Pith. sign in

REVIEW 5 major objections 6 minor 50 references

Underwater object detection in sonar imagery with detection transformer and Zero-shot neural architecture search

T0 review · 5 major / 6 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read NAS-DETR claims that a training-free, entropy-based search — not hand-tuned design — picks the backbone that makes its sonar detector the best on two underwater benchmarks while staying real-time.

desk verdict A sensible engineering attempt undermined by contradictory tables and missing error bars—the claimed SOTA sonar detection results are not currently supported. read the letter →

arxiv 2505.06694 v1 pith:T4VJPF6I submitted 2025-05-10 cs.CV cs.AI

XiaoTong Gu , Shengyu Tang , Yiming Cao , Changdong Yu This is my paper · ORCID
classification cs.CVcs.AI
keywords underwaterobjectdetectionsonarimagerytransformer(DETR)zero-shotneuralarchitecturesearchmaximumentropydifferentialevolutionaryalgorithmCNN-transformerhybridbackbone
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

NAS-DETR sets out to show that the hardest part of building a sonar detector — choosing a backbone for images that are low-resolution, noisy, and feature-sparse — can be automated by a search that never trains a single candidate network. The search maximizes a differential-entropy proxy for the information capacity of a CNN-transformer backbone, using an evolutionary algorithm under a FLOPs budget, and the winning architecture is trained exactly once inside a full detector. On the URPC2021 and URPC2022 underwater benchmarks this detector reaches 0.538 and 0.492 mean average precision, edging out the RT-DETR baseline by 2.5% and 3.4% at 71.2–73.8 frames per second. If the training-free proxy really ranks architectures the way trained accuracy does, then architecture search for detection in degraded imagery becomes nearly free: the only expensive step left is the single training of the chosen network.

What carries the argument

The load-bearing object is the zero-shot fitness function $Z(G)$: a weighted average of differential-entropy estimates across the six backbone stages, each stage scored as $\log \operatorname{var}(h_D) + \log C_{\mathrm{in}}$. The variances are propagated analytically — for CNN layers, output variance is the product of kernel-area and channel factors across layers, and for transformer blocks the derivation treats the softmax attention map as a near-linear scaling matrix, yielding $\sigma^2(y) \approx d_{\mathrm{model}}^2\, d_{\mathrm{feedforward}}\, S^2\, \sigma^2(x)$ — so the differential-entropy upper bound $\frac{1}{2}\ln(2\pi e\sigma^2)$ can be maximized without any training. A numerical scale-normalization step $F' = F/\sigma(F)$ keeps the variance finite in deep stacks, with a correction factor relating scaled and unscaled variance. An evolutionary algorithm then mutates kernel width, depth, channels, bottleneck ratio, hidden dimension, and feedforward dimension over 20,000 rounds, keeping the top-ranked individuals under a FLOPs ceiling, and the single highest-scoring backbone is trained once, inside the full detector.

What would settle it

Train ten to twenty backbone architectures sampled across the full range of fitness scores $Z(G)$ inside the otherwise identical detector, and compute the rank correlation between $Z(G)$ and the resulting mean average precision; if the correlation is weak or negative, the entropy proxy is miscalibrated and the reported gains come from the detector components or the FLOPs prior rather than from the search. A second, targeted check would measure the actual variance of the transformer block output at initialization and compare it with the paper's formula $\sigma^2(y) \approx d_{\mathrm{model}}^2\, d_{\mathrm{feedforward}}\, S^2\, \sigma^2(x)$; a large mismatch would confirm that treating softmax attention as a linear scaling matrix breaks the derivation.

Watch

Extended reading notes

Core claim

The paper's central claim is that a zero-shot neural architecture search grounded in the maximum-entropy principle — maximizing the differential entropy of output feature maps, estimated by propagating the variance of Gaussian-initialized activations through CNN and transformer blocks — finds a CNN-transformer hybrid backbone better matched to sonar imagery than backbones designed for optical images. Wrapped in a Deformable-DETR-style decoder with a feature pyramid, query selection, a denoising training task, and a hybrid Varifocal-plus-GIoU loss, the NAS-found backbone achieves what the authors report as state-of-the-art results on URPC2021 (0.538 mmAP) and URPC2022 (0.492 mmAP), surpassing RT-DETR by 2.5% and 3.4% while running at 71.2–73.8 FPS, or 288–294.9 FPS after TensorRT quantization. The paper further claims that the proxy is interpretable: Spearman rank correlation shows depth strongly drives the score ($\rho \approx 0.83$–$0.90$), width and kernel size contribute positively, and transformer dimensions correlate with the score only when CNN capacity is held fixed, which the authors attribute to FLOPs-induced competition between the two module types.

Load-bearing premise

The search rests on the assumption that the training-free entropy score ranks candidate backbones in the same order as their accuracy after real training, and the paper never checks that ranking against even one trained network.

Editorial extensions

If this is right

  • The detector reports state-of-the-art mean average precision on both benchmarks (0.538 and 0.492 mmAP), with the winner on each dataset coming from a different entropy-weight setting (A1 for URPC2021, A2 for URPC2022).
  • Stepwise ablations (RT-DETR → MAE-DETR → NAS-DETR) improve mmAP at each step, which the paper reads as evidence that the entropy-driven search adds real gain on top of the detector design.
  • The searched backbone keeps the full detector real-time at 71.2–73.8 FPS (288–294.9 FPS after TensorRT quantization), so the accuracy gain does not sacrifice deployability.
  • Spearman analysis indicates that backbone depth is the dominant driver of the search score ($\rho = 0.826$–$0.904$), so the search objective implicitly prioritizes depth within the FLOPs budget.

Reading between the lines

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

  • Because the fitness score is never compared with trained accuracy in the paper, the natural next experiment is to train a sample of searched backbones spanning the full range of $Z(G)$ and check that their measured mAP ranking matches the entropy ranking; that would turn the proxy from an assumption into a validated design tool.
  • The variance-propagation formula makes entropy grow multiplicatively with depth, which suggests the search is effectively exploring a depth-versus-width Pareto frontier under the FLOPs constraint; the same optimization could be recast with an explicit Pareto objective.
  • The recipe — entropy proxy, evolutionary search, one final training run — is domain-agnostic in principle, so it should transfer to other low-resolution, high-noise detection tasks such as medical ultrasound, radar, or degraded optical imagery; rerunning the identical search on one such dataset would test that directly.
  • The best entropy-weight setting differs between the two datasets (A1 on URPC2021, A2 on URPC2022), implying the six stage weights act as dataset-dependent hyperparameters; treating them as part of the search space, rather than fixed choices, would make the method fully automatic.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 6 minor

Summary. The paper proposes NAS-DETR, a DETR-style sonar object detector whose CNN-Transformer backbone is selected by a zero-shot, maximum-entropy neural architecture search. Candidate backbones are scored with a differential-entropy fitness function, then combined with a feature pyramid network and a deformable-attention decoder that uses content-position decoupled query initialization, denoising training, and a hybrid loss. Experiments on URPC2021 and URPC2022 report state-of-the-art mmAP of 0.538 and 0.492 at 71.2-73.8 FPS, together with an ablation study and a Spearman correlation analysis between architecture parameters and the entropy-based fitness score.

Significance. If the results hold, the paper addresses an important applied problem and the zero-shot search would be practically valuable because it avoids training during architecture search. The motivation is clear, the experimental comparison is broad, and the idea of searching a CNN-Transformer backbone for low-resolution, noisy sonar images is reasonable. However, the empirical support is currently insufficient: the central SOTA numbers come from single runs with no variance information, one headline result is internally contradictory across tables, the zero-shot fitness is never validated against trained detection accuracy, and the theoretical variance-propagation derivation contains unjustified equalities. The paper also does not provide code or trained models, so the tables are the only evidence for the claims.

major comments (5)
  1. [§4.3 (Table 6) vs. §4.4 (Table 9)] The headline URPC2022 result is internally inconsistent. Section 4.3 and Table 6 report mmAP 0.492 for NAS-DETR(A2) and use it as the SOTA comparison against RT-DETR (0.476), while Table 9 reports NAS-DETR(A1) at 0.492 and NAS-DETR(A2) at 0.488. Since A1 and A2 are the two entropy-weight configurations of the proposed search, the paper does not specify which searched architecture produced the headline result. This contradiction prevents the reader from reproducing the central empirical claim and must be resolved by correcting the tables or rerunning the experiments.
  2. [Tables 4-10] All detection metrics are reported as single values with no standard deviations, number of seeds, or statistical tests. The claimed advantages over RT-DETR are 1.3 mmAP points on URPC2021 and 1.6 mmAP points on URPC2022; on test sets of roughly 800-1000 images, these margins are within the range of typical training randomness for DETR-style detectors. The central SOTA claim is therefore not empirically established; at minimum, three to five independent seeds with reported means and variances, or a significance test, are needed.
  3. [§3.1, Eqs. (15)-(22)] The variance-propagation derivation for the Transformer block is not valid as written. Equation (18) applies Lemma 2 to attention weights a_ij that are stated to lie in (0,1), yet the text asserts E(a_ij)=0, which is impossible for positive values; the independence of attention weights from value features and the treatment of softmax as a negligible scaling are asserted without justification. Equation (21) is dimensionally inconsistent: a product of d_model, d_ff, and S times the input variance cannot equal the variance of y_ij. Equation (22) then multiplies a variance by a sum of logarithms as if it were an entropy. This undercuts the theoretical basis of the fitness function in Eq. (29).
  4. [§3.2 Eq. (29) and §4.4 Tables 8-9] The zero-shot fitness function Z(G) is never validated against trained detection accuracy. The paper shows that the selected backbones perform well, but it does not show that architectures with higher Z(G) actually yield higher mAP after training, nor does it compare Z(G) against random backbone selection. In addition, the ablation in Tables 8-9 changes the decoder, query initialization, and loss function simultaneously with the backbone, so the 1.2-1.4% gains cannot be attributed to the NAS search alone. An isolated backbone-only comparison, or a correlation between Z(G) and trained mAP across a sample of backbones, is required to support the search claim.
  5. [§4.6.3, Eq. (29), Table 11] The correlation analysis is largely tautological. Since the fitness Z(G) in Eq. (29) is a weighted sum of log(var(h^D)) + log(C_in), and the derivation in Eqs. (13)-(22) makes the score increase with depth and width by construction, the strong Spearman correlations between depth and score and between channel count and score are expected from the definition rather than empirical discoveries about sonar architectures. Table 11 also contains an internal inconsistency: the text describes feedforward dimension as showing a significant negative correlation with a reported value of ρ=0.904, but 0.904 is positive. The interpretability claim should be reframed and the numerical reporting corrected.
minor comments (6)
  1. [Throughout] There are multiple typos, including 'Univesity' in the affiliations, 'Porposed' in the section title, 'NAS-SDETR' in Sections 1 and 3, and 'the the' in Section 4.3; these should be corrected.
  2. [Tables 5 and 7] Several numeric entries are concatenated without separators (for example, '0.8870.9050.9590.95' in Table 5 and similar entries in Table 7), making the category-wise mAP50 values unreadable.
  3. [Tables 11-12] The table captions and the text are inconsistent: Table 11 is captioned 'under A1' but reports rows for both Score(A1) and Score(A2), and Table 12 is captioned 'under A2' but similarly mixes settings. The abbreviations 'hdn' and 'btn' are used without definition.
  4. [§4.6.4] The text references 'Table??' instead of a real table number, and the sampling description in Eq. (39) is incomplete because the text does not define how the differential entropy value E_total is computed for the paired samples.
  5. [§4.5] The TensorRT quantization results (294.9 and 288 FPS) are reported without any quantization setup, precision, or reproducibility details, so the deployment claim cannot be assessed.
  6. [References] Reference [43] is cited for the differential-entropy upper bound of a neural network, but the cited paper is about fast FFT-based convolution performance; this appears to be a citation error and should be replaced with the correct source.

Circularity Check

1 steps flagged · score 4.0 of 10

Section 4.6 correlation analysis is definitional (Eq. 14/29 build depth and width into the entropy score), but the headline SOTA benchmark is empirical and independent.

  1. self definitional [Section 4.6, around Eqs. (14), (22), (28)-(29) and Table 11]
    "Notably, the differential entropy was calculated using Eq29 to ensure that the analysis directly reflects the algorithmic behavior of the search process. / HL = LX i=1 log cik2 i (14)"

    Eq. (14) defines the CNN contribution to differential entropy as a sum over layers of log(c_i k_i^2); Eqs. (28)-(29) define the search score Z as a weighted average of log(variance) plus log(C_in) over the six backbone blocks. Depth L enters as the number of additive log terms, and channel count and kernel size enter as the summands, so the score is constructed from the very parameters that Section 4.6 then Spearman-correlates with the score. The reported 'positive correlations' (L: rho=0.904/0.826; channels and kernel size: 0.247-0.443) are a restatement of the definitions rather than an empirical discovery, and the paper even notes that L dominates the raw entropy and adds log(C_in) to encourage width, building the correlation into the objective.

full rationale

The only step that reduces to its own inputs is the Section 4.6 correlation analysis. Eq. (14)/(22) make depth, width, and kernel size algebraic constituents of the differential entropy, and Eq. (28)-(29) turn that same expression into the NAS fitness score; correlating the score with those parameters is therefore definitional. This is a genuine circularity for the paper's 'interpretability' contribution, but it is not the paper's central empirical claim. The headline results (mmAP 0.538/0.492 on URPC2021/2022) are benchmarks against external methods and do not derive from the entropy formula, so the SOTA claim has independent content. Two correctness risks noted by the reviewers -- the URPC2022 tables (Table 6 vs Table 9) disagree on whether A1 or A2 is best, and no error bars or repeated-seed statistics are reported -- are significant but are reproducibility/statistical concerns, not circularity. No load-bearing self-citation chain or imported uniqueness theorem was found; references to the authors' own YOLO papers are peripheral related work. Overall, the paper has one definitional result presented as empirical, but its central evaluation remains externally anchored, so a moderate score is appropriate.

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

The central claim depends on a heuristic fitness function whose weights are hand-chosen, a proxy that is assumed to predict trained performance, and a variance propagation derivation that requires unstated independence assumptions. No new physical or mathematical entities are introduced.

free parameters (5)
  • Entropy weights a1..a6 = A1: {0,0,1,1,2,4}; A2: {0,0,1,1,3,6}
    Hand-chosen weights in Eq. (29) balance differential entropy across six feature scales; two settings are tested but there is no principled derivation or sensitivity analysis.
  • Loss weights = λ_cls:λ_box:λ_dn = 1:2.5:0.5; λ_L1:λ_GIoU = 5:2
    Set via grid search (Section 3.4), so fitted to the datasets used in the paper.
  • Denoising noise sigma = 0.1
    Default value for Gaussian noise added to ground-truth boxes in Eq. (34); no sensitivity study is provided.
  • VFL modulation factor gamma = 2.0
    Default in Varifocal Loss, no ablation is reported.
  • Width compensation term log(C_in) = not stated
    Ad hoc addition to fitness function Z' (Eq. 28) to prevent depth from dominating; no derivation is given.
assumptions (4)
  • ad hoc to paper H(S_v) ≈ H(F') (Eq. 4)
    Asserts that the differential entropy of network vertices equals the entropy of the output feature map under Gaussian assumptions; no proof is given and it is load-bearing for the proxy.
  • domain assumption Weights and inputs follow N(0,1) (Eq. 8)
    Standard simplification for variance propagation, but real initializations (e.g., He or Xavier) and data distributions differ; results are presented as exact.
  • ad hoc to paper Attention weights are independent of values and softmax is negligible (Eqs. 18-20)
    Treats A=Softmax(QK^T/√d) as having sub-unit variance and independent of V; softmax outputs are correlated with Q, K, and hence X and V, so the variance formula is not justified.
  • domain assumption Higher differential entropy implies higher representational capacity (Lemma 1 and refs [41-43])
    The link between feature-map variance under random initialization and trained task performance is imported from prior zero-shot NAS literature; it is not revalidated in this paper.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Underwater object detection in sonar imagery with detection transformer and Zero-shot neural architecture search." pith.science (2026). https://pith.science/paper/T4VJPF6I

@misc{pith2026250506694,
  author       = {Pith},
  title        = {Pith review of: Underwater object detection in sonar imagery with detection transformer and Zero-shot neural architecture search},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/T4VJPF6I}},
  note         = {Machine review of arXiv:2505.06694}
}
read the original abstract

Underwater object detection using sonar imagery has become a critical and rapidly evolving research domain within marine technology. However, sonar images are characterized by lower resolution and sparser features compared to optical images, which seriously degrades the performance of object detection.To address these challenges, we specifically propose a Detection Transformer (DETR) architecture optimized with a Neural Architecture Search (NAS) approach called NAS-DETR for object detection in sonar images. First, an improved Zero-shot Neural Architecture Search (NAS) method based on the maximum entropy principle is proposed to identify a real-time, high-representational-capacity CNN-Transformer backbone for sonar image detection. This method enables the efficient discovery of high-performance network architectures with low computational and time overhead. Subsequently, the backbone is combined with a Feature Pyramid Network (FPN) and a deformable attention-based Transformer decoder to construct a complete network architecture. This architecture integrates various advanced components and training schemes to enhance overall performance. Extensive experiments demonstrate that this architecture achieves state-of-the-art performance on two Representative datasets, while maintaining minimal overhead in real-time efficiency and computational complexity. Furthermore, correlation analysis between the key parameters and differential entropy-based fitness function is performed to enhance the interpretability of the proposed framework. To the best of our knowledge, this is the first work in the field of sonar object detection to integrate the DETR architecture with a NAS search mechanism.

Figures

Figures reproduced from arXiv: 2505.06694 by the authors.

Figure 1
Figure 1. The proposed NAS-DETR framework for sonar image object detection. [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Schematic diagram of the two feedforward modes under [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Schematic diagram of the proposed architecture search strategy and fitness function calculation. [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Mutation operation of the CNN-Transformer hybrid [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: Architectural design of the Decoder and Head components. [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]
Figure 6
Figure 6. Figure 6: Distribution of different types of target samples on the [PITH_FULL_IMAGE:figures/full_fig_p011_6.png]
Figure 7
Figure 7. Figure 7: Visualization of detection results of NAS-DETR and methods in the benchmark on URPC2021. [PITH_FULL_IMAGE:figures/full_fig_p012_7.png]
Figure 8
Figure 8. Figure 8: Visualization of detection results of NAS-DETR and methods in the benchmark on URPC2022. [PITH_FULL_IMAGE:figures/full_fig_p014_8.png]
Figure 9
Figure 9. Figure 9: Sampling distribution of architecture parameter - differential entropy data pairs for various network architectures. [PITH_FULL_IMAGE:figures/full_fig_p017_9.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

50 extracted references · 41 canonical work pages

  1. [1]

    Yolov3-dpfin: A dual-path feature fusion neural network for robust real-time sonar target de- tection.IEEE Sensors Journal, 20(7):3745–3756, 2019

    Wanzeng Kong, Jichen Hong, Mingyang Jia, Jinliang Yao, Weihua Cong, Hua Hu, and Haigang Zhang. Yolov3-dpfin: A dual-path feature fusion neural network for robust real-time sonar target de- tection.IEEE Sensors Journal, 20(7):3745–3756, 2019. 1

    work page 2019

  2. [2]

    Side-scan sonar image segmentation based on multi-channel fusion convolution neural networks.IEEE Sensors Journal, 22(6):5911– 5928, 2022

    Zhen Wang, Jianxin Guo, Wenzhun Huang, and Shanwen Zhang. Side-scan sonar image segmentation based on multi-channel fusion convolution neural networks.IEEE Sensors Journal, 22(6):5911– 5928, 2022. 1

    work page 2022

  3. [3]

    Francisco Francisco and Jan Sundberg. Detection of visual signa- tures of marine mammals and fish within marine renewable energy farms using multibeam imaging sonar.Journal of Marine Science and Engineering, 7(2):22, 2019. 1

    work page 2019

  4. [4]

    Speckle noise reduction in sonar image based on adaptive redundant dictionary.Journal of marine science and engineering, 8(10):761, 2020

    Yifan Huang, Weixiang Li, and Fei Yuan. Speckle noise reduction in sonar image based on adaptive redundant dictionary.Journal of marine science and engineering, 8(10):761, 2020. 1

    work page 2020

  5. [5]

    Object detection in sonar images.Electronics, 9(7):1180, 2020

    Divas Karimanzira, Helge Renkewitz, David Shea, and Jan Albiez. Object detection in sonar images.Electronics, 9(7):1180, 2020. 1, 2

    work page 2020

  6. [6]

    Deep learning algorithms for sonar imagery analysis and its application in aquaculture: A review.IEEE Sensors Journal, 23(23):28549–28563, 2023

    Yingqian Chai, Huihui Yu, Ling Xu, Daoliang Li, and Yingyi Chen. Deep learning algorithms for sonar imagery analysis and its application in aquaculture: A review.IEEE Sensors Journal, 23(23):28549–28563, 2023. 1, 2 18

    work page 2023

  7. [7]

    Fast target detection in synthetic aperture sonar imagery: A new algorithm and large-scale performance analysis

    David P Williams. Fast target detection in synthetic aperture sonar imagery: A new algorithm and large-scale performance analysis. IEEE Journal of Oceanic Engineering, 40(1):71–92, 2014. 1

    work page 2014

  8. [8]

    Automatic detection of underwater chain links using a forward-looking sonar

    Natalia Hurt ´os, Narc´ıs Palomeras, Sharad Nagappa, and Joaquim Salvi. Automatic detection of underwater chain links using a forward-looking sonar. In2013 MTS/IEEE OCEANS-Bergen, pages 1–7. IEEE, 2013. 1

    work page 2013

Show all 50 references
  1. [9]

    Salient object detection: from pixels to segments.Image and Vision Computing, 31(1):31–42, 2013

    Victoria Yanulevskaya, Jasper Uijlings, and Jan-Mark Geusebroek. Salient object detection: from pixels to segments.Image and Vision Computing, 31(1):31–42, 2013. 1

    2013

  2. [10]

    Small target detection method based on low-rank sparse matrix factoriza- tion for side-scan sonar images.Remote Sensing, 15(8):2054, 2023

    Ju He, Jianfeng Chen, Hu Xu, and Muhammad Saad Ayub. Small target detection method based on low-rank sparse matrix factoriza- tion for side-scan sonar images.Remote Sensing, 15(8):2054, 2023. 1

    2023

  3. [11]

    A fast physics-based, en- vironmentally adaptive underwater object detection algorithm

    David P Williams and Johannes Groen. A fast physics-based, en- vironmentally adaptive underwater object detection algorithm. In OCEANS 2011 IEEE-Spain, pages 1–7. IEEE, 2011. 1

    2011

  4. [12]

    Robust sonar- based underwater object recognition against angle-of-view varia- tion.IEEE Sensors Journal, 16(4):1013–1025, 2015

    Hyeonwoo Cho, Jeonghwe Gu, and Son-Cheol Yu. Robust sonar- based underwater object recognition against angle-of-view varia- tion.IEEE Sensors Journal, 16(4):1013–1025, 2015. 1

    2015

  5. [13]

    Semisupervised lo- cal fusion approach for mine detection in sonar data.International Journal of Intelligent Systems, 30(11):1161–1183, 2015

    Mohamed Maher Ben Ismail and Ouiem Bchir. Semisupervised lo- cal fusion approach for mine detection in sonar data.International Journal of Intelligent Systems, 30(11):1161–1183, 2015. 1

    2015

  6. [14]

    Enhanced fuzzy-based local informa- tion algorithm for sonar image segmentation.IEEE transactions on image processing, 29:445–460, 2019

    Avi Abu and Roee Diamant. Enhanced fuzzy-based local informa- tion algorithm for sonar image segmentation.IEEE transactions on image processing, 29:445–460, 2019. 1

    2019

  7. [15]

    Profile fitting-based small target de- tection in water for side-scan sonar image

    Zhanshuo Liu, Xiufen Ye, Shuxiang Guo, Huiming Xing, Zengchao Hao, and Yao Li. Profile fitting-based small target de- tection in water for side-scan sonar image. In2021 IEEE Interna- tional Conference on Mechatronics and Automation (ICMA), pages 275–280. IEEE, 2021. 2

    2021

  8. [16]

    Multi-object detection and tracking, based on dnn, for autonomous vehicles: A review.IEEE Sensors Journal, 21(5):5668–5677,

    Ratheesh Ravindran, Michael J Santora, and Mohsin M Jamali. Multi-object detection and tracking, based on dnn, for autonomous vehicles: A review.IEEE Sensors Journal, 21(5):5668–5677,

  9. [17]

    Improvement of road instance segmentation algorithm based on the modified mask r-cnn.Electronics, 12(22):4699, 2023

    Chenxia Wan, Xianing Chang, and Qinghui Zhang. Improvement of road instance segmentation algorithm based on the modified mask r-cnn.Electronics, 12(22):4699, 2023. 2

    2023

  10. [18]

    Yolo-sgc: A dangerous driving behavior detection method with multiscale spatial-channel feature aggregation.IEEE Sensors Journal, 2024

    Ruijie Li, Changdong Yu, Xiangrong Qin, Xin An, Jinpeng Zhao, Wenhui Chuai, and Baisheng Liu. Yolo-sgc: A dangerous driving behavior detection method with multiscale spatial-channel feature aggregation.IEEE Sensors Journal, 2024. 2, 3

    2024

  11. [19]

    Yolo-mrs: An efficient deep learning- based maritime object detection method for unmanned surface ve- hicles.Applied Ocean Research, 153:104240, 2024

    Changdong Yu, Haoke Yin, Chenyi Rong, Jiayi Zhao, Xiao Liang, Ruijie Li, and Xinrong Mo. Yolo-mrs: An efficient deep learning- based maritime object detection method for unmanned surface ve- hicles.Applied Ocean Research, 153:104240, 2024. 2, 3

    2024

  12. [20]

    Yolov3: An incremental improve- ment.arXiv preprint arXiv:1804.02767, 2018

    Joseph Redmon and Ali Farhadi. Yolov3: An incremental improve- ment.arXiv preprint arXiv:1804.02767, 2018. 2, 3

    2018 arXiv

  13. [21]

    Real-time vehicle detection based on improved yolo v5.Sustainability, 14(19):12274, 2022

    Yu Zhang, Zhongyin Guo, Jianqing Wu, Yuan Tian, Haotian Tang, and Xinming Guo. Real-time vehicle detection based on improved yolo v5.Sustainability, 14(19):12274, 2022. 2, 3

    2022

  14. [22]

    A review on yolov8 and its advancements

    Mupparaju Sohan, Thotakura Sai Ram, Rami Reddy, and Ch Venkata. A review on yolov8 and its advancements. InInterna- tional Conference on Data Intelligence and Cognitive Informatics, pages 529–545. Springer, 2024. 2

    2024

  15. [23]

    Yolov11: An overview of the key architectural enhancements.arXiv preprint arXiv:2410.17725, 2024

    Rahima Khanam and Muhammad Hussain. Yolov11: An overview of the key architectural enhancements.arXiv preprint arXiv:2410.17725, 2024. 2

    2024 arXiv

  16. [24]

    Mlffnet: Multilevel feature fusion network for object de- tection in sonar images.IEEE Transactions on Geoscience and Remote Sensing, 60, 2022

    Zhen Wang, Jianxin Guo, Leya Zeng, Chuanlei Zhang, and Buhong Wang. Mlffnet: Multilevel feature fusion network for object de- tection in sonar images.IEEE Transactions on Geoscience and Remote Sensing, 60, 2022. 2

    2022

  17. [25]

    Multi-scale fusion and efficient feature extraction for en- hanced sonar image object detection.Expert Systems with Appli- cations, 256:124958, 2024

    Pengfei Shi, Qi He, Sisi Zhu, Xinyu Li, Xinnan Fan, and Yuanxue Xin. Multi-scale fusion and efficient feature extraction for en- hanced sonar image object detection.Expert Systems with Appli- cations, 256:124958, 2024. 2

    2024

  18. [26]

    A review of underwater mine detection and clas- sification in sonar imagery.Electronics, 10(23):2943, 2021

    Stanisław Ho ˙zy´n. A review of underwater mine detection and clas- sification in sonar imagery.Electronics, 10(23):2943, 2021. 2

    2021

  19. [27]

    Sonar im- age intelligent processing in seabed pipeline detection: review and application.Measurement Science and Technology, 35(4):045405,

    Bo Shi, Tianyu Cao, Qiqi Ge, Yuan Lin, and Zitao Wang. Sonar im- age intelligent processing in seabed pipeline detection: review and application.Measurement Science and Technology, 35(4):045405,

  20. [28]

    End-to-end object detection with transformers

    Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov, and Sergey Zagoruyko. End-to-end object detection with transformers. InEuropean conference on computer vision, pages 213–229. Springer, 2020. 3

    2020

  21. [31]

    Teach-detr: Better training detr with teachers.IEEE Transactions on Pattern Analysis and Machine In- telligence, 45(12):15759–15771, 2023

    Linjiang Huang, Kaixin Lu, Guanglu Song, Liang Wang, Si Liu, Yu Liu, and Hongsheng Li. Teach-detr: Better training detr with teachers.IEEE Transactions on Pattern Analysis and Machine In- telligence, 45(12):15759–15771, 2023. 3

    2023

  22. [32]

    Dn-detr: Accelerate detr training by introducing query de- noising

    Feng Li, Hao Zhang, Shilong Liu, Jian Guo, Lionel M Ni, and Lei Zhang. Dn-detr: Accelerate detr training by introducing query de- noising. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 13619–13627, 2022. 3

    2022

  23. [33]

    Efficient detr: improving end-to-end object detector with dense prior.arXiv preprint arXiv:2104.01318, 2021

    Zhuyu Yao, Jiangbo Ai, Boxun Li, and Chi Zhang. Efficient detr: improving end-to-end object detector with dense prior.arXiv preprint arXiv:2104.01318, 2021. 3

    2021 arXiv

  24. [34]

    Sparse detr: Efficient end-to-end object detection with learnable sparsity.arXiv preprint arXiv:2111.14330, 2021

    Byungseok Roh, JaeWoong Shin, Wuhyun Shin, and Saehoon Kim. Sparse detr: Efficient end-to-end object detection with learnable sparsity.arXiv preprint arXiv:2111.14330, 2021. 3

    2021 arXiv

  25. [35]

    An- chor detr: Query design for transformer-based detector

    Yingming Wang, Xiangyu Zhang, Tong Yang, and Jian Sun. An- chor detr: Query design for transformer-based detector. InPro- ceedings of the AAAI conference on artificial intelligence, vol- ume 36, pages 2567–2575, 2022. 3

    2022

  26. [36]

    A survey on evolutionary neural architecture search.IEEE transactions on neural networks and learning sys- tems, 34(2):550–570, 2021

    Yuqiao Liu, Yanan Sun, Bing Xue, Mengjie Zhang, Gary G Yen, and Kay Chen Tan. A survey on evolutionary neural architecture search.IEEE transactions on neural networks and learning sys- tems, 34(2):550–570, 2021. 3 19

    2021

  27. [37]

    Nas-fcos: efficient search for object detection architectures.International Journal of Computer Vision, 129:3299–3312, 2021

    Ning Wang, Yang Gao, Hao Chen, Peng Wang, Zhi Tian, Chunhua Shen, and Yanning Zhang. Nas-fcos: efficient search for object detection architectures.International Journal of Computer Vision, 129:3299–3312, 2021. 3

    2021

  28. [38]

    Sm-nas: Structural-to-modular neural architecture search for object detection

    Lewei Yao, Hang Xu, Wei Zhang, Xiaodan Liang, and Zhenguo Li. Sm-nas: Structural-to-modular neural architecture search for object detection. InProceedings of the AAAI conference on artifi- cial intelligence, volume 34, pages 12661–12668, 2020. 3

    2020

  29. [39]

    Au- todet: pyramid network architecture search for object detection

    Zhihang Li, Teng Xi, Gang Zhang, Jingtuo Liu, and Ran He. Au- todet: pyramid network architecture search for object detection. International Journal of Computer Vision, 129:1087–1105, 2021. 3

    2021

  30. [40]

    Zen-nas: A zero-shot nas for high- performance image recognition

    Ming Lin, Pichao Wang, Zhenhong Sun, Hesen Chen, Xiuyu Sun, Qi Qian, Hao Li, and Rong Jin. Zen-nas: A zero-shot nas for high- performance image recognition. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 347–356,

  31. [41]

    Mae-det: Revisiting maximum entropy principle in zero-shot nas for efficient object detection.arXiv preprint arXiv:2111.13336, 2021

    Zhenhong Sun, Ming Lin, Xiuyu Sun, Zhiyu Tan, Hao Li, and Rong Jin. Mae-det: Revisiting maximum entropy principle in zero-shot nas for efficient object detection.arXiv preprint arXiv:2111.13336, 2021. 3

    2021 arXiv

  32. [42]

    Deepmad: Mathematical architecture design for deep convolutional neural network

    Xuan Shen, Yaohua Wang, Ming Lin, Yilun Huang, Hao Tang, Xi- uyu Sun, and Yanzhi Wang. Deepmad: Mathematical architecture design for deep convolutional neural network. InProceedings of the IEEE/CVF conference on computer vision and pattern recogni- tion, pages 6163–6173, 2023. 3

    2023

  33. [43]

    Fast convolutional nets with fbfft: A gpu performance evaluation.arXiv preprint arXiv:1412.7580, 2014

    Nicolas Vasilache, Jeff Johnson, Michael Mathieu, Soumith Chin- tala, Serkan Piantino, and Yann LeCun. Fast convolutional nets with fbfft: A gpu performance evaluation.arXiv preprint arXiv:1412.7580, 2014. 4

    2014 arXiv

  34. [44]

    Cambridge University Press Cambridge, MA, USA, 2022

    Daniel A Roberts, Sho Yaida, and Boris Hanin.The principles of deep learning theory, volume 46. Cambridge University Press Cambridge, MA, USA, 2022. 4, 7

    2022

  35. [45]

    Deformable detr: Deformable transformers for end- to-end object detection.arXiv preprint arXiv:2010.04159, 2020

    Xizhou Zhu, Weijie Su, Lewei Lu, Bin Li, Xiaogang Wang, and Jifeng Dai. Deformable detr: Deformable transformers for end- to-end object detection.arXiv preprint arXiv:2010.04159, 2020. 7

    2010 arXiv

  36. [46]

    Detrs beat yolos on real-time object detection

    Yian Zhao, Wenyu Lv, Shangliang Xu, Jinman Wei, Guanzhong Wang, Qingqing Dang, Yi Liu, and Jie Chen. Detrs beat yolos on real-time object detection. InProceedings of the IEEE/CVF con- ference on computer vision and pattern recognition, pages 16965– 16974, 2024. 7

    2024

  37. [47]

    End-to-end object detection with transformers

    Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov, and Sergey Zagoruyko. End-to-end object detection with transformers. InEuropean conference on computer vision, pages 213–229. Springer, 2020. 8

    2020

  38. [48]

    Dab-detr: Dynamic anchor boxes are better queries for detr.arXiv preprint arXiv:2201.12329, 2022

    Shilong Liu, Feng Li, Hao Zhang, Xiao Yang, Xianbiao Qi, Hang Su, Jun Zhu, and Lei Zhang. Dab-detr: Dynamic anchor boxes are better queries for detr.arXiv preprint arXiv:2201.12329, 2022. 9

    2022 arXiv

  39. [49]

    Dn-detr: Accelerate detr training by introducing query de- noising

    Feng Li, Hao Zhang, Shilong Liu, Jian Guo, Lionel M Ni, and Lei Zhang. Dn-detr: Accelerate detr training by introducing query de- noising. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 13619–13627, 2022. 9

    2022

  40. [50]

    Dual spatial attention network for underwater object detec- tion with sonar imagery.IEEE Sensors Journal, 24(5):6998–7008,

    Zikang Li, Zhuojun Xie, Puhong Duan, Xudong Kang, and Shutao Li. Dual spatial attention network for underwater object detec- tion with sonar imagery.IEEE Sensors Journal, 24(5):6998–7008,

  41. [51]

    A dataset with multibeam forward-looking sonar for underwater object detection.Scientific Data, 9(1):739, 2022

    Kaibing Xie, Jian Yang, and Kang Qiu. A dataset with multibeam forward-looking sonar for underwater object detection.Scientific Data, 9(1):739, 2022. 10

    2022

  42. [52]

    The proof and measurement of associa- tion between two things.The American journal of psychology, 100(3/4):441–471, 1987

    Charles Spearman. The proof and measurement of associa- tion between two things.The American journal of psychology, 100(3/4):441–471, 1987. 15 20

    1987

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.