Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

EPBC-YOLOv8: An efficient and accurate improved YOLOv8 underwater detector based on an attention mechanism

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

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

Pith's one-line read EPBC-YOLOv8 is a modified YOLOv8 that achieves mAP@0.5 of 76.7% and 79.0% on the underwater benchmarks URPC2019 and URPC2020, exceeding the original YOLOv8 by 2.3 and 0.7 percentage points while using fewer parameters and FLOPs.

desk verdict A routine YOLOv8 modification with plausible but statistically unsupported gains: +2.3 mAP on URPC2019, +0.7 on URPC2020, no error bars, and a garbled ablation table. read the letter →

arxiv 2502.05788 v1 pith:ZIBFDKUV submitted 2025-02-09 cs.CV cs.AI

Xing Jiang , Xiting Zhuang , Jisheng Chen , Jian Zhang This is my paper · ORCID
classification cs.CVcs.AI
keywords underwaterobjectdetectionYOLOv8attentionmechanismEMApointwiseconvolutionBiFPNCARAFEURPC2019
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 claims that a modified YOLOv8 detector, called EPBC-YOLOv8, detects marine organisms in degraded underwater images more accurately and with less computation than the standard model. On the public URPC2019 and URPC2020 benchmarks it reports mAP@0.5 of 76.7% and 79.0%, which are 2.3 and 0.7 percentage points above YOLOv8, while the n-sized model uses 2.6 million parameters and 7.6 GFLOPs versus 3.0 million and 8.2 GFLOPs for YOLOv8n. The gain is attributed to four modifications: EMA attention in the backbone, pointwise-convolution-based FasterPW blocks, a BiFPN-inspired weighted feature pyramid, and CARAFE content-aware upsampling. The paper's ablation identifies C2f_EMA as the largest single contributor, alone raising mAP@0.5 by 1.3 points.

What carries the argument

The machinery is a chain of four plug-in components inside YOLOv8. C2f_EMA inserts the Efficient Multi-Scale Attention module, with a widened 5x5 branch, into the backbone's C2f blocks; FasterPW swaps partial convolution for 1x1 pointwise convolution inside FasterNeXt-style blocks; WFPN replaces plain Concat with a weighted Concat along BiFPN-inspired bidirectional cross-scale links; and CARAFE upsampling predicts content-aware reassembly kernels instead of using interpolation. The ablation singles out C2f_EMA as the largest single contributor, raising mAP@0.5 by 1.3 points over the baseline.

What would settle it

Re-running EPBC-YOLOv8 and YOLOv8n on the official URPC training and validation files with identical hyperparameters and several random seeds; if the 2.3-point and 0.7-point mAP margins do not reproduce, or if the splits differ from those in the cited papers, the claimed improvement is not established.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that combining attention, lightweight convolution, weighted cross-scale fusion, and content-aware upsampling in a single YOLOv8 architecture yields a detector that is both lighter and more accurate for underwater organisms. The reported numbers are mAP@0.5 of 76.7% on URPC2019 and 79.0% on URPC2020, beating the original YOLOv8 by 2.3 and 0.7 points respectively. The authors also report that each of the four modules individually improves over the baseline and that the best result comes when all four are used together.

Load-bearing premise

The central comparison assumes the URPC2019 and URPC2020 validation sets used here are exactly the same as those behind the previously published numbers, so the reported margins are like-for-like.

Editorial extensions

If this is right

  • The full EPBC-YOLOv8n model beats YOLOv8n on both benchmarks (76.7 vs 74.4 on URPC2019, 79.0 vs 78.3 on URPC2020) while using fewer parameters and FLOPs.
  • Every one of the four proposed modules improves individually on the baseline, and the complete combination is better than any subset tested.
  • The C2f_EMA module alone is responsible for most of the gain over the baseline according to the ablation study.
  • The paper's per-class results show the largest improvement on the hardest category, waterweeds, whose AP rises from 32.3% to 45.5%.

Reading between the lines

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

  • The claimed margin over the baseline may depend on the choice of validation split; re-evaluating all compared models on identical, standard URPC splits with multiple seeds would settle whether the advantage is stable.
  • Because the four modules act on generic YOLO components (C2f blocks, neck concatenation, and upsampling), a likely testable extension is porting the same recipe to newer YOLO versions or to other one-stage detectors.
  • The paper does not report small-object-specific average precision; if the benefit of CARAFE and C2f_EMA is mainly for small targets, a COCO-style AP_s analysis on URPC would make that mechanism explicit.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper proposes EPBC-YOLOv8, an underwater object detector built on YOLOv8n by adding four components: an EMA-based attention module integrated into C2f (C2f_EMA), a FasterNeXt-derived block with pointwise convolutions (FasterPW), a BiFPN-inspired weighted feature pyramid network (WFPN), and CARAFE content-aware upsampling. The authors report mAP@0.5 of 76.7% on URPC2019 and 79.0% on URPC2020, corresponding to +2.3 and +0.7 points over the original YOLOv8n, with modest reductions in parameters and FLOPs. A 15-configuration ablation study is used to attribute the gain to the four modules.

Significance. If the reported gains are reproducible, EPBC-YOLOv8 is a slightly more accurate and lighter YOLOv8 variant for underwater detection, and the combination of established modules is a reasonable incremental contribution. The paper's strengths are its use of public benchmarks, the intention to evaluate each module through ablation, and the promised public code and data. However, the central accuracy claim currently rests on single training runs without error bars or significance testing; differences of 0.1-0.7 mAP points on validation sets of 800-942 images are plausibly within seed-to-seed noise. The efficiency claim is supported only by parameter/FLOP counts, not by measured latency or throughput. With repeated seeded runs, a corrected ablation table, and a fully specified training protocol, the work could become a useful practical study.

major comments (4)
  1. [Experimental Details; Tables 3 and 4] The central claim of +2.3 and +0.7 mAP over YOLOv8n is based on a single training run per configuration, with no error bars, no repeated seeds, and no significance test. On a validation set of 800 images (URPC2020), a 0.7-point difference is well within the typical run-to-run variation of YOLO-scale detectors, so the URPC2020 improvement is not statistically established. Similarly, the ablation differences of 0.1-0.2 points (e.g., 76.6 vs 76.4 vs 76.7 in Table 6) are smaller than unmeasured stochasticity and cannot be interpreted. Please report the mean and standard deviation over at least 3-5 seeds for the baseline and the proposed model, and state whether the improvement is significant (e.g., a paired test across seeds).
  2. [Table 6] The ablation table is rendered as an unparsable grid of checkmarks and mAP values; for example, the string '76.0√ √ 75.8√ √ 75.8√ √ 75.1√ √ 75.4√ √ 76.0√ √ √ 76.2...' does not make clear which modules are included in each row. This prevents the reader from verifying the claim that all four modules contribute and that the full combination is best. Please replace Table 6 with a conventional ablation table: one row per configuration, explicit module columns (C2f_EMA, FasterPW, WFPN, CARAFE), and separate columns for mAP@0.5, parameters, and FLOPs.
  3. [Parameter Settings; Table 1] Table 1 lists both the initial and the final learning rate as 1e-2, which is inconsistent with the standard YOLOv8 cosine annealing schedule and would imply a constant learning rate. The paper does not specify the optimizer, momentum, weight decay, augmentation, or whether the YOLOv8 baseline in Tables 3-4 was trained under exactly this protocol or taken from another source. Without a complete training protocol, the comparison may not be apples-to-apples. Please specify the exact schedule and hyperparameters, and state for each baseline whether it was reproduced in this work or quoted from literature.
  4. [Benchmark Testing and Implementation Details] The comparisons against Boosting RCNN (Table 3, from Ref. 37) and YOLOv5x/Faster-RCNN (Table 4, from Ref. 38) presume that the train/validation splits of URPC2019 and URPC2020 are identical to those used in the cited papers. The manuscript does not state how the splits were obtained or whether the baseline numbers were re-evaluated on the same split. If the splits differ, the claimed margins over those methods are not meaningful. Please specify the exact split (file indices or split definition) and, ideally, re-run the principal baselines under the same protocol.
minor comments (5)
  1. [C2f_EMA] The text first states that the EMA module was incorporated into the neck part of YOLOv8, but the following paragraph and Table 5 describe modifications to the first C2f in the backbone; please reconcile this inconsistency.
  2. [Result Analysis] The description of the loss curves mentions 'the loss function for keypoint detection', but the model is an object detector without keypoint heads; please correct or clarify what is plotted.
  3. [FasterPW] The text cites reference [44] for PWConv as a local channel context aggregator, but the reference list contains only 38 entries; please fix the citation.
  4. [Table 4] The YOLOv8 row in Table 4 does not specify the variant (n/s/m/l/x) or per-class AP values; please identify the exact model used and report the per-class APs for completeness.
  5. [Conclusions] The Abstract and Conclusions state the URPC2020 improvement as a positive result, yet Section 'Comparative Experiments' acknowledges the improvement is 'not very significant'; please align the wording with the statistical evidence.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: this is an empirical integration paper whose reported mAP values are measured, not derived.

full rationale

The paper contains no derivation chain that could be circular: EPBC-YOLOv8 is defined as YOLOv8 plus four externally published components (EMA, FasterPW, WFPN-like weighted fusion, CARAFE), and the central claims are the measured mAP@0.5 values in Tables 3, 4, and 6. Equations (1)-(6) define pooling, IoU, precision, recall, AP, and mAP; none of these equations rewrites the model output as an input. The 2.3% and 0.7% gains over YOLOv8 are arithmetic differences between the paper's own runs (74.4 to 76.7 on URPC2019; 78.3 to 79.0 on URPC2020), not fitted parameters renamed as predictions. The only self-citations are refs 4 and 38, used in Table 4 as benchmark numbers for YOLOv5x, Faster-RCNN, and a literature method; those numbers are externally published results and do not define the proposed model's mAP, which is measured with the paper's own training and evaluation protocol. The unresolved split-comparability and seed-variability concerns are correctness or robustness risks, not instances of circularity. Therefore the circularity score is 0.

Assumptions & free parameters 1 free parameters · 3 assumptions · 0 invented entities

The paper contributes no new theory or derived quantities. The central claim depends on the validity of YOLOv8 and the four borrowed modules, the correctness of the benchmark splits, and the specific training hyperparameters, all of which are taken as given from prior work or the authors' choices.

free parameters (1)
  • Training hyperparameters (epochs=100, batch=16, LR=1e-2, NMS IoU=0.7, image size=640) = epochs=100, batch=16, LR=1e-2, NMS IoU=0.7, img=640
    These are chosen by hand and reported without a tuning study; the claimed mAP values depend on them, but they are standard settings, so they do not amount to a fitted physical parameter.
assumptions (3)
  • domain assumption YOLOv8 architecture and its official pretrained weights are an appropriate, valid baseline for underwater object detection.
    Invoked in Section 'Benchmark Testing and Implementation Details' ('we use official document as the pre-trained weight file for all experiments'); if the baseline is misconfigured, the reported +2.3% gain is meaningless.
  • domain assumption EMA, CARAFE, FasterNeXt, and BiFPN behave on this task as described in their source papers (refs 8,13,15,12) with no hidden interactions.
    The paper does not re-derive or validate these modules; it assumes their published designs and performance characteristics transfer.
  • domain assumption Standard public splits of URPC2019 and URPC2020 are used without leakage.
    The paper does not specify how splits were created; the validation mAP is only comparable if the splits match prior work.

how reviews work

0 comments
Cite this review

Pith. "Pith review of EPBC-YOLOv8: An efficient and accurate improved YOLOv8 underwater detector based on an attention mechanism." pith.science (2026). https://pith.science/paper/ZIBFDKUV

@misc{pith2026250205788,
  author       = {Pith},
  title        = {Pith review of: EPBC-YOLOv8: An efficient and accurate improved YOLOv8 underwater detector based on an attention mechanism},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZIBFDKUV}},
  note         = {Machine review of arXiv:2502.05788}
}
read the original abstract

In this study, we enhance underwater target detection by integrating channel and spatial attention into YOLOv8's backbone, applying Pointwise Convolution in FasterNeXt for the FasterPW model, and leveraging Weighted Concat in a BiFPN-inspired WFPN structure for improved cross-scale connections and robustness. Utilizing CARAFE for refined feature reassembly, our framework addresses underwater image degradation, achieving mAP at 0.5 scores of 76.7 percent and 79.0 percent on URPC2019 and URPC2020 datasets, respectively. These scores are 2.3 percent and 0.7 percent higher than the original YOLOv8, showcasing enhanced precision in detecting marine organisms.

Figures

Figures reproduced from arXiv: 2502.05788 by the authors.

Figure 1
Figure 1. The structure of EPBC-YOLOv8. • Parallel Structure: EMA employs a parallel architecture to process input data, differing from the sequential layer structure of traditional Convolutional Neural Networks (CNNs). This parallel structure improves the efficiency of model training and enhances the model’s accuracy when dealing with multi-scale features. • Feature Map Grouping: EMA groups input feature maps along the chann… view at source ↗
Figure 2
Figure 2. Schematic diagram of EMA. Here, ’g’ represents grouping, ’X Avg Pool’ represents 1D horizontal global pooling, and ’Y Avg Pool’ represents 1D vertical global pooling. zc = 1 H ×W ∑ j ∑ i xc (i, j) (1) Here, zc represents the output associated with the c-th channel. The primary purpose of this output is to encode global information, thereby capturing and modeling long-range dependencies. In the YOLOv8 architecture, t… view at source ↗
Figure 3
Figure 3. The structure of the C2f_EMA module. In the Bottleneck module, T/F indicates whether a shortcut is used. T stands for true, and F stands for false. 4/16 [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (8 more)
Figure 4
Figure 4. Figure 4: The construction process of the FasterPW structure. This structure adopts a lightweight design concept, initially ap￾plying PWConv to the FasterPWBlock network, then combining multiple FasterPWBlock and multiple ConvModules to construct an efficient and lightweight fea…
Figure 5
Figure 5. Figure 5: Comparison of different FPN structures, including the structures of (a) FPN, (b) PANet, (c) NASFPN, (d) SimplifiedPANet, (e) BiFPN and (f)WFPN networks. CARAFE In multi-scale image labeling detection, feature upsampling is a key step. Traditional upsampling techniques …
Figure 6
Figure 6. Figure 6: The overall framework of CARAFE, consisting of the kernel prediction module and content-aware assembly module. and lower computational complexity, and is more sensitive to details. Furthermore, the semantically enriched upsampling features generated by CARAFE significa…
Figure 7
Figure 7. Figure 7: Example images from the URPC2019 dataset. 7/16 [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]
Figure 8
Figure 8. Figure 8: Comparison of object detection results between different models [PITH_FULL_IMAGE:figures/full_fig_p010_8.png]
Figure 9
Figure 9. Figure 9: Trends of various parameters of the EPBC-YOLOv8 model over 100 epochs. Conclusions and Future Work In this study, we propose EPBC-YOLOv8,an improved underwater organism detection framework, for complex underwater environments. In the backbone network of YOLOv8, we intr…
Figure 10
Figure 10. Figure 10: Confusion matrix of the EPBC-YOLOv8 model [PITH_FULL_IMAGE:figures/full_fig_p013_10.png]
Figure 11
Figure 11. Figure 11: Precision-Recall (PR) curve of the EPBC-YOLOv8 model. 13/16 [PITH_FULL_IMAGE:figures/full_fig_p013_11.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Improve Underwater Object Detection through YOLOv12 Architecture and Physics-informed Augmentation

    cs.CV 2025-06 conditional novelty 4.0 of 10

    Applying YOLOv12 with physics-flavored augmentations yields high reported mAP on four underwater detection benchmarks, but the claims are weakened by missing code, variance, and inconsistent speed numbers.

Reference graph

Works this paper leans on

36 extracted references · 23 canonical work pages · cited by 1 Pith paper

  1. [2]

    Potential Efficiency of Earth Observation for Optimum Fishing Zone Detection of the Pelagic Sardinella Aurita Species along the Mediterranean Coast of Egypt

    Ali, E.M.; Zanaty, N.; Abou El-Magd, I. Potential Efficiency of Earth Observation for Optimum Fishing Zone Detection of the Pelagic Sardinella Aurita Species along the Mediterranean Coast of Egypt. Fishes 2022, 7, 97.https://doi.org/10.3390/fishes7030097

    work page doi:10.3390/fishes7030097 2022

  2. [3]

    Selvaraj, J.J.; Rosero-Henao, L.V .; Cifuentes-Ossa, M.A. Projecting Future Changes in Distributions of Small-Scale Pelagic Fisheries of the Southern Colombian Pacific Ocean.Heliyon 2022, 8, e08975.https://doi.org/10.1016/j.heliyon.2022.e08975

    work page doi:10.1016/j.heliyon.2022.e08975 2022

  3. [4]

    A Marine Organism Detection Framework Based on the Joint Optimization of Image Enhancement and Object Detection

    Zhang, X.; Fang, X.; Pan, M.; Yuan, L.; Zhang, Y .; Yuan, M.; Lv, S.; Yu, H. A Marine Organism Detection Framework Based on the Joint Optimization of Image Enhancement and Object Detection. Sensors 2021, 21, 7205. https://doi.org/10.3390/s21217205

    work page doi:10.3390/s21217205 2021

  4. [5]

    Enhancing Underwater Image via Adaptive Color and Contrast Enhancement, and Denois- ing.Engineering Applications of Artificial Intelligence 2022 , 111, 104759

    Li, X.; Hou, G.; Li, K.; Pan, Z. Enhancing Underwater Image via Adaptive Color and Contrast Enhancement, and Denois- ing.Engineering Applications of Artificial Intelligence 2022 , 111, 104759. https://doi.org/10.1016/j.engappai.2022.104759

    work page arXiv 2022

  5. [6]

    Enhanced Frequency Fusion Network with Dynamic Hash Attention for Image Denoising.Information Fusion 2023 , 92, 420–434

    Jiang, B.; Li, J.; Li, H.; Li, R.; Zhang, D.; Lu, G. Enhanced Frequency Fusion Network with Dynamic Hash Attention for Image Denoising.Information Fusion 2023 , 92, 420–434 . https://doi.org/10.1016/j.inffus.2022.12.015

    work page doi:10.1016/j.inffus.2022.12.015 2023

  6. [7]

    Attention Guided Low-Light Image Enhancement with a Large Scale Low-Light Simulation Dataset 2020

    Lv, F.; Li, Y .; Lu, F. Attention Guided Low-Light Image Enhancement with a Large Scale Low-Light Simulation Dataset 2020

    work page 2020

  7. [9]

    Mnih, V .; Heess, N.; Graves, A.; kavukcuoglu, koray Recurrent Models of Visual Attention. In Proceedings of the Advances in Neural Information Processing Systems; Ghahramani, Z., Welling, M., Cortes, C., Lawrence, N., Weinberger, K.Q., Eds.; Curran Associates, Inc., 2014; V ol. 27

    work page 2014

  8. [10]

    Feature Pyramid Networks for Object Detection

    Lin, T.-Y .; Dollar, P.; Girshick, R.; He, K.; Hariharan, B.; Belongie, S. Feature Pyramid Networks for Object Detection. In Proceedings of the 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR);IEEE: Honolulu, HI, July 2017; pp. 936–944.https://doi.org/10.1109/CVPR.2017.106

    doi:10.1109/cvpr.2017.106 2017

Show all 36 references
  1. [11]

    NAS-FPN: Learning Scalable Feature Pyramid Architecture for Object Detection

    Ghiasi, G.; Lin, T.-Y .; Le, Q.V . NAS-FPN: Learning Scalable Feature Pyramid Architecture for Object Detection. In Proceedings of the 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR); IEEE, June 2019. https://doi.org/10.1109/cvpr.2019.00720

    2019

  2. [12]

    EfficientDet: Scalable and Efficient Object Detection

    Tan, M.; Pang, R.; Le, Q.V . EfficientDet: Scalable and Efficient Object Detection. In Proceedings of the 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR); IEEE: Seattle, W A, USA, June 2020; pp. 10778–10787. https://doi.org/10.1109/icassp49357.2023.10096516

    2020

  3. [13]

    CARAFE: Content-Aware ReAssembly of FEatures

    Wang, J.; Chen, K.; Xu, R.; Liu, Z.; Loy, C.C.; Lin, D. CARAFE: Content-Aware ReAssembly of FEatures. In Proceedings of the 2019 IEEE/CVF International Conference on Computer Vision (ICCV); IEEE, October 2019. https://doi.org/10.1109/iccv.2019.00310

    2019

  4. [14]

    YOLOv8 by Ultralytics

    Jocher, G. YOLOv8 by Ultralytics. 2023. Available online: https://github.com/ultralytics/ultralytics(accessed on 15 February 2023)

    2023

  5. [16]

    Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation

    Girshick, R.; Donahue, J.; Darrell, T.; Malik, J. Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation. In Proceedings of the 2014 IEEE Conference on Computer Vision and Pattern Recognition; IEEE, June

    2014

  6. [17]

    Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks

    Ren, S.; He, K.; Girshick, R.; Sun, J. Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks. IEEE Transactions on Pattern Analysis and Machine Intelligence 2017 , 39, 1137–1149 . https://doi.org/10.1109/tpami.2016.2577031

    2017

  7. [18]

    You Only Look Once: Unified, Real-Time Object Detection

    Redmon, J.; Divvala, S.; Girshick, R.; Farhadi, A. You Only Look Once: Unified, Real-Time Object Detection. In Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR); IEEE, June 2016. https://doi.org/10.1109/cvpr.2016.91 14/16

    2016 doi

  8. [19]

    YOLO9000: Better, Faster, Stronger

    Redmon, J.; Farhadi, A. YOLO9000: Better, Faster, Stronger. In Proceedings of the 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR); IEEE, July 2017. https://doi.org/10.1109/cvpr.2017.690

    2017 doi

  9. [20]

    Yolov3: An incremental improvement

    Redmon, J.; Farhadi, A. Yolov3: An incremental improvement. arXiv 2018, arXiv:1804.02767

    2018 arXiv

  10. [21]

    Yolov4: Optimal speed and accuracy of object detection

    Bochkovskiy, A.; Wang, C.Y .; Liao, H.Y .M. Yolov4: Optimal speed and accuracy of object detection. arXiv 2020, arXiv:2004.10934

    2020 arXiv

  11. [22]

    YOLOv5 by Ultralytics

    Jocher, G. YOLOv5 by Ultralytics. 2022. Available online: https://github.com/ultralytics/yolov5(accessed on 22 December 2022)

    2022

  12. [23]

    YOLOv6: A Single-Stage Object Detection Framework for Industrial Applications 2022.arXiv 2022, arXiv:2209.02976

    Li, C.; Li, L.; Jiang, H.; Weng, K.; Geng, Y .; Li, L.; Ke, Z.; Li, Q.; Cheng, M.; Nie, W.; et al. YOLOv6: A Single-Stage Object Detection Framework for Industrial Applications 2022.arXiv 2022, arXiv:2209.02976

    2022 arXiv

  13. [24]

    YOLOv7: Trainable Bag-of-Freebies Sets New State-of-the-Art for Real- Time Object Detectors

    Wang, C.-Y .; Bochkovskiy, A.; Liao, H.-Y .M. YOLOv7: Trainable Bag-of-Freebies Sets New State-of-the-Art for Real- Time Object Detectors. In Proceedings of the 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR); IEEE, June 2023.https://doi.org/10.1109/...

    2023

  14. [25]

    SSD: Single Shot MultiBox Detector

    Liu, W.; Anguelov, D.; Erhan, D.; Szegedy, C.; Reed, S.; Fu, C.-Y .; Berg, A.C. SSD: Single Shot MultiBox Detector. In Lecture Notes in Computer Science; Springer International Publishing, 2016; pp. 21–37 ISBN 978-3-319-46448-0

    2016

  15. [26]

    Focal Loss for Dense Object Detection

    Lin, T.-Y .; Goyal, P.; Girshick, R.; He, K.; Dollar, P. Focal Loss for Dense Object Detection. In Proceedings of the 2017 IEEE International Conference on Computer Vision (ICCV); IEEE, October 2017.https://doi.org/10.1109/iccv.2017.324

    2017 doi

  16. [27]

    Underwater Object Detection Using TC-YOLO with Attention Mechanisms

    Liu, K.; Peng, L.; Tang, S. Underwater Object Detection Using TC-YOLO with Attention Mechanisms. Sensors 2023, 23, 2567.https://doi.org/10.3390/s23052567

    2023 doi

  17. [28]

    Fish Detection under Occlusion Using Modified You Only Look Once v8 Integrating Real-Time Detection Transformer Features

    Li, E.; Wang, Q.; Zhang, J.; Zhang, W.; Mo, H.; Wu, Y . Fish Detection under Occlusion Using Modified You Only Look Once v8 Integrating Real-Time Detection Transformer Features. Applied Sciences 2023, 13, 12645. https://doi.org/10.3390/app13231264

    2023 doi

  18. [29]

    Student Behavior Detection in the Classroom Based on Improved YOLOv8

    Chen, H.; Zhou, G.; Jiang, H. Student Behavior Detection in the Classroom Based on Improved YOLOv8. Sensors 2023, 23, 8385. https://doi.org/10.3390/s23208385

    2023 doi

  19. [30]

    CBAM: Convolutional Block Attention Module

    Woo, S.; Park, J.; Lee, J.-Y .; Kweon, I.S. CBAM: Convolutional Block Attention Module. InLecture Notes in Computer Science; Springer International Publishing, 2018; pp. 3–19 ISBN 978-3-030-01234-2

    2018

  20. [31]

    Residual Attention Network for Image Classification

    Wang, F.; Jiang, M.; Qian, C.; Yang, S.; Li, C.; Zhang, H.; Wang, X.; Tang, X. Residual Attention Network for Image Classification. In Proceedings of the 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR); IEEE: Honolulu, HI, USA, July 2017; pp. 6450–6458.h...

    2017 doi

  21. [32]

    Meta-SR: A Magnification-Arbitrary Network for Super-Resolution

    Hu, X.; Mu, H.; Zhang, X.; Wang, Z.; Tan, T.; Sun, J. Meta-SR: A Magnification-Arbitrary Network for Super-Resolution. In Proceedings of the 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR); IEEE: Long Beach, CA, USA, June 2019; pp. 1575–1584. https:/...

    2019

  22. [33]

    Cattle Body Detection Based on YOLOv5-EMA for Precision Livestock Farming.Animals 2023, 13, 3535

    Hao, W.; Ren, C.; Han, M.; Zhang, L.; Li, F.; Liu, Z. Cattle Body Detection Based on YOLOv5-EMA for Precision Livestock Farming.Animals 2023, 13, 3535. [https://doi.org/10.3390/ani13223535]

    2023 doi

  23. [34]

    MobileNetV2: Inverted Residuals and Linear Bottlenecks

    Sandler, M.; Howard, A.; Zhu, M.; Zhmoginov, A.; Chen, L.-C. MobileNetV2: Inverted Residuals and Linear Bottlenecks. In Proceedings of the 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition; IEEE, June 2018. https://doi.org/10.1109/cvpr.2018.00474]

    2018

  24. [35]

    MSA-YOLO: A Remote Sensing Object Detection Model Based on Multi-Scale Strip Attention.Sensors 2023, 23, 6811

    Su, Z.; Yu, J.; Tan, H.; Wan, X.; Qi, K. MSA-YOLO: A Remote Sensing Object Detection Model Based on Multi-Scale Strip Attention.Sensors 2023, 23, 6811. https://doi.org/10.3390/s23156811]

    2023 doi

  25. [36]

    Transformer-Based Object Detection with Deep Feature Fusion Using Carafe Operator in Remote Sensing Image.EAI Endorsed Transactions on Energy Web2023, 10

    Chen, S.; Wang, B.; Zhong, C. Transformer-Based Object Detection with Deep Feature Fusion Using Carafe Operator in Remote Sensing Image.EAI Endorsed Transactions on Energy Web2023, 10. https://doi.org/10.4108/ew.3404

    doi

  26. [37]

    Boosting R-CNN: Reweighting R-CNN Samples by RPN’s Error for Underwater Object Detection.Neurocomputing 2023 , 530, 150–164

    Song, P.; Li, P.; Dai, L.; Wang, T.; Chen, Z. Boosting R-CNN: Reweighting R-CNN Samples by RPN’s Error for Underwater Object Detection.Neurocomputing 2023 , 530, 150–164 . https://doi.org/10.1016/j.neucom.2023.01.088

    2023 doi

  27. [38]

    An Improved YOLOv5-Based Underwater Object-Detection Framework.Sensors 2023, 23, 3693

    Zhang, J.; Zhang, J.; Zhou, K.; Zhang, Y .; Chen, H.; Yan, X. An Improved YOLOv5-Based Underwater Object-Detection Framework.Sensors 2023, 23, 3693. https://doi.org/10.3390/s23073693 Author contributions statement Conceptualization, J.Z.; Methodology, X.Z. and X.J.; Software, ...

    2023 doi

  28. [2014]

    https://doi.org/10.1109/cvpr.2014.81

    2014 doi

Pith tools

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