target detection,”IEEE transactions on geoscience and remote sensing, vol. 59, no. 11, pp. 9813–9824, 2021

    2021

  30. [38]

    Dense nested attention network for infrared small target detection,

    B. Li, C. Xiao, L. Wang, Y . Wang, Z. Lin, M. Li, W. An, and Y . Guo, “Dense nested attention network for infrared small target detection,” IEEE Transactions on Image Processing, vol. 32, pp. 1745–1758, 2023

    2023

  31. [39]

    UIU-Net: U-net in u-net for infrared small object detection,

    X. Wu, D. Hong, and J. Chanussot, “UIU-Net: U-net in u-net for infrared small object detection,”IEEE Transactions on Image Processing, vol. 32, pp. 364–376, 2023

    2023

  32. [40]

    Receptive-field and direction induced attention network for infrared dim small target detection with a large-scale dataset irdst,

    H. Sun, J. Bai, F. Yang, and X. Bai, “Receptive-field and direction induced attention network for infrared dim small target detection with a large-scale dataset irdst,”IEEE Transactions on Geoscience and Remote Sensing, vol. 61, pp. 1–13, 2023

    2023

  33. [41]

    ISTDU-Net: Infrared small-target detection u-net,

    Q. Hou, L. Zhang, F. Tan, Y . Xi, H. Zheng, and N. Li, “ISTDU-Net: Infrared small-target detection u-net,”IEEE Geoscience and Remote Sensing Letters, vol. 19, pp. 1–5, 2022

    2022

  34. [42]

    Attention-guided pyramid context networks for detecting infrared small target under complex background,

    T. Zhang, L. Li, S. Cao, T. Pu, and Z. Peng, “Attention-guided pyramid context networks for detecting infrared small target under complex background,”IEEE Transactions on Aerospace and Electronic Systems, vol. 59, no. 4, pp. 4250–4261, 2023

    2023

  35. [43]

    SCTransNet: Spatial- channel cross transformer network for infrared small target detection,

    S. Yuan, H. Qin, X. Yan, N. Akhtar, and A. Mian, “SCTransNet: Spatial- channel cross transformer network for infrared small target detection,” IEEE Transactions on Geoscience and Remote Sensing, vol. 62, pp. 1– 15, 2024

    2024

  36. [44]

    A spatial-temporal feature-based detection framework for infrared dim JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015 13 small target,

    J. Du, H. Lu, L. Zhang, M. Hu, S. Chen, Y . Deng, X. Shen, and Y . Zhang, “A spatial-temporal feature-based detection framework for infrared dim JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015 13 small target,”IEEE Transactions on Geoscience and Remote Sensing, vol. ...

    2015

  37. [45]

    Sstnet: Sliced spatio- temporal network with cross-slice convlstm for moving infrared dim- small target detection,

    S. Chen, L. Ji, J. Zhu, M. Ye, and X. Yao, “Sstnet: Sliced spatio- temporal network with cross-slice convlstm for moving infrared dim- small target detection,”IEEE Transactions on Geoscience and Remote Sensing, vol. 62, pp. 1–12, 2024

    2024

  38. [46]

    Toward dense moving infrared small target detection: New datasets and baseline,

    S. Chen, L. Ji, S. Zhu, M. Ye, H. Ren, and Y . Sang, “Toward dense moving infrared small target detection: New datasets and baseline,”IEEE Transactions on Geoscience and Remote Sensing, vol. 62, pp. 1–13, 2024

    2024

  39. [47]

    Lmaformer: Local motion aware transformer for small moving infrared target detection,

    Y . Huang, X. Zhi, J. Hu, L. Yu, Q. Han, W. Chen, and W. Zhang, “Lmaformer: Local motion aware transformer for small moving infrared target detection,”IEEE Transactions on Geoscience and Remote Sensing, vol. 62, pp. 1–17, 2024

    2024

  40. [48]

    Infrared small target detection in satellite videos: A new dataset and a novel recurrent feature refinement framework,

    X. Ying, L. Liu, Z. Lin, Y . Shi, Y . Wang, R. Li, X. Cao, B. Li, S. Zhou, and W. An, “Infrared small target detection in satellite videos: A new dataset and a novel recurrent feature refinement framework,”IEEE Transactions on Geoscience and Remote Sensing, vol. 63, no. 500281...

    2025

  41. [49]

    Rapid salient object detection with difference convo- lutional neural networks,

    Z. Su, L. Liu, M. M ¨uller, J. Zhang, D. Wofk, M.-M. Cheng, and M. Pietik ¨ainen, “Rapid salient object detection with difference convo- lutional neural networks,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 47, no. 10, pp. 9061–9077, 2025

    2025

  42. [50]

    Face description with local binary patterns: Application to face recognition,

    T. Ahonen, A. Hadid, and M. Pietikainen, “Face description with local binary patterns: Application to face recognition,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 28, no. 12, pp. 2037– 2041, 2006

  43. [51]

    Unsupervised texture segmentation using gabor filters,

    A. K. Jain and F. Farrokhnia, “Unsupervised texture segmentation using gabor filters,”Pattern Recognition, vol. 24, no. 12, pp. 1167–1186, 1991

    1991

  44. [52]

    Local binary convolu- tional neural networks,

    F. Juefei-Xu, V . N. Boddeti, and M. Savvides, “Local binary convolu- tional neural networks,” inIEEE Conference on Computer Vision and Pattern Recognition, 2017, pp. 4284–4293

    2017

  45. [53]

    Gabor convolutional networks,

    S. Luan, C. Chen, B. Zhang, J. Han, and J. Liu, “Gabor convolutional networks,”IEEE Transactions on Image Processing, vol. 27, no. 9, pp. 4357–4366, 2018

    2018

  46. [54]

    Searching central difference convolutional networks for face anti-spoofing,

    Z. Yu, C. Zhao, Z. Wang, Y . Qin, Z. Su, X. Li, F. Zhou, and G. Zhao, “Searching central difference convolutional networks for face anti-spoofing,” inIEEE Conference on Computer Vision and Pattern Recognition, 2020, pp. 5294–5304

    2020

  47. [55]

    DEA-Net: Single image dehazing based on detail-enhanced convolution and content-guided attention,

    Z. Chen, Z. He, and Z.-M. Lu, “DEA-Net: Single image dehazing based on detail-enhanced convolution and content-guided attention,”IEEE Transactions on Image Processing, vol. 33, pp. 1002–1015, 2024

    2024

  48. [56]

    Deformable kernel networks for joint image filtering,

    B. Kim, J. Ponce, and B. Ham, “Deformable kernel networks for joint image filtering,”International Journal of Computer Vision, vol. 129, no. 2, p. 579–600, Feb. 2021

    2021

  49. [57]

    Dynamic gesture recognition using 3d central difference separable residual lstm coordi- nate attention networks,

    L. Geng, J. Chen, Y . Tie, L. Qi, and C. Liang, “Dynamic gesture recognition using 3d central difference separable residual lstm coordi- nate attention networks,”Journal of Visual Communication and Image Representation, vol. 107, p. 104364, 2025

    2025

  50. [58]

    Isnet: Shape matters for infrared small target detection,

    M. Zhang, R. Zhang, Y . Yang, H. Bai, J. Zhang, and J. Guo, “Isnet: Shape matters for infrared small target detection,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2022, pp. 867–876

    2022

  51. [59]

    Unitbox: An ad- vanced object detection network,

    J. Yu, Y . Jiang, Z. Wang, Z. Cao, and T. Huang, “Unitbox: An ad- vanced object detection network,” inProceedings of ACM international conference on Multimedia, 2016, pp. 516–520

    2016

  52. [60]

    Weighted res-unet for high- quality retina vessel segmentation,

    X. Xiao, S. Lian, Z. Luo, and S. Li, “Weighted res-unet for high- quality retina vessel segmentation,” inProceedings of the International Conference on Information Technology in Medicine and Education, 2018, pp. 327–331

    2018

  53. [61]

    Infrared small target detection with scale and location sensitivity,

    Q. Liu, R. Liu, B. Zheng, H. Wang, and Y . FU, “Infrared small target detection with scale and location sensitivity,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2024, pp. 17 490–17 499

    2024

  54. [62]

    Datransnet: Dynamic attention transformer network for infrared small target detection,

    C. Hu, Y . Huang, K. Li, L. Zhang, C. Long, Y . Zhu, T. Pu, and Z. Peng, “Datransnet: Dynamic attention transformer network for infrared small target detection,”IEEE Geoscience and Remote Sensing Letters, vol. 22, pp. 1–5, 2025

    2025

  55. [63]

    HDNet: A hybrid domain network with multiscale high-frequency information enhancement for infrared small-target detection,

    M. Xu, C. Yu, Z. Li, H. Tang, Y . Hu, and L. Nie, “HDNet: A hybrid domain network with multiscale high-frequency information enhancement for infrared small-target detection,”IEEE Transactions on Geoscience and Remote Sensing, vol. 63, pp. 1–15, 2025

    2025

  56. [64]

    RPCANet: Deep unfolding rpca based infrared small target detection,

    F. Wu, T. Zhang, L. Li, Y . Huang, and Z. Peng, “RPCANet: Deep unfolding rpca based infrared small target detection,” inProceedings of the IEEE Winter Conference on Applications of Computer Vision, 2024, pp. 4809–4818

    2024

  57. [65]

    RPCANet++: Deep interpretable robust pca for sparse object segmen- tation,

    F. Wu, Y . Dai, T. Zhang, Y . Ding, J. Yang, M.-M. Cheng, and Z. Peng, “RPCANet++: Deep interpretable robust pca for sparse object segmen- tation,” 2025

    2025

  58. [66]

    Lightweight deep unfolding networks with enhanced robustness for infrared small target detection,

    J. Liu, Y . Han, X. Xiu, J. Zhang, and W. Liu, “Lightweight deep unfolding networks with enhanced robustness for infrared small target detection,” 2025

    2025

  59. [67]

    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,” inAdvances in Neural Information Processing Systems, A. Globerson, L. Mackey, D. Belgrave, A. Fan, U. Paquet, J. Tomczak, and C. Zhang, Eds., vol. 37. Curran ...

    2024

  60. [68]

    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,” 2024

    2024

  61. [69]

    MiM-ISTD: Mamba-in-Mamba for efficient infrared small- target detection,

    T. Chen, Z. Ye, Z. Tan, T. Gong, Y . Wu, Q. Chu, B. Liu, N. Yu, and J. Ye, “MiM-ISTD: Mamba-in-Mamba for efficient infrared small- target detection,”IEEE Transactions on Geoscience and Remote Sensing, vol. 62, pp. 1–13, 2024

    2024

  62. [70]

    IRSAM: Advancing segment anything model for infrared small target detection,

    M. Zhang, Y . Wang, J. Guo, Y . Li, X. Gao, and J. Zhang, “IRSAM: Advancing segment anything model for infrared small target detection,” inEuropean Conference on Computer Vision. Berlin, Heidelberg: Springer-Verlag, 2024, p. 233–249. Chen Hureceived the B.S. and M.E. degree fr...

    2024

  63. [2022]

    degree with the School of Intelligent Systems Engineering, Sun Yat-sen University

    Now, He is pursuing the Ph.D. degree with the School of Intelligent Systems Engineering, Sun Yat-sen University. His research interests include image processing, computer vision, object detection, and recognition. Yuming Zhouis currently pursuing the B.E.degree in Network Engi...

    2019

Pith tools

Reviewed August 3, 2026 · model on record in the stance chip above.