Pith. sign in

REVIEW 4 major objections 7 minor 37 references

Bearded Dragon Activity Recognition Pipeline: An AI-Based Approach to Behavioural Monitoring

T0 review · 4 major / 7 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read The paper shows a YOLO-based pipeline can automatically detect bearded dragon basking, while hunting detection fails on small crickets.

desk verdict A credible object-detection benchmark with a public dataset, but the behavior-recognition claim is only half-supported; hunting is effectively undemonstrated. read the letter →

arxiv 2507.17987 v1 pith:BMLBN7TS submitted 2025-07-23 cs.CV

Arsen Yermukan , Pedro Machado , Feliciano Domingos , Isibor Kennedy Ihianle , Jordan J. Bird , Stefano S. K. Kaburu , Samantha J. Ward This is my paper · ORCID
classification cs.CV
keywords beardeddragonYOLOobjectdetectionactivityrecognitionbaskingbehaviourhuntingreptilemonitoringsmall
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 proposes an end-to-end video pipeline that automates behaviour monitoring for bearded dragons: YOLO object detection locates dragons, heating lamps, and crickets in each frame, temporal interpolation fills short detection gaps, and geometric rules turn those detections into basking and hunting labels. The authors train five YOLO variants (v5, v7, v8, v11, v12) on a custom 1,200-image dataset and select YOLOv8s as the best accuracy–speed compromise, with mAP@0.5:0.95 = 0.855. On video, the basking rule is reported reliable, producing coverage between 16.5% and 100% of frames in basking clips, while hunting is detected in only 0.6% of frames because the cricket detector is weak (mAP@0.5 = 0.392). The paper's claim is that this pipeline is a scalable alternative to manual observation for welfare-relevant behaviours, with hunting detection as a known bottleneck rather than a solved problem.

What carries the argument

The carrying mechanism is a two-stage pipeline. Stage one is a set of YOLO object detectors that output per-frame bounding boxes for bearded dragons, heating lamps, and crickets; the authors compare v5, v7, v8, v11, and v12 and pick YOLOv8s for animals and lamps and YOLOv8n for crickets. Stage two is rule-based spatial logic: basking is declared when the vertical distance $\Delta y$ and off-axis angle $\theta$ between dragon and lamp satisfy $\Delta y \le \beta H$ and $\theta < \theta_{\max}$, and hunting is declared when a cricket's centre disappears from view within Euclidean distance $d < \gamma W$ of a dragon's centre, with $H$ and $W$ the frame height and width. A two-way nearest-frame interpolation bridges short detection gaps and is credited with a 30% boost in continuity.

What would settle it

Take the released 1,200-image dataset, add 30 to 60 minutes of enclosure video with independent human coding of basking and hunting frame by frame, then run the pipeline and compute per-frame agreement between the rule labels and the human labels. If the basking rule's agreement is around chance on videos where a dragon sits near a lamp but does not bask, or if the hunting rule fires on occlusions and exits rather than prey capture, the behaviour-recognition claim is falsified even though detection metrics stay high.

Watch

Extended reading notes

Core claim

On the authors' own terms, the central discovery is that a bounding-box geometry rule can turn YOLO detections into a usable automatic basking log: if a dragon's centre is within $\beta H$ pixels vertically and within angle $\theta_{\max}$ of a detected heating lamp, the frame is labelled basking, and this label matches observed basking episodes across tested clips. The same approach fails for hunting: the rule flags a hunt when a cricket disappears within distance $\gamma W$ of a dragon, but because cricket detection recall is only 0.392, hunting episodes are caught in 0.6% of frames, usually as isolated single-frame events. Across the five tested YOLO variants, YOLOv8s offers the best overall detection balance, with the highest mAP@0.5 and minimal cross-class confusion between dragons and lamps.

Load-bearing premise

The behaviour labels are produced by geometric proxies -- a dragon near and below a lamp is basking, a cricket vanishing near a dragon is hunting -- and neither the threshold values nor the agreement with human-observed behaviour is ever tested, so the behavioural claim rests on these proxies being correct.

Editorial extensions

If this is right

  • If the pipeline works as reported, researchers can obtain automated per-frame basking logs for bearded dragons in controlled enclosures, replacing hand-scored observation for thermoregulatory behaviour.
  • The detection comparison implies YOLOv8s is a suitable default detector for dragon and lamp localisation, with mAP@0.5:0.95 around 0.855 and low lamp-to-dragon confusion.
  • The hunting results imply that no reliable automated hunting label can be produced until cricket detection improves; the authors' own threshold for success is pushing detection above 95% across classes.
  • The 92% coverage figure in basking and idle clips implies the pipeline can distinguish active basking from idle behaviour with few false positives when both objects are visible.

Reading between the lines

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

  • The geometric rule thresholds $\beta$, $\theta_{\max}$, and $\gamma$ are never reported and the rule outputs are never compared with human-annotated behaviour labels, so the behaviour-classification half of the pipeline is unvalidated even if the object detectors are sound.
  • Because hunting is scored by cricket disappearance, a cricket leaving the frame or being occluded would be counted as a hunt; the paper does not disentangle these cases from actual prey capture.
  • A direct extension would be to treat the distance and angle time-series as features for a temporal classifier rather than fixed thresholds, which could separate basking from other lamp-proximity behaviour without requiring new detection infrastructure.
  • The same architecture likely transfers to other reptile species kept in similarly controlled enclosures, but the scarce class of small prey will need dedicated small-object detection heads or higher-resolution inputs before hunting can be monitored.
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 / 7 minor

Summary. This paper describes an end-to-end pipeline for monitoring bearded dragon behaviour from video: five YOLO variants are trained on a custom three-class dataset (dragon, lamp, cricket), the best variant is selected, per-frame detections are temporally interpolated, and rule-based geometric conditions (Eqs. 6-8) classify basking and hunting. Detection performance is reported as precision, recall, mAP@0.5, and mAP@0.5:0.95. Activity evaluation in Section 4.2 uses coverage, mean vertical separation, jitter, and drift. The paper reports reliable basking detection and limited hunting detection due to poor cricket detection.

Significance. The dataset is public and the YOLO comparison is concrete and independently evaluable, which are real strengths. If the behavioural classification were validated, the pipeline would be a useful tool for reptile welfare monitoring. However, the evidence supports only object detection and a basking heuristic; the hunting half of the central claim is undemonstrated, and the activity evaluation lacks human-labelled behavioural ground truth. The abstract overstates the headline result, so the paper as written does not yet deliver on its central claim.

major comments (4)
  1. [Abstract; Table 1] The abstract cites YOLOv8s accuracy as mAP@0.5:0.95 = 0.855, but Table 1 reports YOLOv8s All Classes mAP@0.5 = 0.855 and mAP@0.5:0.95 = 0.522. The abstract therefore conflates the two metrics, and the selection claim of a 'superior balance of accuracy and speed' needs to be based on the correct figures and on measured inference speed, which is not reported.
  2. [§3, Eqs. (7)-(8)] The thresholds beta, theta_max, and gamma in Eqs. (7) and (8) are never reported, and no sensitivity analysis is given. Without these values the behaviour rules cannot be reproduced, and because coverage is computed from these same rules, the thresholds are free parameters that can be tuned to make the rules fire. Please report the exact values, how they were chosen, and how the reported coverage changes as they vary.
  3. [§4.2, Table 4] The activity evaluation reports only coverage, mean vertical difference, jitter, and drift. None of these measures agreement with human-annotated behaviour labels, and no such labels are used anywhere in the paper. The claim that basking detection is 'reliable' therefore does not follow from Table 4; at most it shows that the heuristic fires consistently in some clips. An evaluation with human-labelled frames is needed to support the behaviour-recognition claim.
  4. [§4.2; Table 3] The hunting claim is contradicted by the paper's own results: cricket mAP@0.5 is 0.392, hunting coverage is 0.60% and often consists of single-frame detections, jitter and drift are not applicable, and the text concedes that hunting detection 'was severely hampered' and that false negatives disrupt classification. The abstract's statement that the system identifies hunting as one of two key behaviours is therefore unsupported.
minor comments (7)
  1. [§3; §4] Section 3 states that each model was trained for 200 epochs per class, while Section 4 says training ran over 400 epochs with early stopping; these statements should be reconciled.
  2. [§3] The temporal interpolation is said to boost continuity by 30%, but no comparison or measurement is provided to support this figure.
  3. [§4] The text refers to normalised confusion matrices, but they are not shown; please add the figures or remove the references.
  4. [Table 2; §4] YOLOv8s is chosen as the optimal model, but Table 2 shows that YOLOv7s has a higher maximum F1 score (0.81 versus 0.76); since inference speed is not measured, the selection criterion should be stated explicitly.
  5. [Abstract; §4] The abstract says 'real-time video analysis', but no frame-rate or latency measurements are reported; please qualify this claim.
  6. [§5] The conclusion states that 'other components achieved 92% coverage in basking and idle conditions', but Table 4 reports ranges rather than a single 92% figure, so this value is not traceable.
  7. [Table 3] YOLOv12n is described as 'fused' in Table 3, but the fusion method is not defined anywhere in the text.

Circularity Check

1 steps flagged · score 6.0 of 10

Behavior-rule evaluation is self-referential: basking reliability is measured by coverage of the very rule that defines basking, with no behavioral ground truth.

  1. self definitional [Section 3 Eqs. (6)-(7) define basking; Section 4.2 Table 4 evaluates it via coverage]
    "We apply the following thresholds for basking detection: ∆y≤βH, θ < θ max (7)... Four quantitative metrics were derived: coverage, the percentage of frames in which the behaviour was detected; ... Basking detection achieved coverage from 16.50 to 100.00 percent ... These findings indicate reliable identification of basking."

    Basking is defined by Eq. (7): the rule fires when the dragon's vertical distance and angle relative to the lamp satisfy thresholds. Section 4.2 evaluates this with 'coverage, the percentage of frames in which the behaviour was detected,' and then concludes 'These findings indicate reliable identification of basking.' But 'the behaviour was detected' is just Eq. (7) firing; coverage is a count of the rule's own output, not a comparison with human-annotated basking labels. Table 4 reports only coverage, mean vertical difference, jitter, and drift; no behavioral ground truth is used. The thresholds beta, theta_max, and gamma are never reported, so the rule cannot be externally checked.

full rationale

The object-detection component is evaluated independently on a labeled image dataset with standard metrics (precision, recall, mAP@0.5, mAP@0.5:0.95) and is not circular. The circularity is confined to the behavioral classification stage. Basking and hunting are defined by hand-written geometric rules (Eqs. 6-8), and the activity evaluation (Section 4.2, Table 4) uses 'coverage' — the percentage of frames in which the rule fires — as evidence that 'basking detection proved reliable.' Since no human-annotated behavior labels are used anywhere, coverage is a restatement of the rule's own output, not an independent validation. This makes the headline behavior-recognition claim partially self-referential, even though the underlying detector comparisons remain valid. The abstract also misreports the YOLOv8s result (Table 1 gives mAP@0.5:0.95 = 0.522, not 0.855), but that is an accuracy error rather than circularity.

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

The central behavior claims rest on unvalidated geometric proxies with undisclosed threshold constants. The weights are not available, and the rule equations introduce free parameters without calibration. No new physical or conceptual entities are proposed.

free parameters (5)
  • beta (basking vertical distance threshold) = not reported
    Eq. 7 defines maximum vertical distance between dragon and lamp for basking; no value or calibration is given.
  • theta_max (basking angle threshold) = not reported
    Eq. 7 defines maximum angle from vertical for basking; no value or calibration is given.
  • gamma (hunting proximity threshold) = not reported
    Eq. 8 defines how close a cricket must be to a dragon for a disappearance to count as hunting; no value or calibration is given.
  • Detector confidence threshold = not reported
    Deployment parses detections at some confidence cutoff, but the cutoff is not stated for the activity pipeline.
  • Temporal interpolation gap (frames) = not reported
    The two-way nearest-frame strategy fills detection gaps, but the maximum gap length is not specified.
assumptions (4)
  • domain assumption Basking is equivalent to a dragon being close underneath a heating lamp, expressed by Eq. 7.
    Requires that vertical proximity and angle from lamp centers capture basking; no ethological validation is provided.
  • domain assumption Hunting is equivalent to a cricket disappearing near a dragon, expressed by Eq. 8.
    Assumes prey capture events manifest as cricket disappearance within gamma*W; no ground truth confirms this.
  • domain assumption Bounding-box centers represent animal and lamp positions adequately for behavior classification.
    Eqs. 6 and 8 use only center coordinates; occlusion and pose could break this assumption.
  • domain assumption The annotated dataset is representative of deployment videos.
    Data includes web images plus one animal unit; class distribution is heavily imbalanced (100 cricket images).

how reviews work

0 comments
Cite this review

Pith. "Pith review of Bearded Dragon Activity Recognition Pipeline: An AI-Based Approach to Behavioural Monitoring." pith.science (2026). https://pith.science/paper/BMLBN7TS

@misc{pith2026250717987,
  author       = {Pith},
  title        = {Pith review of: Bearded Dragon Activity Recognition Pipeline: An AI-Based Approach to Behavioural Monitoring},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BMLBN7TS}},
  note         = {Machine review of arXiv:2507.17987}
}
read the original abstract

Traditional monitoring of bearded dragon (Pogona Viticeps) behaviour is time-consuming and prone to errors. This project introduces an automated system for real-time video analysis, using You Only Look Once (YOLO) object detection models to identify two key behaviours: basking and hunting. We trained five YOLO variants (v5, v7, v8, v11, v12) on a custom, publicly available dataset of 1200 images, encompassing bearded dragons (600), heating lamps (500), and crickets (100). YOLOv8s was selected as the optimal model due to its superior balance of accuracy (mAP@0.5:0.95 = 0.855) and speed. The system processes video footage by extracting per-frame object coordinates, applying temporal interpolation for continuity, and using rule-based logic to classify specific behaviours. Basking detection proved reliable. However, hunting detection was less accurate, primarily due to weak cricket detection (mAP@0.5 = 0.392). Future improvements will focus on enhancing cricket detection through expanded datasets or specialised small-object detectors. This automated system offers a scalable solution for monitoring reptile behaviour in controlled environments, significantly improving research efficiency and data quality.

Figures

Figures reproduced from arXiv: 2507.17987 by the authors.

Figure 1
Figure 1. [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗
Figure 2
Figure 2. Annotated Hunting Screenshot 5 Conclusion and Future Work The pipeline integrates detection, interpolation, activity classification, and gen￾eration of annotated videos and logs. Basking was consistently identified with high recall and precision when both dragon and lamp were visible. Idle video clips yielded low false positives. Interpolation maintained tracking continuity through brief occlusions. Hunting recognit… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

37 extracted references · 36 canonical work pages

  1. [18]

    ReptiLearn: An automated home cage system for behavioral experiments in reptiles without human interven- tion

    Tal Eisenberg and Mark Shein-Idelson. “ReptiLearn: An automated home cage system for behavioral experiments in reptiles without human interven- tion”. In:PLOS Biology22.2 (2024), e3002411. (Visited on 11/29/2024)

    work page 2024

  2. [1]

    https://www.thebeardeddragon

    TheBeardedDragon.org.Bearded Dragon Behavior. https://www.thebeardeddragon. org/bearded-dragon/behavior. n.d. (Visited on 10/22/2024)

    work page 2024

  3. [2]

    Inclusive Enrichment for Dragons: Behavioral Responses of Amputee and Non-Amputee Individuals of Bearded Dragons Pogona vitticeps to Different Food Enrichment Items

    Luisa Martins Pereira et al. “Inclusive Enrichment for Dragons: Behavioral Responses of Amputee and Non-Amputee Individuals of Bearded Dragons Pogona vitticeps to Different Food Enrichment Items”. In:J. Zool. Bot. Gard.5.3 (2024), pp. 455–464.doi: 10 . 3390 / jzbg5030030.url: https : //doi.org/10.3390/jzbg5030030 (visited on 10/11/2024)

    work page doi:10.3390/jzbg5030030 2024

  4. [3]

    Identification of animals and recognition of their actions in wildlife videos using deep learning techniques

    F. Schindler and V. Steinhage. “Identification of animals and recognition of their actions in wildlife videos using deep learning techniques”. In:Eco- logical Informatics61 (2021), p. 101215. (Visited on 10/12/2024)

    work page 2021

  5. [4]

    Deep learning-based animal activity recognition with wearable sensors: Overview, challenges, and future directions

    Axiu Mao et al. “Deep learning-based animal activity recognition with wearable sensors: Overview, challenges, and future directions”. In:Com- puters and Electronics in Agriculture211 (2023), p. 108043. (Visited on 10/05/2024)

    work page 2023

  6. [5]

    The conservation status of the world’s reptiles

    Monika B¨ ohm et al. “The conservation status of the world’s reptiles”. In: Biological conservation157 (2013), pp. 372–385

    work page 2013

  7. [6]

    Version 1.0

    Arsen Yermukan and Pedro Machado.Bearded Dragons activity recogni- tion. Version 1.0. Zenodo, June 2025.doi: 10.5281/zenodo.15616848.url: https://doi.org/10.5281/zenodo.15616848

    work page doi:10.5281/zenodo.15616848.url: 2025

  8. [7]

    Surface-dwelling reptiles

    J. D. Willson. “Surface-dwelling reptiles”. In:Reptile Ecology and Con- servation: A Handbook of Techniques. Oxford University Press, UK, 2016, pp. 207–207. (Visited on 10/08/2024)

    work page 2016

Show all 37 references
  1. [8]

    Monitoring small animal usage patterns of suburban wildlife tunnels: Behaviour, design, and recommendations

    D. Pomezanski. “Monitoring small animal usage patterns of suburban wildlife tunnels: Behaviour, design, and recommendations”. PhD thesis. University of Guelph, 2017. (Visited on 10/21/2024)

    2017

  2. [9]

    Developing recommendations for mon- itoring wildlife underpass usage using trail cameras

    D. Pomezanski and L. Bennett. “Developing recommendations for mon- itoring wildlife underpass usage using trail cameras”. In:Environmental Monitoring and Assessment190 (2018), pp. 1–9. (Visited on 10/05/2024)

    2018

  3. [10]

    Effect of Camera Flash on the Behavior of Nocturnal Mammals: Implications Bearded Dragon Activity Recognition 11 for Wildlife Monitoring

    Daniel J. Welbourne, Andrew W. Claridge, and David J. Paull. “Effect of Camera Flash on the Behavior of Nocturnal Mammals: Implications Bearded Dragon Activity Recognition 11 for Wildlife Monitoring”. In:Austral Ecology42 (2017), pp. 666–677.doi: 10.1111/aec.12490. (Visited on...

    2017 doi

  4. [11]

    Camera Trap Methodologies for Long-Term Monitoring: Challenges and Opportunities

    Jamie W. McCallum, Michael Rowcliffe, and Innes Cuthill. “Camera Trap Methodologies for Long-Term Monitoring: Challenges and Opportunities”. In:Biological Conservation168 (2013), pp. 240–247.doi: 10.1016/j.biocon. 2013.10.017. (Visited on 10/28/2024)

    2013 doi

  5. [12]

    Photographic identification in reptiles: A matter of scales

    Roberto Sacchi et al. “Photographic identification in reptiles: A matter of scales”. In:Amphibia-Reptilia31.4 (2010), pp. 489–502.doi: 10.1163/ 017353710X521546. (Visited on 10/16/2024)

    2010

  6. [13]

    Revealing the unknown: Real-time recognition of Gal´ apagos snake species using deep learning

    A. Patel et al. “Revealing the unknown: Real-time recognition of Gal´ apagos snake species using deep learning”. In:Animals10.5 (2020), p. 806.doi: 10.3390/ani10050806. (Visited on 10/02/2024)

    2020 doi

  7. [14]

    Reptile Identifica- tion for Endemic and Invasive Alien Species Using Transfer Learning Ap- proaches

    R. Hern´ andez-L´ opez and C. M. Travieso-Gonz´ alez. “Reptile Identifica- tion for Endemic and Invasive Alien Species Using Transfer Learning Ap- proaches”. In:Sensors24.5 (2024), p. 1372.doi: 10.3390/s24051372. (Vis- ited on 10/06/2024)

    2024 doi

  8. [15]

    Large-scale Video Classification with Convolu- tional Neural Networks

    Andrej Karpathy et al. “Large-scale Video Classification with Convolu- tional Neural Networks”. In:Proceedings of the IEEE Conference on Com- puter Vision and Pattern Recognition. IEEE, 2014, pp. 1725–1732. (Visited on 10/20/2024)

    2014

  9. [16]

    Identification of reptile species using convolutional neural networks (CNN)

    Olvy Diaz Annesa, Condro Kartiko, Agi Prasetiadi, et al. “Identification of reptile species using convolutional neural networks (CNN)”. In:Jurnal RESTI (Rekayasa Sistem Dan Teknologi Informasi)4.5 (2020), pp. 899–

    2020

  10. [17]

    A review of recurrent neural networks: LSTM cells and network architectures

    Yong Yu et al. “A review of recurrent neural networks: LSTM cells and network architectures”. In:Neural computation31.7 (2019), pp. 1235–1270. (Visited on 11/11/2024)

    2019

  11. [19]

    Application of PIT tag and radio telemetry research methods for the effective management of reptiles in Korea national parks

    J. H. Lee et al. “Application of PIT tag and radio telemetry research methods for the effective management of reptiles in Korea national parks”. In:Korean Journal of Environmental Biology27.2 (2009), pp. 146–154. (Visited on 10/19/2024)

    2009

  12. [20]

    Movement patterns and telemetry

    Bruce A Kingsbury and Nathan J Robinson. “Movement patterns and telemetry”. In:Reptile ecology and conservation: a handbook of techniques 110 (2016). (Visited on 10/25/2024)

    2016

  13. [21]

    Terrestrial animal tracking as an eye on life and planet

    Roland Kays et al. “Terrestrial animal tracking as an eye on life and planet”. In:Science348.6240 (2015), aaa2478.doi: 10 . 1126 / science . aaa2478. (Visited on 10/03/2024)

    2015

  14. [22]

    Effects of Tagging and Radio Telemetry on Survival and Behavior of Small Song- birds

    John C. Withey, Thomas D. Bloxton Jr., and John M. Marzluff. “Effects of Tagging and Radio Telemetry on Survival and Behavior of Small Song- birds”. In:Journal of Wildlife Management65.4 (2001), pp. 806–815.doi: 10.2307/3803025. (Visited on 10/17/2024). 12Yermukan et al

    2001 doi

  15. [23]

    Biotelemetry: A mechanistic approach to ecology

    Steven J. Cooke et al. “Biotelemetry: A mechanistic approach to ecology”. In:Trends in Ecology and Evolution19.6 (2004), pp. 334–343.doi: 10 . 1016/j.tree.2004.04.003. (Visited on 10/12/2024)

    2004

  16. [24]

    The Higher the Better: The Influence of Camera Trap Height on Wildlife Detection

    Paul D. Meek, Guy-Anthony Ballard, and Greg Falzon. “The Higher the Better: The Influence of Camera Trap Height on Wildlife Detection”. In: PLOS One9 (2014), e93250.doi: 10.1371/journal.pone.0093250. (Visited on 10/14/2024)

    2014 doi

  17. [25]

    Deep learning models for real-time human activity recognition with smartphones

    Shaohua Wan et al. “Deep learning models for real-time human activity recognition with smartphones”. In:mobile networks and applications25.2 (2020), pp. 743–755. (Visited on 12/22/2024)

    2020

  18. [26]

    A Deep Convolutional Neural Net- work for Reptile Species Classification

    M. Ahmed, Y. Zong, and R. Zhao. “A Deep Convolutional Neural Net- work for Reptile Species Classification”. In:International Journal of Com- puter Vision and Pattern Recognition34.3 (2020), pp. 201–217. (Visited on 12/15/2024)

    2020

  19. [27]

    Improved Small Object Detection Algorithm CRL- YOLOv5

    Zhiyuan Wang et al. “Improved Small Object Detection Algorithm CRL- YOLOv5”. In:Sensors24.19 (2024), p. 6437.doi: 10 . 3390 / s24196437. (Visited on 04/22/2025)

    2024

  20. [28]

    Improved YOLOv7 for Small Object Detection Algo- rithm Based on Feature Enhancement

    Wei Zhang et al. “Improved YOLOv7 for Small Object Detection Algo- rithm Based on Feature Enhancement”. In:Applied Sciences13.16 (2023), p. 9316.doi: 10.3390/app13169316. (Visited on 04/05/2025)

    2023 doi

  21. [29]

    SOD-YOLOv8: Enhancing YOLOv8 for Small Object Detection in Traffic Scenes

    Boshra Khalili and Andrew W. Smyth. “SOD-YOLOv8: Enhancing YOLOv8 for Small Object Detection in Traffic Scenes”. In:arXiv preprint arXiv:2408.04786 (2024).url: https://arxiv.org/abs/2408.04786 (visited on 04/14/2025)

    2024 arXiv

  22. [30]

    YOLOv11 Optimization for Ef- ficient Resource Utilization

    Areeg Fahad Rasheed and M. Zarkoosh. “YOLOv11 Optimization for Ef- ficient Resource Utilization”. In:arXiv preprint arXiv:2412.14790(2024). url: https://arxiv.org/abs/2412.14790 (visited on 04/03/2025)

    2024 arXiv

  23. [31]

    YOLOv12: Attention-Centric Real-Time Object Detectors

    Smarterjie Sun et al. “YOLOv12: Attention-Centric Real-Time Object Detectors”. In:GitHub Repository(2025).url: https : / / github . com / sunsmarterjie/yolov12 (visited on 04/20/2025)

    2025

  24. [32]

    Yolo Versions Architecture

    Rusul Hussein Hasan, Rasha Majid Hassoo, and Inaam Salman Aboud. “Yolo Versions Architecture”. In:International Journal of Advances in Scientific Research and Engineering9.11 (2023), p. 73

    2023

  25. [33]

    Ac- cessed: 2025-04-21

    Ultralytics.YOLOv11: Attention-Centric Real-Time Object Detection. Ac- cessed: 2025-04-21. 2025.url: https : / / docs . ultralytics . com / models / yolo11/

    2025

  26. [34]

    Ac- cessed: 2025-04-21

    Ultralytics.YOLOv12: Attention-Centric Real-Time Object Detectors. Ac- cessed: 2025-04-21. 2025.url: https : / / docs . ultralytics . com / models / yolo12/

    2025

  27. [35]

    A multi-scale small object detection algo- rithm SMA-YOLO for UA V remote sensing images

    Shilong Zhou and Qian Lei. “A multi-scale small object detection algo- rithm SMA-YOLO for UA V remote sensing images”. In:Scientific Re- ports15.1 (2025), p. 8270.doi: 10.1038/s41598-025-92344-7. (Visited on 04/26/2025)

    2025 doi

  28. [36]

    Precision and speed: LSOD-YOLO for lightweight small object detection

    Hezheng Wang et al. “Precision and speed: LSOD-YOLO for lightweight small object detection”. In:Expert Systems with Applications(2025).doi: 10.1016/j.eswa.2025.126440. (Visited on 04/16/2025)

    2025

  29. [906]

    (Visited on 11/14/2024)

    2024

Pith tools

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