Pith. sign in

REVIEW 4 major objections 1 minor 52 references

EPANet: Efficient Path Aggregation Network for Underwater Fish Detection

T0 review · 4 major / 1 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read This paper claims EPANet outperforms existing detectors on underwater fish detection in both accuracy and speed while keeping parameter count comparable or lower.

desk verdict Abstract-only submission; the state-of-the-art claim is unsupported because no experiments are in the text, though the architecture idea is coherent. read the letter →

arxiv 2508.00528 v1 pith:EWYQTEBJ submitted 2025-08-01 cs.CV

Jinsong Yang , Zeyuan Hu , Yichen Li This is my paper · ORCID
classification cs.CV
keywords underwaterfishdetectionsmallobjectfeaturepyramidnetworkefficientdeeplearningbottleneckdesignEPA-FPNMS-DDSP
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

Underwater fish detection is hard because fish are small, blend into cluttered backgrounds, and look like their surroundings. The paper introduces EPANet, a detection network built from two pieces: an efficient path aggregation feature pyramid network (EPA-FPN) and a multi-scale diverse-division short path bottleneck (MS-DDSP bottleneck). The claim is that these structural changes, rather than attention modules, improve small-object feature integration, and that EPANet beats state-of-the-art methods in accuracy and inference speed while keeping parameter complexity comparable or lower. A sympathetic reader should care because it suggests accurate fish counting and monitoring can run on lightweight underwater hardware.

What carries the argument

Two named modules carry the argument. EPA-FPN is a feature pyramid network that connects features across distant scales with long-range skip connections and cross-layer fusion paths, so high-level semantics and low-level spatial detail are combined early. The MS-DDSP bottleneck is a modified residual bottleneck that splits its input channels into finer groups and applies different convolutional kernels to each group, increasing feature diversity without widening the network. Together they replace attention-based feature enhancement with cheaper structural integration.

What would settle it

Run EPANet and a matched attention-based detector of the same parameter and FLOP budget on the same underwater fish benchmark; if the attention baseline ties or beats EPANet, or if removing either the EPA-FPN or the MS-DDSP component leaves accuracy unchanged, the claimed mechanism is not what carries the result.

Watch

Extended reading notes

Core claim

EPANet's central claim is that complementary feature integration across scales, without relying on attention mechanisms, can solve the small-object problem in underwater fish detection. EPA-FPN adds long-range skip connections between features at different scales so semantic and spatial information reinforce each other, and uses cross-layer fusion to make the integration cheap. The MS-DDSP bottleneck divides feature channels into finer groups and processes them with different convolution operations, increasing local diversity and representational capacity. On benchmark underwater fish detection datasets, EPANet reportedly reaches state-of-the-art detection accuracy and faster inference with comparable or lower parameter counts. If correct, the discovery is that accuracy and efficiency can both be won through architectural composition rather than added attention.

Load-bearing premise

The gain rests on the assumption that the long-range cross-scale connections and fine-grained split convolutions, not extra capacity or training luck, are what make small fish easier to detect.

Editorial extensions

If this is right

  • If the accuracy and speed claims hold, underwater fish monitoring can be deployed on low-power, small-form-factor devices rather than GPU servers.
  • Attention modules may be unnecessary for small-object detection; structural multi-scale fusion alone could reach state-of-the-art accuracy.
  • The EPA-FPN and MS-DDSP components are modular and could be transplanted into other single-stage detectors to cut computational cost.
  • Benchmark results would establish a new efficiency-accuracy trade-off point for underwater object detection.

Reading between the lines

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

  • The same cross-scale aggregation could transfer to other cluttered small-object domains, such as aerial images, medical slides, or weed detection, since the design does not use fish-specific cues.
  • A direct test: replace the MS-DDSP bottleneck with a standard bottleneck holding width and depth fixed; the paper's reasoning predicts a measurable accuracy drop on small specimens that a FLOPs-matched attention module would not recover.
  • If the gains come mostly from EPA-FPN, the bottleneck design might be simplified further, pushing the parameter count below what the paper reports.
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 / 1 minor

Summary. The manuscript, as submitted, consists solely of an abstract. It proposes EPANet for underwater fish detection (UFD), comprising an efficient path aggregation feature pyramid network (EPA-FPN) and a multi-scale diverse-division short path bottleneck (MS-DDSP). The abstract claims that EPANet outperforms state-of-the-art methods in detection accuracy and inference speed while maintaining comparable or lower parameter complexity, on benchmark UFD datasets. No further content—no experimental results, datasets, tables, figures, implementation details, or references—is provided in the full text.

Significance. If the performance claims were substantiated, the proposed architecture could be of interest to the UFD community: the ideas of long-range cross-scale skip connections for semantic-spatial complementarity and finer-grained feature division with diverse convolutions are plausible directions for lightweight detectors. However, because the manuscript contains no empirical evidence whatsoever, the significance cannot be assessed. The claim of state-of-the-art accuracy and speed is entirely unsupported, and the architectural descriptions are too high-level to allow reproducibility or informed judgment. The contribution, as it stands, is an unverified assertion rather than a demonstrated result.

major comments (4)
  1. [Abstract] The central claim—that EPANet outperforms state-of-the-art methods in detection accuracy and inference speed—is unsupported. The manuscript contains no experimental section, no dataset names, no evaluation metrics, no baseline comparisons, and no result tables. Without this evidence, the claim is untestable and cannot be accepted.
  2. [Full Text (empty)] The full text of the manuscript is empty, so none of the usual elements of a scientific paper are present: no introduction, no related work, no architectural details beyond the abstract's brief description, no training or inference protocols, no ablation studies, and no analysis. The paper as submitted provides no basis for evaluating the proposed method.
  3. [Abstract (EPA-FPN and MS-DDSP)] The descriptions of the two key components are too vague to assess their novelty or technical correctness. For example, the abstract states that EPA-FPN 'introduces long-range skip connections across disparate scales' and that MS-DDSP 'extends the conventional bottleneck structure,' but no concrete formulas, layer configurations, or design choices are given. This prevents any meaningful evaluation of the proposed contributions.
  4. [Abstract (state-of-the-art comparison)] The abstract refers to 'state-of-the-art methods' without identifying any specific prior work. Without a clear list of compared methods, their architectures, or the benchmarks used, the claimed superiority cannot be contextualized or verified.
minor comments (1)
  1. [General] The manuscript is incomplete as submitted; the full-text body is blank, leaving only the abstract. This is a fundamental presentation issue that must be addressed before any substantive review can occur.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity found: the submitted text contains only an abstract-level empirical claim, with no derivation chain, fitted parameters, or self-citations to reduce.

full rationale

The provided full text is effectively blank apart from the abstract; there are no equations, no experimental protocols, no baseline definitions, and no references. The central claim that EPANet outperforms state-of-the-art methods in accuracy and inference speed while maintaining comparable or lower parameter complexity is an empirical assertion. Its support would require experimental evidence, but the absence of that evidence is a missing-evidence problem, not circularity. No step defines one quantity in terms of another by construction, no fitted parameter is renamed as a prediction, and no load-bearing uniqueness theorem is imported from the authors' prior work. Because the paper's architectural description is purely qualitative, there is no derivation chain to walk and nothing to reduce to its own inputs. The appropriate finding is therefore no significant circularity, score 0.

Assumptions & free parameters 0 free parameters · 3 assumptions · 2 invented entities

Because the full text is not provided, the ledger can only capture assumptions visible from the abstract. The main unseen inputs are the benchmark datasets, the training protocol, and the soundness of the baseline implementations.

assumptions (3)
  • domain assumption Underwater fish detection benchmark datasets used in the experiments are valid, comparable, and correctly labeled.
    The abstract claims performance on benchmark UFD datasets but does not name them.
  • domain assumption Existing feature pyramid networks and bottleneck architectures are a sound foundation that EPANet extends.
    The paper builds on established deep learning components; the abstract does not question them.
  • domain assumption The reported inference speed comparisons used identical hardware and software settings.
    Speed comparisons can be skewed by implementation details; no protocol is given.
invented entities (2)
  • EPA-FPN (efficient path aggregation feature pyramid network)
    purpose: Integrate features across scales with long-range skip connections and cross-layer fusion for small fish detection.
    This is a newly named architectural module; no independent validation outside the paper is reported in the abstract.
  • MS-DDSP bottleneck
    purpose: Increase local feature diversity by dividing features into fine groups and applying diverse convolutions.
    Another new module introduced to improve representation capacity; no standalone evidence is provided.

how reviews work

0 comments
Cite this review

Pith. "Pith review of EPANet: Efficient Path Aggregation Network for Underwater Fish Detection." pith.science (2026). https://pith.science/paper/EWYQTEBJ

@misc{pith2026250800528,
  author       = {Pith},
  title        = {Pith review of: EPANet: Efficient Path Aggregation Network for Underwater Fish Detection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/EWYQTEBJ}},
  note         = {Machine review of arXiv:2508.00528}
}
read the original abstract

Underwater fish detection (UFD) remains a challenging task in computer vision due to low object resolution, significant background interference, and high visual similarity between targets and surroundings. Existing approaches primarily focus on local feature enhancement or incorporate complex attention mechanisms to highlight small objects, often at the cost of increased model complexity and reduced efficiency. To address these limitations, we propose an efficient path aggregation network (EPANet), which leverages complementary feature integration to achieve accurate and lightweight UFD. EPANet consists of two key components: an efficient path aggregation feature pyramid network (EPA-FPN) and a multi-scale diverse-division short path bottleneck (MS-DDSP bottleneck). The EPA-FPN introduces long-range skip connections across disparate scales to improve semantic-spatial complementarity, while cross-layer fusion paths are adopted to enhance feature integration efficiency. The MS-DDSP bottleneck extends the conventional bottleneck structure by introducing finer-grained feature division and diverse convolutional operations, thereby increasing local feature diversity and representation capacity. Extensive experiments on benchmark UFD datasets demonstrate that EPANet outperforms state-of-the-art methods in terms of detection accuracy and inference speed, while maintaining comparable or even lower parameter complexity.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

52 extracted references · 49 canonical work pages

  1. [1]

    Deep learning-based visual detection of marine organisms: A survey

    Ning Wang, Tingkai Chen, Shaoman Liu, Rongfeng Wang, Hamid Reza Karimi, and Yejin Lin. Deep learning-based visual detection of marine organisms: A survey. Neurocomputing , 532:1--32, 2023

    work page 2023

  2. [2]

    From classical approach to deep-learning: A review on underwater target segmentation with sonar image

    Wenyu Cai, Jifeng Zhu, and Meiyan Zhang. From classical approach to deep-learning: A review on underwater target segmentation with sonar image. Neurocomputing , 637:130087, 2025

    work page 2025

  3. [3]

    Review on deep learning techniques for marine object recognition: Architectures and algorithms

    Ning Wang, Yuanyuan Wang, and Meng Joo Er. Review on deep learning techniques for marine object recognition: Architectures and algorithms. Control Engineering Practice , 118:104458, 2022

    work page 2022

  4. [4]

    A systematic review and analysis of deep learning-based underwater object detection

    Shubo Xu, Minghua Zhang, Wei Song, Haibin Mei, Qi He, and Antonio Liotta. A systematic review and analysis of deep learning-based underwater object detection. Neurocomputing , 527:204--232, 2023

    work page 2023

  5. [5]

    Real-time underwater object detection technology for complex underwater environments based on deep learning

    Hui Zhou, Meiwei Kong, Hexiang Yuan, Yanyan Pan, Xinru Wang, Rong Chen, Weiheng Lu, Ruizhi Wang, and Qunhui Yang. Real-time underwater object detection technology for complex underwater environments based on deep learning. Ecological Informatics , 82:102680, 2024

    work page 2024

  6. [6]

    Advancing underwater vision: A survey of deep learning models for underwater object recognition and tracking

    Mahmoud Elmezain, Lyes Saad Saoud, Atif Sultan, Mohamed Heshmat, Lakmal Seneviratne, and Irfan Hussain. Advancing underwater vision: A survey of deep learning models for underwater object recognition and tracking. IEEE Access , 13:17830--17867, 2025

    work page 2025

  7. [7]

    Underwater image processing and analysis: A review

    Muwei Jian, Xiangyu Liu, Hanjiang Luo, Xiangwei Lu, Hui Yu, and Junyu Dong. Underwater image processing and analysis: A review. Signal Processing: Image Communication , 91:116088, 2021

    work page 2021

  8. [8]

    Rethinking general underwater object detection: Datasets, challenges, and solutions

    Chenping Fu, Risheng Liu, Xin Fan, Puyang Chen, Hao Fu, Wanqi Yuan, Ming Zhu, and Zhongxuan Luo. Rethinking general underwater object detection: Datasets, challenges, and solutions. Neurocomputing , 517:243--256, 2023

    work page 2023

Show all 52 references
  1. [9]

    Review of intelligent detection and health assessment of underwater structures

    Shuai Teng, Airong Liu, Xijun Ye, Jialin Wang, Jiyang Fu, Zhihua Wu, Bingcong Chen, Chao Liu, Haoxiang Zhou, Yuxin Zeng, and Jie Yang. Review of intelligent detection and health assessment of underwater structures. Engineering Structures , 308:117958, 2024

    2024

  2. [10]

    Enhancing aerial object detection with selective frequency interaction network

    Weijie Weng, Mengwan Wei, Junchi Ren, and Fei Shen. Enhancing aerial object detection with selective frequency interaction network. IEEE Transactions on Artificial Intelligence , 1(01):1--12, 2024

    2024

  3. [11]

    Lr-fpn: Enhancing remote sensing object detection with location refined feature pyramid network

    Hanqian Li, Ruinan Zhang, Ye Pan, Junchi Ren, and Fei Shen. Lr-fpn: Enhancing remote sensing object detection with location refined feature pyramid network. arXiv preprint arXiv:2404.01614 , 2024

    2024 arXiv

  4. [12]

    A novel multi-frequency coordinated module for sar ship detection

    Chenchen Qiao, Fei Shen, Xuejun Wang, Ruixin Wang, Fang Cao, Sixian Zhao, and Chang Li. A novel multi-frequency coordinated module for sar ship detection. In 2022 IEEE 34th International Conference on Tools with Artificial Intelligence (ICTAI) , pages 804--811. IEEE, 2022

    2022

  5. [13]

    A novel cross frequency-domain interaction learning for aerial oriented object detection

    Weijie Weng, Weiming Lin, Feng Lin, Junchi Ren, and Fei Shen. A novel cross frequency-domain interaction learning for aerial oriented object detection. In Chinese Conference on Pattern Recognition and Computer Vision (PRCV) , pages 292--305. Springer, 2023

    2023

  6. [14]

    Fahad Hasan Bhuiyan Emon , Md Rakibul Haque, Arif Reza Anwary, and Swakkhar Shatabda

    Abdullah Al Muksit, Fakhrul Hasan, Md. Fahad Hasan Bhuiyan Emon , Md Rakibul Haque, Arif Reza Anwary, and Swakkhar Shatabda. Yolo-fish: A robust fish detection model to detect fish in realistic underwater environment. Ecological Informatics , 72:101847, 2022

    2022

  7. [15]

    Deepfish: Accurate underwater live fish recognition with a deep architecture

    Hongwei Qin, Xiu Li, Jian Liang, Yigang Peng, and Changshui Zhang. Deepfish: Accurate underwater live fish recognition with a deep architecture. Neurocomputing , 187:49--58, 2016

    2016

  8. [16]

    Ceh-yolo: A composite enhanced yolo-based model for underwater object detection

    Jiangfan Feng and Tao Jin. Ceh-yolo: A composite enhanced yolo-based model for underwater object detection. Ecological Informatics , 82:102758, 2024

    2024

  9. [17]

    U-yolov7: A network for underwater organism detection

    Guoyan Yu, Ruilin Cai, Jinping Su, Mingxin Hou, and Ruoling Deng. U-yolov7: A network for underwater organism detection. Ecological Informatics , 75:102108, 2023

    2023

  10. [18]

    An improved feature pyramid network for object detection

    Linxiang Zhu, Feifei Lee, Jiawei Cai, Hongliu Yu, and Qiu Chen. An improved feature pyramid network for object detection. Neurocomputing , 483:127--139, 2022

    2022

  11. [19]

    Weighted parallel decoupled feature pyramid network for object detection

    Bo Han, Lihuo He, Junjie Ke, Chenwei Tang, and Xinbo Gao. Weighted parallel decoupled feature pyramid network for object detection. Neurocomputing , 593:127809, 2024

    2024

  12. [20]

    An embedded feature pyramid network enables bidirectional information flow for object detection and instance segmentation

    Chunning Meng, Zequn Sun, Lianzhi Huo, Tao Li, Shengjiang Chang, and Zhiqing Zhang. An embedded feature pyramid network enables bidirectional information flow for object detection and instance segmentation. Neurocomputing , 619:129196, 2025

    2025

  13. [21]

    Fcanet: A novel feature fusion approach to eeg emotion recognition based on cross-attention networks

    Mingjie Li, Heming Huang, and Kedi Huang. Fcanet: A novel feature fusion approach to eeg emotion recognition based on cross-attention networks. Neurocomputing , 638:130102, 2025

    2025

  14. [22]

    Mff-net: A multi-view feature fusion network for generalized forgery image detection

    Ying Lin, Tenglong Mao, Ziyi Chen, Hong Lu, Zhaoyu Chen, and Yan Kang. Mff-net: A multi-view feature fusion network for generalized forgery image detection. Neurocomputing , 640:130351, 2025

    2025

  15. [23]

    Lightweight oriented object detection with dynamic smooth feature fusion network

    Iftikhar Ahmad, Wei Lu, Si-Bao Chen, Jin Tang, and Bin Luo. Lightweight oriented object detection with dynamic smooth feature fusion network. Neurocomputing , 628:129725, 2025

    2025

  16. [24]

    Path aggregation network for instance segmentation

    Shu Liu , Lu Qi , Haifang Qin , Jianping Shi , and Jiaya Jia . Path aggregation network for instance segmentation. arXiv e-prints , page arXiv:1803.01534, 2018

    2018 arXiv

  17. [25]

    Mingxing Tan, Ruoming Pang, and Quoc V. Le. Efficientdet: Scalable and efficient object detection. 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pages 10778--10787, 2020

    2020

  18. [26]

    Golnaz Ghiasi, Tsung-Yi Lin, and Quoc V. Le. Nas-fpn: Learning scalable feature pyramid architecture for object detection. 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pages 7029--7038, 2019

    2019

  19. [27]

    Attentional feature fusion

    Yimian Dai, Fabian Gieseke, Stefan Oehmcke, Yiquan Wu, and Kobus Barnard. Attentional feature fusion. Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) , pages 3560--3569, 2021

    2021

  20. [28]

    Small object detection network based on progressive enhanced multi-level feature fusion

    Yanshan Li, Fuxing Liu, Yusong Qin, Linhui Dai, Li Zhang, and Weixin Xie. Small object detection network based on progressive enhanced multi-level feature fusion. Neurocomputing , 649:130755, 2025

    2025

  21. [29]

    Rocknet: Deep progressive lithology recognition model based on feature saliency and fusion

    Xiangyuan Zhu, Mincan Li, Zhiming Lan, Jianguo Chen, Zerui Li, and Keqin Li. Rocknet: Deep progressive lithology recognition model based on feature saliency and fusion. Neurocomputing , 616:128898, 2025

    2025

  22. [30]

    Fast r-cnn

    Ross Girshick. Fast r-cnn. 2015 IEEE International Conference on Computer Vision (ICCV) , pages 1440--1448, 2015

    2015

  23. [31]

    Xudong Sun, Pengcheng Wu, and Steven C.H. Hoi. Face detection using deep learning: An improved faster rcnn approach. Neurocomputing , 299:42--50, 2018

    2018

  24. [32]

    Underwater fish detection in sonar image based on an improved faster rcnn

    Di Zhao, Bo Yang, Yinke Dou, and Xiaojia Guo. Underwater fish detection in sonar image based on an improved faster rcnn. 2022 9th International Forum on Electrical Engineering and Automation (IFEEA) , pages 358--363, 2022

    2022

  25. [33]

    Fernandes

    Dewei Yi, Hasan Bayarov Ahmedov, Shouyong Jiang, Yiren Li, Sean Joseph Flinn, and Paul G. Fernandes. Coordinate-aware mask r-cnn with group normalization: A underwater marine animal instance segmentation framework. Neurocomputing , 583:127488, 2024

    2024

  26. [34]

    Sur-net: A deep network for fish detection and segmentation with limited training data

    Huei-Yung Lin, Shih-Lun Tseng, and Jun-Yi Li. Sur-net: A deep network for fish detection and segmentation with limited training data. IEEE Sensors Journal , 22:18035--18044, 2022

    2022

  27. [35]

    Imagpose: A unified conditional framework for pose-guided person generation

    Fei Shen and Jinhui Tang. Imagpose: A unified conditional framework for pose-guided person generation. Advances in neural information processing systems , 37:6246--6266, 2024

    2024

  28. [36]

    Imagdressing-v1: Customizable virtual dressing

    Fei Shen, Xin Jiang, Xin He, Hu Ye, Cong Wang, Xiaoyu Du, Zechao Li, and Jinhui Tang. Imagdressing-v1: Customizable virtual dressing. In Proceedings of the AAAI Conference on Artificial Intelligence , volume 39, pages 6795--6804, 2025

    2025

  29. [37]

    Atrous spatial pyramid convolution for object detection with encoder-decoder

    Feiran Jie, Qingfeng Nie, Mingsuo Li, Ming Yin, and Taisong Jin. Atrous spatial pyramid convolution for object detection with encoder-decoder. Neurocomputing , 464:107--118, 2021

    2021

  30. [38]

    An efficient re-parameterization feature pyramid network on yolov8 to the detection of steel surface defect

    Weining Xie, Weifeng Ma, and Xiaoyong Sun. An efficient re-parameterization feature pyramid network on yolov8 to the detection of steel surface defect. Neurocomputing , 614:128775, 2025

    2025

  31. [39]

    Feature pyramid network with multi-scale prediction fusion for real-time semantic segmentation

    Toan Van Quyen and Min Young Kim. Feature pyramid network with multi-scale prediction fusion for real-time semantic segmentation. Neurocomputing , 519:104--113, 2023

    2023

  32. [40]

    Dynamic feature pyramid networks for object detection

    Mingjian Zhu , Kai Han , Changbin Yu , and Yunhe Wang . Dynamic feature pyramid networks for object detection. arXiv e-prints , page arXiv:2012.00779, 2022

    2012 arXiv

  33. [41]

    Pedestrian-specific bipartite-aware similarity learning for text-based person retrieval

    Fei Shen, Xiangbo Shu, Xiaoyu Du, and Jinhui Tang. Pedestrian-specific bipartite-aware similarity learning for text-based person retrieval. In Proceedings of the 31th ACM International Conference on Multimedia , 2023

    2023

  34. [42]

    Triplet contrastive learning for unsupervised vehicle re-identification

    Fei Shen, Xiaoyu Du, Liyan Zhang, and Jinhui Tang. Triplet contrastive learning for unsupervised vehicle re-identification. arXiv preprint arXiv:2301.09498 , 2023

    2023 arXiv

  35. [43]

    Yolov8 by ultralytics

    Ayush Chaurasia Glenn Jocher and Jing Qiu. Yolov8 by ultralytics. 2023

    2023

  36. [44]

    Yolov10: Real-time end-to-end object detection

    Ao Wang, Hui Chen, Lihao Liu, Kai Chen, Zijia Lin, Jungong Han, and Guiguang Ding. Yolov10: Real-time end-to-end object detection. 38th Conference on Neural Information Processing Systems (NeurIPS 2024) , 2024

    2024

  37. [45]

    Ultralytics yolo11

    Glenn Jocher and Jing Qiu. Ultralytics yolo11. 2024

    2024

  38. [46]

    Mask r-cnn

    Kaiming He, Georgia Gkioxari, Piotr Dollár, and Ross Girshick. Mask r-cnn. 2017 IEEE International Conference on Computer Vision (ICCV) , pages 2980--2988, 2017

    2017

  39. [47]

    Yolov5 by ultralytics

    Glenn Jocher. Yolov5 by ultralytics. 2020

    2020

  40. [48]

    Fish detection method based on improved yolov5

    Lei Li, Guosheng Shi, and Tao Jiang. Fish detection method based on improved yolov5. Aquaculture International , 31:2513--2530, 2023

    2023

  41. [49]

    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. Computer Vision -- ECCV 2020 , pages 213--229, 2020

    2020

  42. [50]

    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. 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pages 16965--16974, 2024

    2024

  43. [51]

    A stochastic approximation method

    Herbert Robbins and Sutton Monro. A stochastic approximation method. The annals of mathematical statistics , page 400–407, 1951

    1951

  44. [52]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.