Pith. sign in

REVIEW 4 major objections 6 minor 59 references

YOLOv4: A Breakthrough in Real-Time Object Detection

T0 review · 4 major / 6 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 This review paper argues that YOLOv4 combined a CSPDarknet53 backbone, PANet neck, and training-time tricks to reach 43.5% AP and 65.7% AP50 at real-time speed on COCO.

desk verdict A competent but unlabeled review of YOLOv4 that overclaims in the abstract and adds nothing new. read the letter →

arxiv 2502.04161 v1 pith:SSNPQWOS submitted 2025-02-06 cs.CV

Athulya Sundaresan Geetha This is my paper · ORCID
classification cs.CV
keywords ObjectDetectionYOLOModelsConvolutionalNeuralNetworksReal-TimeImageProcessingComputerVisionYOLOv4FrameworkCSPDarknet53BagofFreebies
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 descriptive review, not a new experiment. Its central claim is that YOLOv4's architecture—CSPDarknet53 backbone, PANet neck with a modified SPP module, and a three-scale YOLO head—together with a set of training-time and inference-time techniques, produced state-of-the-art real-time object detection on COCO. The evidence it rests on is a reproduced benchmark table: 43.5% AP and 65.7% AP50 at roughly 65 FPS on a Tesla V100. The paper also traces YOLOv1 through YOLOv11, positioning YOLOv4 as the pivotal model in that lineage. A fair reader should treat the contribution as synthesis and explanation of the original YOLOv4 results, with the benchmark numbers inherited from the 2020 paper.

What carries the argument

The load-bearing object is the YOLOv4 pipeline as a three-part architecture: the CSPDarknet53 backbone extracts multi-scale features; the PANet neck, augmented with a modified Spatial Pyramid Pooling block, fuses high-level semantics with low-level spatial detail; and a three-scale YOLO head predicts boxes and classes. Around that pipeline, the paper groups techniques into a 'bag of freebies'—Mosaic and CutMix augmentation, DropBlock, label smoothing, CIoU loss, Self-Adversarial Training, and Cross mini-Batch Normalization—which raise accuracy without affecting inference cost, and a 'bag of specials'—SPP, attention modules, and improved activations—which cost a little inference time for larger accuracy gains. The reproduced benchmark table carries the empirical weight of the argument.

What would settle it

Reproduce Table 2 under controlled conditions: run YOLOv4 CSPDarknet-53 at 608x608 input on a Tesla V100 with a fixed batch size and single precision, and check whether AP, AP50, and FPS match 43.5%, 65.7%, and roughly 62 FPS. If they do not, the paper's central numeric claim fails as stated; benchmarking a later YOLO version on the same split would likewise settle whether YOLOv4 remains the best detector the paper calls it.

Watch

Extended reading notes

Core claim

The discovery, as the paper states it, is that YOLOv4 achieved the best COCO performance among real-time detectors at the time by combining a CSPDarknet53 backbone with Cross-Stage-Partial connections, a PANet neck with Spatial Pyramid Pooling, a three-scale YOLO head, and a set of enhancements: Cross mini-Batch Normalization, Self-Adversarial Training, Weighted-Residual-Connections, CIoU loss, Mosaic augmentation, and DropBlock regularization. The paper reports that this combination reached 43.5% AP and 65.7% AP50 at 62–65 FPS on Volta GPUs. This is presented as a milestone that made high-accuracy detection fast and affordable enough for real-world deployment, before later YOLO versions pushed accuracy higher.

Load-bearing premise

The benchmark numbers are taken from the original YOLOv4 paper and are assumed to be accurate and directly comparable across GPU generations without controlling for batch size, precision, or inference framework, and the paper treats 'best on COCO' as a lasting status even though it describes later versions with higher accuracy.

Editorial extensions

If this is right

  • If the reported design is what produced the gains, the freebies/specials split offers a transferable recipe: improve a detector's training without slowing inference, then add small-cost modules only where they pay for themselves.
  • The table's speed-accuracy trade-off implies input resolution is a practical dial: on Volta GPUs, 416-pixel input runs at about 96 FPS with 41.2% AP, while 608-pixel input runs at about 62 FPS with 43.5% AP.
  • The paper's lineage account implies that later YOLO versions inherited YOLOv4's CSP, PAN, and SPP choices, making YOLOv4's architecture the common ancestor of the family's subsequent improvements.
  • If efficiency and affordability are as claimed, YOLOv4-class detectors are viable on single GPUs for real-time applications such as surveillance and autonomous navigation, which the paper explicitly names.

Reading between the lines

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

  • Beyond the paper's claims, the identical AP values across Maxwell, Pascal, and Volta GPUs in its own table suggest the accuracy numbers are determined by the model and input size, not the GPU; only FPS is hardware-sensitive, so cross-GPU comparisons should control for batch size, precision, and framework before being read as 'best'.
  • The paper's framing implies a speed-accuracy frontier; a direct test would be to benchmark YOLOv4 against YOLOv8 or YOLOv10 on the same COCO split and GPU, measuring how much of the frontier YOLOv4 actually defined.
  • The freebies/specials taxonomy could be applied as an audit tool to newer detectors: classify each component by whether it changes inference cost, then predict which components contributed most to a reported gain before re-running training.
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 / 6 minor

Summary. The manuscript is a descriptive overview of YOLOv4, its architectural components, and its reported COCO performance. It recapitulates the architecture (CSPDarknet53 backbone, PANet neck, SPP, head), lists Bag-of-Freebies and Bag-of-Specials techniques, surveys YOLOv1 through YOLOv11, and reproduces a performance table attributed to Bochkovskiy et al. [23]. The abstract claims that YOLOv4 achieved the best performance on the COCO dataset at 43.5% AP and about 65 FPS on a Tesla V100, and the introduction frames the paper as an in-depth exploration of YOLOv4.

Significance. If the manuscript is read as a tutorial, it offers a compact summary of YOLOv4's ideas and correctly reproduces the headline COCO numbers from the original source. However, the paper contains no new experiments, no code, no derivations, and no quantitative analysis beyond quoting existing results. The unqualified 'best performance on COCO' claim is historically inaccurate without a temporal qualifier and is contradicted by the paper's own description of later YOLO versions with higher AP. The paper's potential value is therefore pedagogical rather than research-oriented, and even that value is undermined by several internal inconsistencies and uncited quantitative claims.

major comments (4)
  1. [Abstract and §4.4, Table 2] The abstract's claim that 'YOLOv4 achieved the best performance on the COCO dataset' is unqualified and is internally contradicted by §2.8–§2.13, which report later YOLO versions with higher AP, e.g., YOLOv6-L at 52.3% AP. The claim needs a temporal and scope qualifier, such as 'best among real-time detectors at the time of its 2020 release.' In addition, the abstract states '~65 frames per second' on a Tesla V100, while Table 2 reports 62 FPS for the V100 at 608 resolution; this numerical discrepancy must be reconciled.
  2. [§4, Figure 1] The architecture described in Section 4 is not the original YOLOv4. Figure 1 is taken from Scaled-YOLOv4 [51] and depicts YOLOv4-P5/P6/P7 with CSPDown layers and detection heads from P3 to P7, whereas the original YOLOv4, which is the subject of the paper and of Table 1, uses a single multi-scale head at 19×19, 38×38, and 76×76. Conflating Scaled-YOLOv4 with YOLOv4 materially misrepresents the model the paper claims to analyze.
  3. [§2.4, YOLOv4] The statement that YOLOv4 delivered 'a 10% increase in detection accuracy and a 12% boost in processing speed' lacks both a defined baseline and a citation. If the comparison is to YOLOv3, that should be stated explicitly and supported by [23] or another source; if it is to a different model, the comparison must be specified. The same paragraph's description of CSPDarknet53 as having '29 convolutional layers' is not obviously consistent with the '53' in the architecture's name and needs a source or correction.
  4. [Title, Abstract, and §1 (scope)] The paper presents no original experiments, derivations, or analysis; its central numerical claims are restatements of [23]. The title and abstract read as if the paper itself established these results. If the manuscript is intended as a review, it should say so explicitly in the title and abstract and cite the primary source for each quoted result. If it is intended as original research, it does not currently meet that bar, and the framing should be revised accordingly.
minor comments (6)
  1. [Table 2 header] The header row contains a typo: 'V olta' should be 'Volta'.
  2. [Abstract] The term 'Cross mini-Batch Normalization' is normally abbreviated CmBN; introducing the abbreviation would improve precision and consistency with the literature.
  3. [References [27, 28]] Reference [28] is a paper on Bangla handwritten character recognition and does not support the claims about YOLOv9 and YOLOv10; the actual YOLOv9 and YOLOv10 publications should be cited.
  4. [§2.5, YOLOR] The subsection describes YOLOR as setting a new standard but gives no quantitative performance numbers; adding at least one concrete comparison would make the claim verifiable.
  5. [Throughout] The spelling of the framework name is inconsistent ('DarkNet' vs. 'Darknet'); one convention should be used throughout.
  6. [Table 2] The annotations '(M)', '(P)', and '(V)' next to the FPS values are not defined in the text or table caption and should be explained.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the paper is a literature-review-style restatement of external YOLOv4 benchmark results, not a derivation from fitted parameters or self-referential claims.

full rationale

The paper's central numerical claims — 43.5% AP, 65.7% AP50, the FPS values, and the full performance table — are explicitly attributed to the external YOLOv4 paper by Bochkovskiy et al. [23]; Table 2 is labelled "From [23]". There is no equation fitted to the target result, no parameter inferred from the COCO numbers and then renamed as a prediction, and no uniqueness theorem or definitional equivalence that forces the reported benchmarks. The author's self-citations in Sections 2.7–2.10 support adjacent descriptive statements about YOLOv5, YOLOv6, YOLOv7, and YOLOv8/YOLOv10, but those claims are not load-bearing for the paper's central YOLOv4 performance figures, which come from an independent external source. The abstract's unqualified "best performance on the COCO dataset" is internally inconsistent with the paper's own later descriptions of higher-AP YOLO versions (e.g., YOLOv6-L at 52.3% AP in Section 2.8), and the abstract's "~65 frames per second" differs from Table 2's 62 FPS on V100 at 608; however, both are correctness or scoping concerns, not circularity, since neither is produced by a self-referential construction from the paper's own inputs. The derivation chain is effectively a restatement of externally reported, independently reproducible benchmark numbers, so no significant circularity is present.

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

This is a survey; it introduces no free parameters and no invented entities. The load-bearing assumptions are that the cited benchmark results and architecture descriptions are correct, and that later YOLO versions were not relevant to the 'best performance' claim at the time of writing.

assumptions (3)
  • domain assumption The COCO performance numbers in Table 2, taken from Bochkovskiy et al. 2020, are accurate.
    The paper's entire central claim about YOLOv4's performance rests on trusting the original benchmark results; no independent verification or error analysis is provided.
  • domain assumption MS COCO is a representative benchmark for object detection and the AP/AP50 metrics are meaningful for comparing detectors.
    The paper uses COCO AP as the measure of 'best performance' without justifying why this benchmark and metric are the appropriate standard.
  • domain assumption The architecture descriptions in Section 4 accurately reflect YOLOv4.
    The paper paraphrases CSPDarknet53, PANet, SPP, and SAM from the original papers and from secondary blog sources without reimplementing the model. This assumption is weakened by the inclusion of Scaled-YOLOv4 material in Figure 1.

how reviews work

0 comments
Cite this review

Pith. "Pith review of YOLOv4: A Breakthrough in Real-Time Object Detection." pith.science (2026). https://pith.science/paper/SSNPQWOS

@misc{pith2026250204161,
  author       = {Pith},
  title        = {Pith review of: YOLOv4: A Breakthrough in Real-Time Object Detection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SSNPQWOS}},
  note         = {Machine review of arXiv:2502.04161}
}
read the original abstract

YOLOv4 achieved the best performance on the COCO dataset by combining advanced techniques for regression (bounding box positioning) and classification (object class identification) using the Darknet framework. To enhance accuracy and adaptability, it employs Cross mini-Batch Normalization, Cross-Stage-Partial-connections, Self-Adversarial-Training, and Weighted-Residual-Connections, as well as CIoU loss, Mosaic data augmentation, and DropBlock regularization. With Mosaic augmentation and multi-resolution training, YOLOv4 achieves superior detection in diverse scenarios, attaining 43.5\% AP (in contrast, 65.7\% AP50) on a Tesla V100 at ~65 frames per second, ensuring efficiency, affordability, and adaptability for real-world environments.

Figures

Figures reproduced from arXiv: 2502.04161 by the authors.

Figure 1
Figure 1. illustrates the YOLOv4-P5, YOLOv4-P6, and YOLOv4-P7 architectures, detailing the feature extraction, integration, and detection processes. At its core, the CSPDarkNet backbone (green blocks) extracts multi-scale features through hierarchical layers, starting from 1xCSPDark (64 channels) up to 7xCSPDark (1024 channels). Each CSP block (shown separately in the bottom-left corner) consists of multiple convolutional lay… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

59 extracted references · 48 canonical work pages

  1. [23]

    Yolov4: Optimal speed and accuracy of object detection

    Alexey Bochkovskiy, Chien-Yao Wang, and Hong-Yuan Mark Liao. Yolov4: Optimal speed and accuracy of object detection. arXiv preprint arXiv:2004.10934, 2020

    arXiv 2004

  2. [51]

    Scaled-yolov4: Scaling cross stage partial network

    Chien-Yao Wang, Alexey Bochkovskiy, and Hong-Yuan Mark Liao. Scaled-yolov4: Scaling cross stage partial network. In Proceedings of the IEEE/cvf conference on computer vision and pattern recognition, pages 13029– 13038, 2021

    work page 2021

  3. [1]

    A layer-wise surface deformation defect detection by convolutional neural networks in laser powder-bed fusion images

    Muhammad Ayub Ansari, Andrew Crampton, and Simon Parkinson. A layer-wise surface deformation defect detection by convolutional neural networks in laser powder-bed fusion images. Materials, 15(20):7166, 2022

    work page 2022

  4. [2]

    Custom lightweight convolutional neural network architecture for automated detection of damaged pallet racking in warehousing & distribution centers

    Muhammad Hussain and Richard Hill. Custom lightweight convolutional neural network architecture for automated detection of damaged pallet racking in warehousing & distribution centers. IEEE Access, 11:58879– 58889, 2023

    work page 2023

  5. [3]

    Lightweight convolutional network for automated photovoltaic defect detection

    Arsalan Zahid, Muhammad Hussain, Richard Hill, and Hussain Al-Aqrabi. Lightweight convolutional network for automated photovoltaic defect detection. In 2023 9th International Conference on Information Technology Trends (ITT), pages 133–138. IEEE, 2023

    work page 2023

  6. [4]

    Comparative study of computational time that hog-based features used for vehicle detection

    Natthariya Laopracha and Khamron Sunat. Comparative study of computational time that hog-based features used for vehicle detection. In Recent Advances in Information and Communication Technology 2017: Proceedings of the 13th International Conference on Computing and Information Technology (IC2IT), pages 275–284. Springer, 2018. 9 A.S.G EETHA .: YOLO V4: A...

    work page 2017

  7. [5]

    On combining classifiers

    J Kittler, M Hatef, Robert PW Duin, and J Matas. On combining classifiers. ieee transaction on pattern analysis and machine intelligence, 1998

    work page 1998

  8. [6]

    Action recognition by dense trajectories

    H Wang, A Klaser, C Schmid, and CL Liu. Action recognition by dense trajectories. cvpr’11, washington, dc, usa. IEEE Computer Society, pages 3169–3176, 2011

    work page 2011

Show all 59 references
  1. [7]

    A database for fine grained activity detection of cooking activities

    Marcus Rohrbach, Sikandar Amin, Mykhaylo Andriluka, and Bernt Schiele. A database for fine grained activity detection of cooking activities. In 2012 IEEE conference on computer vision and pattern recognition , pages 1194–1201. IEEE, 2012

    2012

  2. [8]

    Feature mapping for rice leaf defect detection based on a custom convolutional architecture

    Muhammad Hussain, Hussain Al-Aqrabi, Muhammad Munawar, and Richard Hill. Feature mapping for rice leaf defect detection based on a custom convolutional architecture. Foods, 11(23):3914, 2022

    2022

  3. [9]

    Gun and knife detection based on faster r-cnn for video surveillance

    M Milagro Fernandez-Carrobles, Oscar Deniz, and Fernando Maroto. Gun and knife detection based on faster r-cnn for video surveillance. In Iberian conference on pattern recognition and image analysis, pages 441–452. Springer, 2019

    2019

  4. [10]

    Imagenet classification with deep convolutional neural networks

    Alex Krizhevsky, Ilya Sutskever, and Geoffrey E Hinton. Imagenet classification with deep convolutional neural networks. Communications of the ACM, 60(6):84–90, 2017

    2017

  5. [11]

    Stable and compact design of memristive googlenet neural network

    Huanhuan Ran, Shiping Wen, Kaibo Shi, and Tingwen Huang. Stable and compact design of memristive googlenet neural network. Neurocomputing, 441:52–63, 2021

    2021

  6. [12]

    Brain tumor detection using mask r-cnn

    Mounika Gajja. Brain tumor detection using mask r-cnn. J. Adv. Res. Dyn. Control Syst, 12:101–108, 2020

    2020

  7. [13]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 770–778, 2016

    2016

  8. [14]

    Pedestrian detection based on faster r-cnn

    Shuang Liu, Xing Cui, Jiayi Li, Hui Yang, and Niko Lukaˇc. Pedestrian detection based on faster r-cnn. Interna- tional Journal of Performability Engineering, 15(7):1792, 2019

    2019

  9. [15]

    Rich feature hierarchies for accurate object detection and semantic segmentation

    Ross Girshick, Jeff Donahue, Trevor Darrell, and Jitendra Malik. Rich feature hierarchies for accurate object detection and semantic segmentation. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 580–587, 2014

    2014

  10. [16]

    Classification of picture art style based on vggnet

    Zihan Yang. Classification of picture art style based on vggnet. In Journal of Physics: Conference Series, volume 1774, page 012043. IOP Publishing, 2021

    2021

  11. [17]

    A dynamic multi-mobile agent itinerary planning approach in wireless sensor networks via intuitionistic fuzzy set

    Tariq Alsboui, Richard Hill, Hussain Al-Aqrabi, Hafiz Muhammad Athar Farid, Muhammad Riaz, Shamaila Iram, Hafiz Muhammad Shakeel, and Muhammad Hussain. A dynamic multi-mobile agent itinerary planning approach in wireless sensor networks via intuitionistic fuzzy set. Sensors, 2...

    2022

  12. [18]

    Statistical analysis and development of an ensemble- based machine learning model for photovoltaic fault detection

    Muhammad Hussain, Hussain Al-Aqrabi, and Richard Hill. Statistical analysis and development of an ensemble- based machine learning model for photovoltaic fault detection. Energies, 15(15):5492, 2022

    2022

  13. [19]

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

    Joseph Redmon, Santosh Divvala, Ross Girshick, and Ali Farhadi. You only look once: Unified, real-time object detection. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 779–788, 2016

    2016

  14. [20]

    A gradient guided architecture coupled with filter fused representations for micro-crack detection in photovoltaic cell surfaces

    Muhammad Hussain, Tianhua Chen, Sofya Titrenko, Pan Su, and Mufti Mahmud. A gradient guided architecture coupled with filter fused representations for micro-crack detection in photovoltaic cell surfaces. IEEE Access, 10:58950–58964, 2022

    2022

  15. [21]

    Yolo9000: better, faster, stronger

    Joseph Redmon and Ali Farhadi. Yolo9000: better, faster, stronger. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 7263–7271, 2017

    2017

  16. [22]

    Yolov3: An incremental improvement

    Joseph Redmon and Ali Farhadi. Yolov3: An incremental improvement. arXiv preprint arXiv:1804.02767, 2018

    2018 arXiv

  17. [24]

    Comprehensive guide to ultralytics yolov5, 2023

    Lakshantha Dissanayake, Glenn Jocher, Q Burhan, and Sergiu Waxmann. Comprehensive guide to ultralytics yolov5, 2023

    2023

  18. [25]

    Yolov6: A single-stage object detection framework for industrial applications

    Chuyi Li, Lulu Li, Hongliang Jiang, Kaiheng Weng, Yifei Geng, Liang Li, Zaidan Ke, Qingyuan Li, Meng Cheng, Weiqiang Nie, et al. Yolov6: A single-stage object detection framework for industrial applications. arXiv preprint arXiv:2209.02976, 2022

    2022 arXiv

  19. [26]

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

    Chien-Yao Wang, Alexey Bochkovskiy, and Hong-Yuan Mark Liao. Yolov7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 7464–7475, 2023

    2023

  20. [27]

    Yolo: A brief history, 2023

    Glenn Jocher, Muhammad Rizwan Munawar, and Ayush Chaurasia. Yolo: A brief history, 2023

    2023

  21. [28]

    Isolated bangla handwritten character recognition with convolutional neural network

    Mujadded Al Rabbani Alif, Sabbir Ahmed, and Muhammad Abul Hasan. Isolated bangla handwritten character recognition with convolutional neural network. In2017 20th International conference of computer and information technology (ICCIT), pages 1–6. IEEE, 2017

    2017

  22. [29]

    Yolo object detection explained, 2024

    DataCamp. Yolo object detection explained, 2024. Accessed: 2024-12-13. 10 A.S.G EETHA .: YOLO V4: A B REAKTHROUGH IN REAL -T IME OBJECT DETECTION - FEBRUARY 7, 2025

    2024

  23. [30]

    Using deep convolutional neural network architectures for object classification and detection within x-ray baggage security imagery

    Samet Akcay, Mikolaj E Kundegorski, Chris G Willcocks, and Toby P Breckon. Using deep convolutional neural network architectures for object classification and detection within x-ray baggage security imagery. IEEE transactions on information forensics and security, 13(9):2203–2...

    2018

  24. [31]

    A frame-work assisting the visually impaired people: common object detection and pose estimation in surrounding environment

    Van-Hung Le, Hai Vu, and Thuy Thi Nguyen. A frame-work assisting the visually impaired people: common object detection and pose estimation in surrounding environment. In 2018 5th NAFOSTED conference on information and computer science (NICS), pages 216–221. IEEE, 2018

    2018

  25. [32]

    An evaluation of yolo-based algorithms for hand detection in the kitchen

    Joshua van Staden and Dane Brown. An evaluation of yolo-based algorithms for hand detection in the kitchen. In 2021 International Conference on Artificial Intelligence, Big Data, Computing and Data Communication Systems (icABCD), pages 1–7. IEEE, 2021

    2021

  26. [33]

    Improved yolov4 algorithm for safety management of on-site power system work

    Qiang Li, Feng Zhao, Zhongping Xu, Kexin Li, Jing Wang, Haofeng Liu, Liang Qin, and Kaipei Liu. Improved yolov4 algorithm for safety management of on-site power system work. Energy reports, 8:739–746, 2022

    2022

  27. [34]

    You only learn one representation: Unified network for multiple tasks

    CY Wang, IH Yeh, and HYM Liao. You only learn one representation: Unified network for multiple tasks. arxiv

  28. [35]

    Complete and accurate holly fruits counting using yolox object detection

    Yanchao Zhang, Wenbo Zhang, Jiya Yu, Leiying He, Jianneng Chen, and Yong He. Complete and accurate holly fruits counting using yolox object detection. Computers and Electronics in Agriculture, 198:107062, 2022

    2022

  29. [36]

    Real-time multiple object tracking for safe cooking activities

    Hubert Ngankam, Philippe Dion, Hélène Pigot, and Sylvain Giroux. Real-time multiple object tracking for safe cooking activities. In International Conference on Smart Homes and Health Telematics, pages 192–204. Springer, 2023

    2023

  30. [37]

    Augmented reality based interactive cooking guide

    Isaias Majil, Mau-Tsuen Yang, and Sophia Yang. Augmented reality based interactive cooking guide. Sensors, 22(21):8290, 2022

    2022

  31. [38]

    A comparative analysis of yolov5, yolov8, and yolov10 in kitchen safety

    Athulya Sundaresan Geetha and Muhammad Hussain. A comparative analysis of yolov5, yolov8, and yolov10 in kitchen safety. arXiv preprint arXiv:2407.20872, 2024

    2024 arXiv

  32. [39]

    Comparing yolov5 variants for vehicle detection: A performance analysis

    Athulya Sundaresan Geetha. Comparing yolov5 variants for vehicle detection: A performance analysis. arXiv preprint arXiv:2408.12550, 2024

    2024 arXiv

  33. [40]

    A yolov6-based improved fire detection approach for smart city environments

    Saydirasulov Norkobil Saydirasulovich, Akmalbek Abdusalomov, Muhammad Kafeel Jamil, Rashid Nasimov, Dinara Kozhamzharova, and Young-Im Cho. A yolov6-based improved fire detection approach for smart city environments. Sensors, 23(6):3161, 2023

    2023

  34. [41]

    What is yolov6? a deep insight into the object detection model

    Athulya Sundaresan Geetha. What is yolov6? a deep insight into the object detection model. arXiv preprint arXiv:2412.13006, 2024

    2024 arXiv

  35. [42]

    Detection of guns and knives images based on yolo v7

    Yi Yi Aung and Kyi Zar Oo. Detection of guns and knives images based on yolo v7. In 2024 3rd International Conference on Artificial Intelligence For Internet of Things (AIIoT), pages 1–6. IEEE, 2024

    2024

  36. [43]

    Performance of yolov7 in kitchen safety while handling knife

    Athulya Sundaresan Geetha. Performance of yolov7 in kitchen safety while handling knife. arXiv e-prints, pages arXiv–2501, 2025

    2025

  37. [44]

    Real time object detection with data variation

    Maha Mokrani and Zied Hajaiej. Real time object detection with data variation. Przeglad Elektrotechniczny, 2024(5), 2024

    2024

  38. [45]

    Comparative analysis of yolov8 and yolov10 in vehicle detection: Performance metrics and model efficacy.Vehicles, 6(3):1364– 1382, 2024

    Athulya Sundaresan Geetha, Mujadded Al Rabbani Alif, Muhammad Hussain, and Paul Allen. Comparative analysis of yolov8 and yolov10 in vehicle detection: Performance metrics and model efficacy.Vehicles, 6(3):1364– 1382, 2024

    2024

  39. [46]

    A comprehensive review of yolo architectures in computer vision: From yolov1 to yolov8 and yolo-nas

    Juan Terven, Diana-Margarita Córdova-Esparza, and Julio-Alejandro Romero-González. A comprehensive review of yolo architectures in computer vision: From yolov1 to yolov8 and yolo-nas. Machine Learning and Knowledge Extraction, 5(4):1680–1716, 2023

    2023

  40. [47]

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

    Chien-Yao Wang, I-Hau Yeh, and Hong-Yuan Mark Liao. Yolov9: Learning what you want to learn using programmable gradient information. In European Conference on Computer Vision, pages 1–21. Springer, 2025

    2025

  41. [48]

    Yolov1 to yolov10: A comprehensive review of yolo variants and their application in the agricultural domain

    Mujadded Al Rabbani Alif and Muhammad Hussain. Yolov1 to yolov10: A comprehensive review of yolo variants and their application in the agricultural domain. arXiv preprint arXiv:2406.10139, 2024

    2024 arXiv

  42. [49]

    Nikhileswara Rao

    S. Nikhileswara Rao. Yolov11 architecture explained: Next-level object detection with enhanced speed and accuracy, 2024

    2024

  43. [50]

    Yolov4: A fast and efficient object detection model, 2024

    Nico Klingler. Yolov4: A fast and efficient object detection model, 2024

    2024

  44. [52]

    What is yolov4? a detailed breakdown, 2024

    Jacob Solawetz. What is yolov4? a detailed breakdown, 2024

    2024

  45. [53]

    Yolov4: High-speed and precise object detection, 2024

    Glenn Jocher, Matthew Noyce, Muhammad Rizwan Munawar, and Sergiu Waxmann. Yolov4: High-speed and precise object detection, 2024

    2024

  46. [54]

    Accelerating object detection with yolov4 for real-time applications

    K Senthil Kumar and KMB Safwan. Accelerating object detection with yolov4 for real-time applications. arXiv preprint arXiv:2410.16320, 2024. 11 A.S.G EETHA .: YOLO V4: A B REAKTHROUGH IN REAL -T IME OBJECT DETECTION - FEBRUARY 7, 2025

    2024 arXiv

  47. [55]

    Yolov4 and darknet for pothole detection, 2022

    Sovit Rath. Yolov4 and darknet for pothole detection, 2022

    2022

  48. [56]

    Deployment of ai-based rbf network for photovoltaics fault detection procedure

    Muhammad Hussain, Mahmoud Dhimish, Violeta Holmes, and Peter Mather. Deployment of ai-based rbf network for photovoltaics fault detection procedure. AIMS Electronics and Electrical Engineering, 4(1):1–18, 2019

    2019

  49. [57]

    Domain modelling for a lightweight convolutional network focused on automated exudate detection in retinal fundus images

    Burcu Ataer Aydin, Muhammad Hussain, Richard Hill, and Hussain Al-Aqrabi. Domain modelling for a lightweight convolutional network focused on automated exudate detection in retinal fundus images. In 2023 9th International Conference on Information Technology Trends (ITT), page...

    2023

  50. [58]

    Child emotion recognition via custom lightweight cnn architecture

    Muhammad Hussain and Hussain Al-Aqrabi. Child emotion recognition via custom lightweight cnn architecture. In Kids Cybersecurity Using Computational Intelligence Techniques, pages 165–174. Springer, 2023. 12

    2023

  51. [2021]

    arXiv preprint arXiv:2105.04206, 2021

    2021 arXiv

Pith tools

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