Underwater Waste Detection Using Deep Learning A Performance Comparison of YOLOv7 to 10 and Faster RCNN

HMLS Kumari; HMNS Kumari; UMMPK Nawarathne

arxiv: 2507.18967 · v1 · submitted 2025-07-25 · 💻 cs.CV · cs.AI· cs.LG

Underwater Waste Detection Using Deep Learning A Performance Comparison of YOLOv7 to 10 and Faster RCNN

UMMPK Nawarathne , HMNS Kumari , HMLS Kumari This is my paper

Pith reviewed 2026-05-19 02:58 UTC · model grok-4.3

classification 💻 cs.CV cs.AIcs.LG

keywords underwater waste detectionobject detectionYOLOv8deep learningmarine pollutioncomputer visionenvironmental monitoring

0 comments

The pith

YOLOv8 achieves 80.9% mAP for underwater waste detection, surpassing YOLOv7 to YOLOv10 and Faster R-CNN.

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

This paper evaluates the effectiveness of several advanced object detection models for identifying various types of waste in underwater settings. Researchers trained and tested YOLOv7, YOLOv8, YOLOv9, YOLOv10, and Faster R-CNN on a dataset covering 15 waste classes in conditions with low visibility and changing depths. YOLOv8 delivered the best results at 80.9% mean average precision, credited to its anchor-free design and self-supervised learning capabilities. Such performance points to its usefulness in scaling up efforts to detect and remove pollution from oceans and rivers.

Core claim

Among the five object recognition algorithms examined, YOLOv8 proves most effective at recognizing materials in underwater situations, attaining a mean Average Precision of 80.9% on a 15-class dataset, thanks to its incorporation of improved anchor-free mechanisms and self-supervised learning.

What carries the argument

Comparative evaluation of YOLOv7 through YOLOv10 and Faster R-CNN models on an underwater waste dataset with 15 classes, highlighting YOLOv8's architectural advantages for precise detection in challenging marine environments.

If this is right

Enhances the accuracy and efficiency of underwater pollution monitoring programs.
Supports the development of more scalable systems for marine waste management.
Improves detection capabilities in low-visibility and variable-depth conditions common in real-world scenarios.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

Integration with robotic systems could automate waste collection in underwater environments.
Similar model comparisons might benefit detection tasks in other obscured settings like fog or dense vegetation.
Further optimization could focus on real-time processing speeds for deployment on underwater drones.

Load-bearing premise

The 15-class dataset collected under diverse underwater conditions adequately represents the variety of real-world waste and environments for the performance results to apply generally.

What would settle it

Evaluating the models on a separate underwater dataset from different locations or with additional waste categories where YOLOv8 no longer shows the highest mAP.

read the original abstract

Underwater pollution is one of today's most significant environmental concerns, with vast volumes of garbage found in seas, rivers, and landscapes around the world. Accurate detection of these waste materials is crucial for successful waste management, environmental monitoring, and mitigation strategies. In this study, we investigated the performance of five cutting-edge object recognition algorithms, namely YOLO (You Only Look Once) models, including YOLOv7, YOLOv8, YOLOv9, YOLOv10, and Faster Region-Convolutional Neural Network (R-CNN), to identify which model was most effective at recognizing materials in underwater situations. The models were thoroughly trained and tested on a large dataset containing fifteen different classes under diverse conditions, such as low visibility and variable depths. From the above-mentioned models, YOLOv8 outperformed the others, with a mean Average Precision (mAP) of 80.9%, indicating a significant performance. This increased performance is attributed to YOLOv8's architecture, which incorporates advanced features such as improved anchor-free mechanisms and self-supervised learning, allowing for more precise and efficient recognition of items in a variety of settings. These findings highlight the YOLOv8 model's potential as an effective tool in the global fight against pollution, improving both the detection capabilities and scalability of underwater cleanup operations.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

Standard YOLO comparison on underwater waste data with YOLOv8 on top, but no evidence the win comes from the claimed architectural features rather than training choices.

read the letter

The paper runs a direct comparison of YOLOv7, v8, v9, v10 and Faster R-CNN on a 15-class underwater waste dataset collected under low visibility and varying depths. YOLOv8 reaches 80.9% mAP and the authors link that result to its anchor-free design and self-supervised elements. That is the core claim to evaluate. The work is new only in the narrow sense that it applies these models to this particular pollution-monitoring setting; the architectures and the general approach are already well documented elsewhere. What it does reasonably is collect results across several recent YOLO releases on the same data split, which gives practitioners a quick ordering to try first when they need an off-the-shelf detector for marine debris. The dataset conditions themselves are relevant to real deployment, so the numbers have some applied utility even if they are not surprising. The main weakness is that the experimental controls are not described. The abstract and the reported results give no indication that every model used the same optimizer, learning-rate schedule, number of epochs, batch size, or data-augmentation policy. In the absence of that information, the 10-point gaps that appear between versions could easily trace to implementation details rather than the architectural differences the authors highlight. No error bars, no repeated runs, and no ablation that isolates the anchor-free or self-supervised components are mentioned either. Those omissions make the causal attribution hard to accept at face value. The paper is aimed at applied computer-vision groups working on environmental monitoring rather than readers looking for methodological advances. A serious referee could still be useful if the full methods section supplies the missing training protocol, releases the dataset and code, and adds at least basic statistical checks. Without those additions the central performance claim stays under-supported and the paper would be better as a short technical report than a journal submission.

Referee Report

3 major / 2 minor

Summary. The manuscript reports an empirical comparison of five object detection models—YOLOv7, YOLOv8, YOLOv9, YOLOv10, and Faster R-CNN—on a 15-class underwater waste dataset collected under conditions of low visibility and variable depths. It states that YOLOv8 achieves the highest mean Average Precision (mAP) of 80.9% and attributes this result to architectural elements including improved anchor-free mechanisms and self-supervised learning.

Significance. A well-controlled comparison of these models on underwater imagery could offer practical guidance for environmental monitoring and cleanup robotics. The current manuscript, however, provides no evidence that performance differences arise from the cited architectural features rather than uncontrolled differences in training protocols, making the central claim difficult to interpret or reproduce.

major comments (3)

[Abstract] Abstract: The attribution of the 80.9% mAP to YOLOv8's 'improved anchor-free mechanisms and self-supervised learning' is unsupported. No section describes a unified training protocol (identical optimizer, epoch count, batch size, learning rate schedule, or data augmentation pipeline) applied to all five models; without such isolation, mAP gaps cannot be causally linked to architecture.
[Methods/Experimental Setup] Methods/Experimental Setup (inferred from absence in Abstract and Results): The manuscript supplies no information on dataset partitioning, validation strategy, pre-training usage, or loss weighting. These details are load-bearing for any claim that one detector outperforms the others on the 15-class underwater task.
[Results] Results: No error bars, standard deviations across runs, or statistical significance tests accompany the reported mAP values. A single scalar of 80.9% for YOLOv8 cannot be evaluated for robustness or superiority without these measures.

minor comments (2)

[Title] Title: 'YOLOv7 to 10' is imprecise; it should explicitly list YOLOv7 through YOLOv10 for clarity.
[Abstract] Abstract: The phrase 'self-supervised learning' is used without definition or citation in the context of YOLOv8; standard YOLOv8 training is supervised, so the term requires clarification or removal.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive feedback, which identifies important gaps in methodological transparency and statistical rigor. We address each major comment below and will revise the manuscript to strengthen reproducibility and support for our claims without overstating the current evidence.

read point-by-point responses

Referee: [Abstract] Abstract: The attribution of the 80.9% mAP to YOLOv8's 'improved anchor-free mechanisms and self-supervised learning' is unsupported. No section describes a unified training protocol (identical optimizer, epoch count, batch size, learning rate schedule, or data augmentation pipeline) applied to all five models; without such isolation, mAP gaps cannot be causally linked to architecture.

Authors: We agree that the abstract's attribution requires supporting details to establish a causal link. The manuscript states that models were 'thoroughly trained and tested' under similar conditions, but does not explicitly document a unified protocol. In the revision we will expand the Methods section with a table and description of the shared training configuration (optimizer, epochs, batch size, LR schedule, and augmentations) applied to all models, and we will revise the abstract to qualify the architectural attribution accordingly. revision: yes
Referee: [Methods/Experimental Setup] Methods/Experimental Setup (inferred from absence in Abstract and Results): The manuscript supplies no information on dataset partitioning, validation strategy, pre-training usage, or loss weighting. These details are load-bearing for any claim that one detector outperforms the others on the 15-class underwater task.

Authors: We acknowledge that these implementation details were omitted. The full manuscript text does not contain a dedicated experimental-setup subsection covering splits, validation, pre-training, or loss weighting. In the revised version we will add this information, specifying the train/validation/test partitioning ratios, any cross-validation procedure, use of COCO pre-trained weights, and class-balanced loss weighting for the 15 underwater waste categories. revision: yes
Referee: [Results] Results: No error bars, standard deviations across runs, or statistical significance tests accompany the reported mAP values. A single scalar of 80.9% for YOLOv8 cannot be evaluated for robustness or superiority without these measures.

Authors: We concur that single-run mAP values limit assessment of robustness. The current results section reports only point estimates. For the revision we will rerun the top-performing models with multiple random seeds to report mean mAP and standard deviation, and we will add a statistical comparison (e.g., paired t-test or Wilcoxon test) between YOLOv8 and the next-best model. If resource constraints prevent full multi-run experiments, we will explicitly note this limitation and qualify the superiority claim. revision: partial

Circularity Check

0 steps flagged

No circularity: purely empirical comparison with no derivations

full rationale

The paper reports mAP results from training YOLOv7–v10 and Faster R-CNN on a 15-class underwater dataset. No equations, derivations, or self-citations appear in the provided text. The claim that YOLOv8's 80.9% mAP stems from its architecture is an interpretive statement about observed results rather than a reduction of any prediction to fitted inputs or self-referential definitions. The work is self-contained as an empirical benchmark study.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The paper rests on standard assumptions of deep learning training and evaluation; no new axioms or entities introduced beyond the choice of models and dataset.

free parameters (1)

training hyperparameters
Learning rates, batch sizes, and augmentation parameters fitted during model training on the underwater dataset.

axioms (1)

domain assumption Standard supervised learning assumptions hold for object detection on the collected underwater images.
Invoked implicitly when reporting mAP as measure of effectiveness under low visibility conditions.

pith-pipeline@v0.9.0 · 5783 in / 1021 out tokens · 45894 ms · 2026-05-19T02:58:19.502533+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

YOLOv8 outperformed the others, with a mean Average Precision (mAP) of 80.9%... attributed to YOLOv8's architecture, which incorporates advanced features such as improved anchor-free mechanisms and self-supervised learning
IndisputableMonolith/Foundation/RealityFromDistinction.lean reality_from_one_distinction unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

The models were thoroughly trained and tested on a large dataset containing fifteen different classes under diverse conditions, such as low visibility and variable depths.

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

23 extracted references · 23 canonical work pages

[1]

The impact of debris on marine life,

S. C. Gall and R. C. Thompson, “The impact of debris on marine life,” Mar. Pollut. Bull., vol. 92, no. 1–2, pp. 170–179, Mar. 2015, doi: 10.1016/j.marpolbul.2014.12.041

work page doi:10.1016/j.marpolbul.2014.12.041 2015
[2]

Garbage Detection using Advanced Object Detection Techniques,

D. Patel, F. Patel, S. Patel, N. Patel, D. Shah, and V. Patel, “Garbage Detection using Advanced Object Detection Techniques,” in Proceedings - International Conference on Artificial Intelligence and Smart Systems, ICAIS 2021, Institute of Electrical and Electronics Engineers Inc., Mar. 2021, pp. 526 –531. doi: 10.1109/ICAIS50930.2021.9395916

work page doi:10.1109/icais50930.2021.9395916 2021
[3]

rochman2015

J. A. van F. Chelsea M. Rochman , Mark Anthony Browne , A. J. Underwood and L. A. A.-Z. Richard C. Thompson, “rochman2015”

work page
[4]

Global Coral Reef Monitoring Network and Reef and Rainforest Research Centre: Townsville,

C. Wilkinson and D. Souter, “Global Coral Reef Monitoring Network and Reef and Rainforest Research Centre: Townsville,” AU, USA, pp. 31–36, 2008

work page 2008
[5]

The economic cost and control of marine debris damage in the Asia-Pacific region,

A. McIlgorm, H. F. Campbell, and M. J. Rule, “The economic cost and control of marine debris damage in the Asia-Pacific region,” Ocean Coast. Manag., vol. 54, no. 9, pp. 643 –651, Sep. 2011, doi: 10.1016/j.ocecoaman.2011.05.007

work page doi:10.1016/j.ocecoaman.2011.05.007 2011
[6]

Impacts of pollution on coastal and marine ecosystems including coastal and marine fisheries and approach for management: A review and synthesis,

M. S. Islam and M. Tanaka, “Impacts of pollution on coastal and marine ecosystems including coastal and marine fisheries and approach for management: A review and synthesis,” Apr. 2004. doi: 10.1016/j.marpolbul.2003.12.004

work page doi:10.1016/j.marpolbul.2003.12.004 2004
[7]

The economics of marine litter,

S. Newman, E. Watkins, A. Farmer, P. ten Brink, and J. -P. Schweitzer, “The economics of marine litter,” Mar. Anthropog. litter, pp. 367–394, 2015

work page 2015
[8]

An Image Processing Approach to Identify Plastic Garbage,

A. Rahman, W. R. Bristi, A. T. Misha, and others, “An Image Processing Approach to Identify Plastic Garbage,” in 2021 International Conference on Electronics, Communications and Information Technology (ICECIT), 2021, pp. 1–4

work page 2021
[9]

Garbage Classification Detection Based on Improved YOLOV4,

Q. Chen and Q. Xiong, “Garbage Classification Detection Based on Improved YOLOV4,” J. Comput. Commun., vol. 08, no. 12, pp. 285–294, 2020, doi: 10.4236/jcc.2020.812023

work page doi:10.4236/jcc.2020.812023 2020
[10]

Underwater Image Detection for Cleaning Purposes; Techniques Used for Detection Based on Machine Learning,

O. Goxhaj, N. G. Yilmaz, L. Kouhalvandi, I. Shayea, and A. Azizan, “Underwater Image Detection for Cleaning Purposes; Techniques Used for Detection Based on Machine Learning,” Acta Marisiensis. Ser. Technol., vol. 19, no. 1, pp. 28 –35, Jun. 2022, doi: 10.2478/amset-2022-0006

work page doi:10.2478/amset-2022-0006 2022
[11]

The Object Detection of Underwater Garbage with an Improved YOLOv5 Algorithm,

X. Teng, Y. Fei, K. He, and L. Lu, “The Object Detection of Underwater Garbage with an Improved YOLOv5 Algorithm,” in Proceedings of the 2022 International Conference on Pattern Recognition and Intelligent Systems, 2022, pp. 55–60

work page 2022
[12]

Detection of Underwater Plastic Waste Based on Improved YOLOv5n,

Z. Hu and C. Xu, “Detection of Underwater Plastic Waste Based on Improved YOLOv5n,” pp. 404 –408, Dec. 2022, doi: https://doi.org/10.1109/icftic57696.2022.10075134

work page doi:10.1109/icftic57696.2022.10075134 2022
[13]

Underwater object detection algorithm based on an improved YOLOv8,

F. Zhang et al., “Underwater object detection algorithm based on an improved YOLOv8,” Journal of Marine Science and Engineering, vol. 12, no. 11, p. 1991, 2024

work page 1991
[14]

A lightweight YOLOv8 integrating FasterNet for real -time underwater object detection,

A. Guo, K. Sun, and Z. Zhang, “A lightweight YOLOv8 integrating FasterNet for real -time underwater object detection,” Journal of Real-Time Image Processing, vol. 21, no. 2, p. 49, 2024

work page 2024
[15]

Underwater Plastic Pollution Detection

A. Samal , “Underwater Plastic Pollution Detection.” [Online]. Available: Underwater Plastic Pollution Detection%0A

work page
[16]

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.” [Online]. Available: https://github.com/

work page
[17]

Ultralytics YOLO,

G. Jocher, A. Chaurasia, and J. Qiu, “Ultralytics YOLO,” 2023. [Online]. Available: https://ultralytics.com

work page 2023
[18]

13, 2024

What is YOLOv8? Exploring its Cutting -Edge Features - YOLOv8,” Jan. 13, 2024. https://yolov8.org/what-is-yolov8/

work page 2024
[19]

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.” [Online]. Available: http://pjreddie.com/yolo/

work page
[20]

arXiv preprint arXiv:2405.14458 , year=

A. Wang et al., “YOLOv10: Real -Time End -to-End Object Detection,” May 2024, [Online]. Available: http://arxiv.org/abs/2405.14458

work page arXiv 2024
[21]

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,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 39, no. 6, pp. 1137–1149, 2016

work page 2016
[22]

Scikit-learn: Machine Learning in Python,

F. Pedregosa et al., “Scikit-learn: Machine Learning in Python,” J. Mach. Learn. Res., vol. 12, pp. 2825–2830, 2011

work page 2011
[23]

Ultralytics YOLO Docs

Ultralytics, “Ultralytics YOLO Docs.” [Online]. Available: https://docs.ultralytics.com/guides/yolo-performance- metrics/#object-detection-metrics

work page

[1] [1]

The impact of debris on marine life,

S. C. Gall and R. C. Thompson, “The impact of debris on marine life,” Mar. Pollut. Bull., vol. 92, no. 1–2, pp. 170–179, Mar. 2015, doi: 10.1016/j.marpolbul.2014.12.041

work page doi:10.1016/j.marpolbul.2014.12.041 2015

[2] [2]

Garbage Detection using Advanced Object Detection Techniques,

D. Patel, F. Patel, S. Patel, N. Patel, D. Shah, and V. Patel, “Garbage Detection using Advanced Object Detection Techniques,” in Proceedings - International Conference on Artificial Intelligence and Smart Systems, ICAIS 2021, Institute of Electrical and Electronics Engineers Inc., Mar. 2021, pp. 526 –531. doi: 10.1109/ICAIS50930.2021.9395916

work page doi:10.1109/icais50930.2021.9395916 2021

[3] [3]

rochman2015

J. A. van F. Chelsea M. Rochman , Mark Anthony Browne , A. J. Underwood and L. A. A.-Z. Richard C. Thompson, “rochman2015”

work page

[4] [4]

Global Coral Reef Monitoring Network and Reef and Rainforest Research Centre: Townsville,

C. Wilkinson and D. Souter, “Global Coral Reef Monitoring Network and Reef and Rainforest Research Centre: Townsville,” AU, USA, pp. 31–36, 2008

work page 2008

[5] [5]

The economic cost and control of marine debris damage in the Asia-Pacific region,

A. McIlgorm, H. F. Campbell, and M. J. Rule, “The economic cost and control of marine debris damage in the Asia-Pacific region,” Ocean Coast. Manag., vol. 54, no. 9, pp. 643 –651, Sep. 2011, doi: 10.1016/j.ocecoaman.2011.05.007

work page doi:10.1016/j.ocecoaman.2011.05.007 2011

[6] [6]

Impacts of pollution on coastal and marine ecosystems including coastal and marine fisheries and approach for management: A review and synthesis,

M. S. Islam and M. Tanaka, “Impacts of pollution on coastal and marine ecosystems including coastal and marine fisheries and approach for management: A review and synthesis,” Apr. 2004. doi: 10.1016/j.marpolbul.2003.12.004

work page doi:10.1016/j.marpolbul.2003.12.004 2004

[7] [7]

The economics of marine litter,

S. Newman, E. Watkins, A. Farmer, P. ten Brink, and J. -P. Schweitzer, “The economics of marine litter,” Mar. Anthropog. litter, pp. 367–394, 2015

work page 2015

[8] [8]

An Image Processing Approach to Identify Plastic Garbage,

A. Rahman, W. R. Bristi, A. T. Misha, and others, “An Image Processing Approach to Identify Plastic Garbage,” in 2021 International Conference on Electronics, Communications and Information Technology (ICECIT), 2021, pp. 1–4

work page 2021

[9] [9]

Garbage Classification Detection Based on Improved YOLOV4,

Q. Chen and Q. Xiong, “Garbage Classification Detection Based on Improved YOLOV4,” J. Comput. Commun., vol. 08, no. 12, pp. 285–294, 2020, doi: 10.4236/jcc.2020.812023

work page doi:10.4236/jcc.2020.812023 2020

[10] [10]

Underwater Image Detection for Cleaning Purposes; Techniques Used for Detection Based on Machine Learning,

O. Goxhaj, N. G. Yilmaz, L. Kouhalvandi, I. Shayea, and A. Azizan, “Underwater Image Detection for Cleaning Purposes; Techniques Used for Detection Based on Machine Learning,” Acta Marisiensis. Ser. Technol., vol. 19, no. 1, pp. 28 –35, Jun. 2022, doi: 10.2478/amset-2022-0006

work page doi:10.2478/amset-2022-0006 2022

[11] [11]

The Object Detection of Underwater Garbage with an Improved YOLOv5 Algorithm,

X. Teng, Y. Fei, K. He, and L. Lu, “The Object Detection of Underwater Garbage with an Improved YOLOv5 Algorithm,” in Proceedings of the 2022 International Conference on Pattern Recognition and Intelligent Systems, 2022, pp. 55–60

work page 2022

[12] [12]

Detection of Underwater Plastic Waste Based on Improved YOLOv5n,

Z. Hu and C. Xu, “Detection of Underwater Plastic Waste Based on Improved YOLOv5n,” pp. 404 –408, Dec. 2022, doi: https://doi.org/10.1109/icftic57696.2022.10075134

work page doi:10.1109/icftic57696.2022.10075134 2022

[13] [13]

Underwater object detection algorithm based on an improved YOLOv8,

F. Zhang et al., “Underwater object detection algorithm based on an improved YOLOv8,” Journal of Marine Science and Engineering, vol. 12, no. 11, p. 1991, 2024

work page 1991

[14] [14]

A lightweight YOLOv8 integrating FasterNet for real -time underwater object detection,

A. Guo, K. Sun, and Z. Zhang, “A lightweight YOLOv8 integrating FasterNet for real -time underwater object detection,” Journal of Real-Time Image Processing, vol. 21, no. 2, p. 49, 2024

work page 2024

[15] [15]

Underwater Plastic Pollution Detection

A. Samal , “Underwater Plastic Pollution Detection.” [Online]. Available: Underwater Plastic Pollution Detection%0A

work page

[16] [16]

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.” [Online]. Available: https://github.com/

work page

[17] [17]

Ultralytics YOLO,

G. Jocher, A. Chaurasia, and J. Qiu, “Ultralytics YOLO,” 2023. [Online]. Available: https://ultralytics.com

work page 2023

[18] [18]

13, 2024

What is YOLOv8? Exploring its Cutting -Edge Features - YOLOv8,” Jan. 13, 2024. https://yolov8.org/what-is-yolov8/

work page 2024

[19] [19]

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.” [Online]. Available: http://pjreddie.com/yolo/

work page

[20] [20]

arXiv preprint arXiv:2405.14458 , year=

A. Wang et al., “YOLOv10: Real -Time End -to-End Object Detection,” May 2024, [Online]. Available: http://arxiv.org/abs/2405.14458

work page arXiv 2024

[21] [21]

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,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 39, no. 6, pp. 1137–1149, 2016

work page 2016

[22] [22]

Scikit-learn: Machine Learning in Python,

F. Pedregosa et al., “Scikit-learn: Machine Learning in Python,” J. Mach. Learn. Res., vol. 12, pp. 2825–2830, 2011

work page 2011

[23] [23]

Ultralytics YOLO Docs

Ultralytics, “Ultralytics YOLO Docs.” [Online]. Available: https://docs.ultralytics.com/guides/yolo-performance- metrics/#object-detection-metrics

work page