Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

YOLOv1 to YOLOv11: A Comprehensive Survey of Real-Time Object Detection Innovations and Challenges

T0 review · 4 major / 5 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 survey claims to provide a systematic, up-to-date review of YOLO architectures from v1 through v11 by organizing the family's evolution along five innovation axes: backbone, neck, detection head, loss and assignment, and training…

desk verdict A useful but sloppy YOLO survey that needs a correction pass before it can be trusted as a reference. read the letter →

arxiv 2508.02067 v1 pith:HZ7BHNPE submitted 2025-08-04 cs.CV

Manikanta Kotthapalli , Deepika Ravipati , Reshma Bhatia This is my paper
classification cs.CV
keywords ObjectDetectionYOLOReal-timeDeepLearningSurveyBenchmarksArchitectureEvolution
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper is a survey of the YOLO family of real-time object detectors, aiming to provide a systematic, up-to-date review from v1 through v11. It claims the series' development can be understood along five innovation axes—backbone, neck, detection head, loss and assignment, and training strategy—and it consolidates architectural changes, benchmark numbers, deployment trade-offs, and applications for each version. If the survey succeeds, a reader could use it as a single reference to follow how YOLO moved from a single regression network to an anchor-free, multi-task, and eventually NMS-free detector family, and to see which open problems remain.

What carries the argument

The organizing mechanism of the paper is the five-axis innovation taxonomy (backbone, neck, detection head, loss and assignment, training strategies). This taxonomy is what converts a sequence of version histories into a comparative framework, and it is operationalized by the summary tables (Table II for performance benchmarks and Table III for publication modality). The taxonomy does the argumentative work: it lets the authors attribute each version's contribution to a specific architectural component and then identify trends such as the shift toward anchor-free and NMS-free detection.

What would settle it

Compare Table II against the original papers: if YOLOv4's 43.5% and YOLOv7's 56.8% appear in the column headed mAP@0.5 while the cited sources report those values as average precision (AP), then the survey's benchmark synthesis is factually wrong and the quantitative claims do not hold.

Watch

Extended reading notes

Core claim

On its own terms, the paper's central claim is that a five-axis innovation taxonomy—backbone, neck, detection head, loss and assignment, and training strategy—captures the evolution of the YOLO series from v1 through v11. Each version is presented as a change on one or more of these axes: YOLOv1 reframes detection as a single regression problem; YOLOv2 adds anchors and multi-scale training; YOLOv3 introduces a residual backbone and multi-scale prediction; YOLOv4 contributes the bag of freebies and bag of specials; YOLOv5 focuses on engineering and exportability; YOLOv6 introduces a decoupled head; YOLOv7 develops E-ELAN; YOLOv8 moves to anchor-free unified tasks; and YOLOv9 uses GELAN. The survey then reads YOLOv10 and YOLOv11 as continuations of this design, with NMS-free training and efficiency gains.

Load-bearing premise

The review's reliability rests on the accuracy of the benchmark numbers and version attributions it aggregates from the original YOLO papers and Ultralytics documentation, so its central claim fails if a reported metric is mislabeled or a cited reference does not actually describe the version it is attached to.

Editorial extensions

If this is right

  • If the survey is correct, a practitioner can use the five-axis taxonomy to choose a YOLO variant by deciding which component matters most for a given deployment.
  • The reported benchmarks imply that anchor-free and NMS-free designs have become the dominant direction, because later versions remove hand-crafted components without losing speed.
  • The survey's account implies that YOLO's advantage is the speed-accuracy balance, so any fair comparison against two-stage detectors must weight latency as a primary metric.
  • The challenges section identifies small-object detection, high-IoU localization, training complexity, and domain shift as the problems future YOLO versions are expected to solve.
  • The applications section treats YOLO as a general-purpose perception module rather than a pure detector, which would make its evolution relevant beyond computer vision research.

Reading between the lines

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

  • The five-axis taxonomy could be tested on real-time detectors outside the YOLO family; if their innovations also fit the axes, the taxonomy would be a general design space rather than a historical record.
  • A follow-up verification of YOLOv10 and v11 numbers against independent benchmarks would be a natural extension, since the survey's coverage of those versions relies primarily on vendor documentation.
  • The taxonomy suggests that changing only the loss/assignment axis should affect convergence and localization but not throughput, a prediction that could be checked by ablating a single version.
  • Given the survey's emphasis on deployment trade-offs, future YOLO work may be evaluated as much on edge latency and domain-shift resilience as on COCO mAP, a shift the survey itself anticipates but does not advocate.
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. This manuscript is a survey of the YOLO family of real-time object detectors, covering versions from YOLOv1 through YOLOv11. It proposes a five-axis taxonomy (backbone, neck, detection head, loss/assignment, training strategies), devotes one section to each major version, provides a comparative table of benchmarks and architectural features, discusses application domains, and lists open challenges and future research directions. The paper explicitly claims in its conclusion to provide 'a systematic, up-to-date review of YOLO architectures from v1 through v11.'

Significance. If the survey's factual content were reliable, it would be a useful single reference for practitioners and researchers seeking an overview of YOLO's architectural evolution, benchmark trends, and deployment considerations. The broad historical narrative is largely consistent with the known development of the YOLO family, and the taxonomy-driven organization is a reasonable structuring device. However, the paper's value as a reference depends on the accuracy of its aggregated tables, metric labels, and citations, and several load-bearing errors currently undermine that value. These errors are concrete and fixable, so the contribution is potentially salvageable through revision.

major comments (4)
  1. [Abstract and Section XII] The manuscript is internally inconsistent about which YOLO version is the latest. The abstract says 'from YOLOv1's pioneering regression-based detection to the latest YOLOv9,' and Section XII begins 'YOLOv9 [29] is the latest evolution in the YOLO series,' yet the title, Section XVI, Table IV, and the conclusion (Section XIX-A) all cover YOLOv10 and YOLOv11. This inconsistency directly contradicts the paper's central claim of being an up-to-date review through v11 and must be resolved.
  2. [Table II and Section XIII-B] The main comparative table mislabels its headline metric. The column is headed 'mAP@0.5 (COCO),' but the YOLOv4 row reports 43.5% (AP), the YOLOv7 row reports 56.8% (AP), and the YOLOv9 row reports 56.0%+ without specifying the threshold, while Section XIII-B acknowledges that 'later versions like YOLOv4, YOLOv7, and YOLOv9 adopted the more rigorous AP.' Mixing AP and mAP@0.5 in a single column makes cross-version numeric comparisons in Section XIII-B and subsequent discussion misleading, and it invalidates the table as a reliable benchmark summary.
  3. [References [34] and Section XVI-A] Reference [34] is dated 2023 but describes YOLOv11, which Section XVI-B correctly states was released by Ultralytics in 2024; as printed, the citation cannot be correct and must be corrected or removed. Additionally, Section XVI-A contains an unedited placeholder '[?]' at the end of the paragraph describing YOLOv10's performance, which is not acceptable in a manuscript submitted for review.
  4. [Table III] The publication-modality table is misleading: it lists YOLOv6 and YOLOv7 as having 'Yes' in the 'Peer-Reviewed Paper' column, but the cited sources for these versions are arXiv preprints, not peer-reviewed publications. The table's note defines 'Yes' ambiguously ('peer-reviewed venue or open-source release'), which conflates two distinct properties; the table should be corrected or the column headers should be clarified so that the peer-review status of each version is accurately represented.
minor comments (5)
  1. [Abstract] The abstract contains a duplicated period ('advancements.. Beyond') and should be proofread for punctuation and grammar.
  2. [Section I] The introduction says the paper presents a comprehensive review 'from YOLOv1 through YOLOv9, with brief discussions of emerging models like YOLOv10 and YOLOv11,' but the title and conclusion claim coverage through v11; the stated scope should be aligned with the actual title and content.
  3. [Section XVI-A] The discussion of YOLOv10 cites no reference for the version's introducing paper; a citation to the original YOLOv10 paper should be added.
  4. [Section XVI-C] The reported figure 'YOLOv11n achieves a mAP of 39.5' does not specify the IoU threshold or benchmark variant; adding the metric definition would improve precision.
  5. [Section XIII-B] The claim that 'YOLOv9-L further advanced detection performance with over 56% mAP' conflicts with the typical reported AP for YOLOv9-L on COCO; the authors should verify this number and report the exact metric and source.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation found; the survey aggregates external sources, and its data-quality issues are accuracy concerns, not circularity.

full rationale

This paper is a literature survey, not a derivation or empirical study. It contains no fitted parameters, no equations, no predictions derived from internal assumptions, and no reliance on the authors' prior work. The central claim is that the survey provides a systematic, up-to-date review of YOLO architectures; that claim is supported or undermined by citation accuracy and benchmark-table correctness, which are external-evidence issues rather than circularity. The noted problems in Table II (mixing AP values into a column labeled mAP@0.5) and reference [34] (dated 2023 while describing YOLOv11) are real reliability defects, but they are not cases where a claimed result reduces to its own inputs by construction. The unedited '[?]' placeholder in Section XVI-A similarly indicates incomplete editing, not a circular step. None of the enumerated circularity patterns can be exhibited with a quote showing a definitional equivalence, a fitted input renamed as a prediction, or a load-bearing self-citation chain. Under the review rules, an honest non-finding is appropriate, so the circularity score is 0.

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

No free parameters or invented entities. The paper's load-bearing assumptions are about the fidelity of its citations and benchmark tables.

assumptions (2)
  • domain assumption Quoted benchmark numbers are faithful to the original papers they cite.
    The review's value as a reference rests on accurate transmission of metrics; Table II mixes AP and mAP@0.5 values, so this assumption is partially violated.
  • domain assumption Referenced publications exist and describe the respective YOLO versions.
    Reference [34] attributes a 2023 journal article to YOLOv11, which was released in 2024, suggesting the citation is fabricated or misattributed.

how reviews work

0 comments
Cite this review

Pith. "Pith review of YOLOv1 to YOLOv11: A Comprehensive Survey of Real-Time Object Detection Innovations and Challenges." pith.science (2026). https://pith.science/paper/HZ7BHNPE

@misc{pith2026250802067,
  author       = {Pith},
  title        = {Pith review of: YOLOv1 to YOLOv11: A Comprehensive Survey of Real-Time Object Detection Innovations and Challenges},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/HZ7BHNPE}},
  note         = {Machine review of arXiv:2508.02067}
}
read the original abstract

Over the past decade, object detection has advanced significantly, with the YOLO (You Only Look Once) family of models transforming the landscape of real-time vision applications through unified, end-to-end detection frameworks. From YOLOv1's pioneering regression-based detection to the latest YOLOv9, each version has systematically enhanced the balance between speed, accuracy, and deployment efficiency through continuous architectural and algorithmic advancements.. Beyond core object detection, modern YOLO architectures have expanded to support tasks such as instance segmentation, pose estimation, object tracking, and domain-specific applications including medical imaging and industrial automation. This paper offers a comprehensive review of the YOLO family, highlighting architectural innovations, performance benchmarks, extended capabilities, and real-world use cases. We critically analyze the evolution of YOLO models and discuss emerging research directions that extend their impact across diverse computer vision domains.

Figures

Figures reproduced from arXiv: 2508.02067 by the authors.

Figure 1
Figure 1. Intersection over Union (IoU) is computed as the ratio of the [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Radar plot of the relative emphasis of each YOLO version on the five [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

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

  1. A Leaf-Level Dataset for Soybean-Cotton Detection and Segmentation

    cs.CV 2025-03 unverdicted novelty 6.0 of 10

    A new leaf-instance dataset for soybean-cotton detection and segmentation collected across growth stages and conditions from commercial farms is presented and validated with YOLOv11.

Reference graph

Works this paper leans on

40 extracted references · 26 canonical work pages · cited by 1 Pith paper

  1. [34]

    Yolov11: Revolutionizing agricultural fruitlet detection with enhanced accuracy and real-time deployment,

    P. Rajendran, R. P. Rajasekaran, P. K. Jena, M. N. Wunnava, M. L. Wunnava, and S. Chitrakala, “Yolov11: Revolutionizing agricultural fruitlet detection with enhanced accuracy and real-time deployment,” Computer Standards & Interfaces , vol. 89, p. 103764, 2023. [Online]. Available: https://doi.org/10.1016/j.csi.2023.103764 13

    work page arXiv 2023

  2. [29]

    Yolov9: Learning what you want to learn using programmable gradient information,

    C.-Y . Wang, A. Bochkovskiy, and H.-Y . M. Liao, “Yolov9: Learning what you want to learn using programmable gradient information,” 2024, arXiv preprint arXiv:2402.13616

    arXiv 2024

  3. [1]

    Rich feature hierarchies for accurate object detection and semantic segmentation,

    R. Girshick, J. Donahue, T. Darrell, and J. Malik, “Rich feature hierarchies for accurate object detection and semantic segmentation,” Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 580–587, 2014

    work page 2014

  4. [2]

    Fast r-cnn,

    R. Girshick, “Fast r-cnn,” Proceedings of the IEEE international con- ference on computer vision , pp. 1440–1448, 2015

    work page 2015

  5. [3]

    Faster r-cnn: Towards real-time object detection with region proposal networks,

    S. Ren, K. He, R. Girshick, and J. Sun, “Faster r-cnn: Towards real-time object detection with region proposal networks,” Advances in neural information processing systems , vol. 28, 2015

    2015

  6. [4]

    You only look once: Unified, real-time object detection,

    J. Redmon, S. Divvala, R. Girshick, and A. Farhadi, “You only look once: Unified, real-time object detection,” Proceedings of the IEEE conference on computer vision and pattern recognition , pp. 779–788, 2016

    work page 2016

  7. [5]

    Discriminatively trained deformable part models, release 1,

    P. Felzenszwalb, D. McAllester, and D. Ramanan, “Discriminatively trained deformable part models, release 1,” University of Chicago, Tech. Rep, vol. 2, no. 2, 2008

    work page 2008

  8. [6]

    Se- lective search for object recognition,

    J. R. Uijlings, K. E. van de Sande, T. Gevers, and A. W. Smeulders, “Se- lective search for object recognition,” International journal of computer vision, vol. 104, no. 2, pp. 154–171, 2013

    work page 2013

Show all 40 references
  1. [7]

    Ssd: Single shot multibox detector,

    W. Liu, D. Anguelov, D. Erhan, C. Szegedy, S. Reed, C.-Y . Fu, and A. C. Berg, “Ssd: Single shot multibox detector,” in European Conference on Computer Vision. Springer, 2016, pp. 21–37

    2016

  2. [8]

    Going deeper with convolutions,

    C. Szegedy, W. Liu, Y . Jia, P. Sermanet, S. Reed, D. Anguelov, D. Erhan, V . Vanhoucke, and A. Rabinovich, “Going deeper with convolutions,” in Proceedings of the IEEE conference on computer vision and pattern recognition, 2015, pp. 1–9

    2015

  3. [9]

    Yolo9000: Better, faster, stronger,

    J. Redmon and A. Farhadi, “Yolo9000: Better, faster, stronger,” Proceed- ings of the IEEE conference on computer vision and pattern recognition, pp. 7263–7271, 2017

    2017

  4. [10]

    Batch normalization: Accelerating deep network training by reducing internal covariate shift,

    S. Ioffe and C. Szegedy, “Batch normalization: Accelerating deep network training by reducing internal covariate shift,” International Conference on Machine Learning , pp. 448–456, 2015

    2015

  5. [11]

    Yolov3: An incremental improvement,

    J. Redmon and A. Farhadi, “Yolov3: An incremental improvement,” arXiv preprint arXiv:1804.02767 , 2018

    2018 arXiv

  6. [12]

    Deep residual learning for image recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” Proceedings of the IEEE conference on computer vision and pattern recognition , pp. 770–778, 2016

    2016

  7. [13]

    Focal loss for dense object detection,

    T.-Y . Lin, P. Goyal, R. Girshick, K. He, and P. Dollar, “Focal loss for dense object detection,” in Proceedings of the IEEE international conference on computer vision , 2017, pp. 2980–2988

    2017

  8. [14]

    Yolov4: Op- timal speed and accuracy of object detection,

    A. Bochkovskiy, C.-Y . Wang, and H.-Y . M. Liao, “Yolov4: Op- timal speed and accuracy of object detection,” arXiv preprint arXiv:2004.10934, 2020

    2004 arXiv

  9. [15]

    Cspnet: A new backbone that can enhance learning capability of cnn,

    C.-Y . Wang, H.-Y . M. Liao, Y .-H. Wu, P.-Y . Chen, J.-W. Hsieh, and I.- H. Yeh, “Cspnet: A new backbone that can enhance learning capability of cnn,” Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, pp. 390–391, 2020

    2020

  10. [16]

    Cutmix: Reg- ularization strategy to train strong classifiers with localizable features,

    S. Yun, D. Han, S. J. Oh, S. Chun, J. Choe, and Y . Yoo, “Cutmix: Reg- ularization strategy to train strong classifiers with localizable features,” Proceedings of the IEEE/CVF International Conference on Computer Vision, pp. 6023–6032, 2019

    2019

  11. [17]

    Dropblock: A regularization method for convolutional networks,

    G. Ghiasi, T.-Y . Lin, and Q. V . Le, “Dropblock: A regularization method for convolutional networks,” Advances in neural information processing systems, vol. 31, 2018

    2018

  12. [18]

    Mish: A self regularized non-monotonic activation function,

    D. Misra, “Mish: A self regularized non-monotonic activation function,” arXiv preprint arXiv:1908.08681 , 2019

    1908 arXiv

  13. [19]

    Spatial pyramid pooling in deep convolutional networks for visual recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Spatial pyramid pooling in deep convolutional networks for visual recognition,” in European conference on computer vision . Springer, 2014, pp. 346–361

    2014

  14. [20]

    Path aggregation network for instance segmentation,

    S. Liu, L. Qi, H. Qin, and J. Jia, “Path aggregation network for instance segmentation,” in Proceedings of the IEEE conference on computer vision and pattern recognition , 2018, pp. 8759–8768

    2018

  15. [21]

    ultralytics/yolov5: v1.0 - first release,

    G. Jocher, “ultralytics/yolov5: v1.0 - first release,” 2020. [Online]. Available: https://doi.org/10.5281/zenodo.3983579

    2020 doi

  16. [22]

    mixup: Beyond empirical risk minimization,

    H. Zhang, M. Cisse, Y . N. Dauphin, and D. Lopez-Paz, “mixup: Beyond empirical risk minimization,” in International Conference on Learning Representations, 2018

    2018

  17. [23]

    Yolov6: A single-stage ob- ject detection framework for industrial applications,

    Z. Li, H. Wang, X. Chen, and T. Lin, “Yolov6: A single-stage ob- ject detection framework for industrial applications,” arXiv preprint arXiv:2209.02976, 2022

    2022 arXiv

  18. [24]

    Repvgg: Making vgg-style convnets great again,

    X. Ding, X. Zhang, J. Han, and G. Ding, “Repvgg: Making vgg-style convnets great again,” Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pp. 13 733–13 742, 2021

    2021

  19. [25]

    Fcos: Fully convolutional one-stage object detection,

    Z. Tian, C. Shen, H. Chen, and T. He, “Fcos: Fully convolutional one-stage object detection,” Proceedings of the IEEE/CVF International Conference on Computer Vision , pp. 9627–9636, 2019

    2019

  20. [26]

    Yolov7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors,

    C.-Y . Wang, A. Bochkovskiy, and H.-Y . M. Liao, “Yolov7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors,” in arXiv preprint arXiv:2207.02696 , 2022

    2022 arXiv

  21. [27]

    Ultralytics yolov8,

    G. Jocher, A. Chaurasia, T. Qiu, and L. Hogan, “Ultralytics yolov8,” 2023, https://github.com/ultralytics/ultralytics

    2023

  22. [28]

    Objects as points,

    X. Zhou, D. Wang, and P. Kr ¨ahenb¨uhl, “Objects as points,” in arXiv preprint arXiv:1904.07850, 2019

    1904 arXiv

  23. [30]

    Efficientdet: Scalable and efficient object detection,

    M. Tan, R. Pang, and Q. V . Le, “Efficientdet: Scalable and efficient object detection,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2020, pp. 10 781–10 790

    2020

  24. [31]

    Ota: Optimal transport assignment for object detection,

    Z. Ge, S. Liu, F. Wang, and J. Sun, “Ota: Optimal transport assignment for object detection,” Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pp. 303–312, 2021

    2021

  25. [32]

    Yolov11: Release notes and model overview,

    Ultralytics, “Yolov11: Release notes and model overview,” https://docs. ultralytics.com/models/yolov11/, 2024, accessed: 2025-06-07

    2024

  26. [33]

    Ultralytics yolov11 models: Comparison and performance,

    ——, “Ultralytics yolov11 models: Comparison and performance,” https: //docs.ultralytics.com/models/, 2024, accessed: 2025-06-07

    2024

  27. [35]

    Domain adaptive yolo for cross-domain object detection,

    C. Chen, M. Li, L. Wang, Y . Xu, and Y . Wang, “Domain adaptive yolo for cross-domain object detection,” IEEE Transactions on Multimedia , vol. 24, pp. 2320–2331, 2022

    2022

  28. [36]

    Stac: Semi-supervised learning for object detection via strong-to-weak consistency,

    K. Sohn, M.-Y . Zhong, C.-L. Yang, C. Zhang, H. Li, Z. Wang, G. Hua, and H. Lee, “Stac: Semi-supervised learning for object detection via strong-to-weak consistency,” in Proceedings of the IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition , 2021, pp. 10 958– 10 967

    2021

  29. [37]

    Robust-yolo: Noise and occlusion aware object detection,

    Q. Zhao, X. Peng, W. Zhou, K. Wang, and Y . Tang, “Robust-yolo: Noise and occlusion aware object detection,” Neurocomputing, vol. 422, pp. 343–352, 2021

    2021

  30. [38]

    Transy- olo: A transformer-based yolo for object detection,

    B. Li, Y . Zhang, X. Han, Y . Jiang, and X. Han, “Transy- olo: A transformer-based yolo for object detection,” arXiv preprint arXiv:2205.10117, 2022

    2022 arXiv

  31. [39]

    Yolo- nas: Neural architecture search for object detection,

    A. Chakraborty, T. Wu, X. Shen, K. Yang, Y . Ma, and Y . Wang, “Yolo- nas: Neural architecture search for object detection,” arXiv preprint arXiv:2303.16199, 2023

    2023 arXiv

  32. [40]

    Yolov7-pose: En- hancing human pose estimation using yolov7 backbone,

    C.-Y . Wang, A. Bochkovskiy, and H.-Y . M. Liao, “Yolov7-pose: En- hancing human pose estimation using yolov7 backbone,” arXiv preprint arXiv:2207.09302, 2022

    2022 arXiv

Pith tools

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