Pith. sign in

REVIEW 4 major objections 4 minor 39 references

Detecting and Classifying Defective Products in Images Using YOLO

T0 review · 4 major / 4 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read One YOLO network finds machine-part defects at 0.91 mAP in 25 ms

desk verdict Routine YOLO application on a private dataset with no baseline or ablations; the 'improved' claim is unsupported and the paper should be desk-rejected. read the letter →

arxiv 2412.16935 v1 pith:JEDCLSTA submitted 2024-12-22 cs.CV

Zhen Qi , Liwei Ding , Xiangtian Li , Jiacheng Hu , Bin Lyu , Ao Xiang This is my paper · ORCID
classification cs.CV
keywords ProductdefectdetectionYOLOdeeplearningreal-timeindustrialautomationobjectmachinepartinspectionmeanaverageprecision
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 reports that an improved YOLO detector can locate and classify common machine-part defects—scratches, cracks, wear, broken teeth, burrs, deformation, and rust—in product images while running fast enough for factory inspection. The authors aim to show that a single one-pass neural network, upgraded with multi-scale feature extraction and fusion, can replace slow manual checks and simple machine-vision pipelines. On the paper's own private dataset of bearings, gears, bolts, and mixed defects, the model reaches an overall mean average precision (mAP) of 0.91, with precision 0.94 and recall 0.92, and processes 1280×720 frames in 25–32 ms (31–40 FPS) depending on scene complexity. If the reported results hold, the main consequence is that real-time automated product quality inspection becomes practical; the authors also note that 1080p inputs slow to 50 ms and 20 FPS.

What carries the argument

The carrying machinery is a YOLO-style single-pass detector augmented in three places. The backbone uses an improved ResC2Net block: the feature map is split into sub-feature maps, each passed through residual (Res2Net-style) branches, then concatenated, so the network captures features at several scales in one pass. The neck adds cross-layer connections, partial convolution (PConv) to cut parameters, and an SPPF (Spatial Pyramid Pooling-Fast) module that max-pools at several scales to add global context. The output layer makes multi-scale predictions. Training uses the Adam optimizer, an IoU-based localization loss, and a weighted confidence loss to keep rare defect types from being ignored.

What would settle it

Re-running the same training recipe on a publicly available surface-defect image set and getting an overall mAP well below 0.91, or measuring no improvement over a standard YOLO baseline when the ResC2Net, PConv, and SPPF changes are ablated, would contradict the reported advantage. Because the paper's tables refer to a private dataset, a direct check would need the authors to release the dataset or a third party to reproduce the Table 2 and Table 3 numbers on equivalent data.

Watch

Extended reading notes

Core claim

The central claim is that the improved YOLO model—a detector whose backbone uses an improved ResC2Net residual structure, whose neck fuses multi-scale features with PConv and SPPF pooling, and whose output layer predicts at multiple scales—detects and classifies five defect types across machine parts with precision 0.94, recall 0.92, F1 0.93, and overall mAP 0.91 on the paper's test set. Detection times are 25 ms for simple backgrounds, 30 ms for complex backgrounds, and 32 ms for multi-target scenes, corresponding to 40, 33, and 31 frames per second; high-resolution 1920×1080 inputs run at 50 ms and 20 FPS. The authors maintain that these numbers meet real-time industrial inspection requirements while keeping per-class mAP between 0.89 and 0.93.

Load-bearing premise

The evaluation assumes the private dataset described in Table 1 exists, is balanced across defect types, and is correctly annotated; every reported precision, recall, and mAP number depends on that unverifiable resource.

Editorial extensions

If this is right

  • If the reported numbers hold, the model can run on a production line at 31–40 FPS and catch the five studied defect classes at a 0.5 IoU threshold while maintaining per-class mAP above 0.89.
  • A 25 ms average per 1280×720 frame means a single GPU can keep up with many conveyor lines without stopping production.
  • High-resolution inspection at 1920×1080 remains feasible at 20 FPS, so the same model can trade speed for fine detail when needed.
  • The reported stability of mAP above 0.91 across bearings, gears, bolts, and mixed defects supports the idea that one detector, rather than a separate model per part type, is enough for these categories.

Reading between the lines

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

  • Editorial inference: The reported per-class mAP spread (0.89–0.93) suggests visually subtle or low-frequency defects such as rust and deformation are the hardest cases; a testable extension is to add a new defect type and watch whether per-class mAP tracks the number of training examples.
  • Editorial inference: Because the paper does not ablate its three architecture changes, a reader cannot attribute the 0.91 mAP to any one component; comparing ResC2Net, PConv, and SPPF individually against a plain YOLO baseline would settle which modification matters.
  • Editorial inference: The controlled-lighting, 80/20 split dataset leaves open whether the model survives factory-floor shifts in camera, lighting, or part pose; evaluating on images from a different acquisition setup is a direct robustness test the paper does not run.
  • Editorial inference: The paper's own limitation note that high-resolution inputs drop to 20 FPS suggests the real-time advantage depends on choosing input resolution carefully; a production deployment would likely need resolution-adaptive inference.
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 / 4 minor

Summary. This paper proposes an "improved YOLO" detector for classifying defects in images of machine parts such as bearings, gears, and bolts. Section 3 describes a private dataset with four subsets, an architecture built around modified ResC2Net, SPPF, and PConv modules, and a training protocol using the Adam optimizer over 200 epochs on an NVIDIA Tesla V100. Section 4 reports per-defect precision, recall, F1, and mAP values in Table 2 and real-time latency and frame-rate measurements in Table 3, claiming an overall mAP of 0.91 and 20–40 FPS depending on scenario. Sections 5–6 discuss practical application and limitations. The central claim is that the proposed modifications improve YOLO for industrial defect detection while maintaining real-time operation.

Significance. If the reported results are correct, the claimed real-time mAP around 0.91 on machine-part defect detection would be practically relevant for automated quality inspection. The paper has clear organization and states several training details, including learning rate, batch size, weight decay, epoch count, and GPU hardware. However, the significance is entirely contingent on results that are not currently verifiable: there is no comparison with an unmodified YOLO baseline, no ablation of the proposed modules, no error bars or multiple-run statistics, and no release of the dataset, code, or trained weights. As presented, the paper provides descriptive performance numbers rather than evidence for an improvement, so its scientific contribution is not established.

major comments (4)
  1. [Section 4, Table 2 and Section 3.2] The paper's central claim of an "improved" YOLO model is not supported because no experiment compares the proposed model with the original YOLO algorithm or with any other detection baseline. Table 2 reports only the proposed model's precision, recall, F1, and mAP values. In addition, the contributions of the ResC2Net, SPPF, and PConv components described in Section 3.2 are never ablated, so the reported performance cannot be attributed to the proposed modifications.
  2. [Section 3.1, Table 1 and Section 4] The evaluation rests entirely on a private dataset that is described only by aggregate image counts per subset. No images, annotation files, per-class bounding-box counts, training/validation/test split indices, or annotation examples are provided, and no code or trained weights are released. Consequently, the precision, recall, and mAP values in Tables 2 and 3 cannot be independently reproduced or verified, which is load-bearing because the paper's only evidence is empirical.
  3. [Section 3.2, Equations (1) and (2)] Equations (1) and (2) are under-specified to the point of preventing reproduction. In Equation (1), the final term is written f3(x) rather than f_n(x), and the number of sub-feature maps n is never defined; f_i(x) is described only as "the processing result of the i-th sub-feature map," with no residual branch structure, channel counts, or split procedure. In Equation (2), MaxPool1 through MaxPooln are given without kernel sizes, strides, pooling scales, or output dimensions. The architecture described in the text is therefore not implementable from the information supplied.
  4. [Section 4, Tables 2 and 3] All metrics are point estimates from a single 80/20 split of the data, with no confidence intervals, standard deviations, or multiple-run statistics, and no per-class test-sample sizes are reported. With only four defect classes, the differences among mAP values (0.89 to 0.93) may not be meaningful; however, the text draws comparative conclusions, such as bolt-deformation detection being "relatively weaker," without any statistical support.
minor comments (4)
  1. [Section 3.1] The preprocessing description says images were standardized "such as resizing and grayscaling," but the same paragraph states that color jittering was applied as an augmentation; if images are grayscale, color jittering is meaningless, so the preprocessing and augmentation pipeline should be clarified.
  2. [Section 3.2] The notation in Equation (1) should be corrected from f3(x) to f_n(x), and the relationship between the module name "ResC2Net" and the referenced "Res2Net" residual networks should be stated precisely.
  3. [References] Many references are unrelated to object detection (for example, [1], [2], and [4]), while standard YOLO references and the original sources for SPPF and PConv are not cited, making it difficult to situate the proposed architecture in the literature.
  4. [Section 3.3] The grid-search procedure is described as "fine-tuning" hyperparameters, but no search ranges, number of trials, or selection criterion are given, so the claim that the final settings were selected by grid search cannot be assessed.

Circularity Check

0 steps flagged · score 2.0 of 10

No circular derivation: reported accuracy and speed are empirical measurements, not predictions derived from fitted assumptions.

full rationale

No circular reduction is present in the claimed derivation chain. The paper reports supervised training and evaluation of a YOLO variant on a private dataset: Section 4 presents precision, recall, F1, mAP, and FPS values in Tables 2 and 3 as measurements on a held-out test split. Equations 1 and 2 define the ResC2Net and SPPF modules, and Equation 6 defines IoU; none of these definitions by construction implies the reported mAP or timing values. Hyperparameters were tuned on the validation set, and early stopping was used, which is standard practice and not a case of fitting parameters to the test-set metric. The central weakness is not circularity but unfalsifiability: the dataset is private, no per-class statistics or annotation examples are provided, and no baseline or ablation isolates the contribution of the proposed modules. That is a reproducibility and correctness concern, not an equivalence between input and output. The only self-citation is reference [7], an Improved YOLOv5 paper by overlapping author Qi Z, cited in Section 2 in the sentence 'the research subjects are machine parts[7]'; it is not used to justify the ResC2Net, SPPF, or PConv changes, nor the mAP or FPS claims. It therefore does not constitute load-bearing self-citation, and no circular step can be exhibited.

Assumptions & free parameters 5 free parameters · 2 assumptions · 0 invented entities

The central empirical claim rests on hand-chosen hyperparameters and on the unverified quality and availability of a private dataset. No new entities are introduced; the architecture is a combination of existing modules.

free parameters (5)
  • Learning rate = 0.001
    Chosen by grid search (Section 3.3); no sensitivity analysis.
  • Batch size = 16
    Chosen by grid search (Section 3.3); no sensitivity analysis.
  • Weight decay = 0.0005
    Chosen by grid search (Section 3.3); no sensitivity analysis.
  • Training epochs = 200
    Set in Section 3.3 with early stopping; no analysis of epoch count.
  • IoU threshold = 0.5
    Standard threshold used for metrics (Section 4); not varied.
assumptions (2)
  • domain assumption The private dataset images are representative of real production-line conditions and correctly annotated by experts.
    Section 3.1 states the images were collected in controlled environments and a 'rigorous review process' was used; no data or annotation examples are provided, so this is an unverifiable domain assumption.
  • domain assumption The YOLO baseline and the added modules (ResC2Net, SPPF, PConv) behave as standard implementations without implementation bugs.
    Section 3.2 describes the modules at a high level with no code, so correct implementation is assumed.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Detecting and Classifying Defective Products in Images Using YOLO." pith.science (2026). https://pith.science/paper/JEDCLSTA

@misc{pith2026241216935,
  author       = {Pith},
  title        = {Pith review of: Detecting and Classifying Defective Products in Images Using YOLO},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JEDCLSTA}},
  note         = {Machine review of arXiv:2412.16935}
}
read the original abstract

With the continuous advancement of industrial automation, product quality inspection has become increasingly important in the manufacturing process. Traditional inspection methods, which often rely on manual checks or simple machine vision techniques, suffer from low efficiency and insufficient accuracy. In recent years, deep learning technology, especially the YOLO (You Only Look Once) algorithm, has emerged as a prominent solution in the field of product defect detection due to its efficient real-time detection capabilities and excellent classification performance. This study aims to use the YOLO algorithm to detect and classify defects in product images. By constructing and training a YOLO model, we conducted experiments on multiple industrial product datasets. The results demonstrate that this method can achieve real-time detection while maintaining high detection accuracy, significantly improving the efficiency and accuracy of product quality inspection. This paper further analyzes the advantages and limitations of the YOLO algorithm in practical applications and explores future research directions.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

39 extracted references · 31 canonical work pages

  1. [7]

    Improved YOLOv5 Based on Attention Mechanism and FasterNet for Foreign Object Detection on Railway and Airway tracks[J]

    Qi Z, Ma D, Xu J, et al. Improved YOLOv5 Based on Attention Mechanism and FasterNet for Foreign Object Detection on Railway and Airway tracks[J]. arXiv preprint arXiv:2403.08499, 2024

    arXiv 2024

  2. [1]

    Stock market analysis and prediction using LSTM: A case study on technology stocks

    Li, Zhenglin, et al. "Stock market analysis and prediction using LSTM: A case study on technology stocks." Innovations in Applied Engineering and Technology (2023): 1-6

    work page 2023

  3. [2]

    Large Language Model (LLM) AI Text Generation Detection based on Transformer Deep Learning Algorithm

    Mo, Yuhong, et al. "Large Language Model (LLM) AI Text Generation Detection based on Transformer Deep Learning Algorithm." International Journal of Engineering and Management Research 14.2 (2024): 154-159

    work page 2024

  4. [3]

    Automated pneumonia detection in chest x-ray images using deep learning model

    Li, Shaojie, Yuhong Mo, and Zhenglin Li. "Automated pneumonia detection in chest x-ray images using deep learning model." Innovations in Applied Engineering and Technology (2022): 1-6

    work page 2022

  5. [4]

    Password complexity prediction based on roberta algorithm

    Mo, Yuhong, et al. "Password complexity prediction based on roberta algorithm." Applied Science and Engineering Journal for Advanced Research 3.3 (2024): 1-5

    work page 2024

  6. [5]

    A comprehensive evaluation and comparison of enhanced learning methods

    Song, Jintong, et al. "A comprehensive evaluation and comparison of enhanced learning methods." Academic Journal of Science and Technology 10.3 (2024): 167-171

    work page 2024

  7. [6]

    Spam detection and classification based on distilbert deep learning algorithm

    Liu, Tianrui, et al. "Spam detection and classification based on distilbert deep learning algorithm." Applied Science and Engineering Journal for Advanced Research 3.3 (2024): 6-10

    work page 2024

  8. [8]

    Comparative analysis of x-ray image classification of pneumonia based on deep learning algorithm algorithm

    Ma, Danqing, et al. "Comparative analysis of x-ray image classification of pneumonia based on deep learning algorithm algorithm." Research Gate 8 (2024)

    work page 2024

Show all 39 references
  1. [9]

    A neural matrix decomposition recommender system model based on the multimodal large language model[J]

    Xiang A, Huang B, Guo X, et al. A neural matrix decomposition recommender system model based on the multimodal large language model[J]. arXiv preprint arXiv:2407.08942, 2024

    2024 arXiv

  2. [10]

    Transformer-Based Classification Outcome Prediction for Multimodal Stroke Treatment[J]

    Ma D, Wang M, Xiang A, et al. Transformer-Based Classification Outcome Prediction for Multimodal Stroke Treatment[J]. arXiv preprint arXiv:2404.12634, 2024

    2024 arXiv

  3. [11]

    A Multimodal Fusion Network For Student Emotion Recognition Based on Transformer and Tensor Product[J]

    Xiang A, Qi Z, Wang H, et al. A Multimodal Fusion Network For Student Emotion Recognition Based on Transformer and Tensor Product[J]. arXiv preprint arXiv:2403.08511, 2024

    2024 arXiv

  4. [12]

    The cloud-based design of unmanned constant temperature food delivery trolley in the context of artificial intelligence

    Dai, Shuying, et al. "The cloud-based design of unmanned constant temperature food delivery trolley in the context of artificial intelligence." Journal of Computer Technology and Applied Mathematics 1.1 (2024): 6-12

    2024

  5. [13]

    Make Scale Invariant Feature Transform “Fly

    Mo, Yuhong, et al. "Make Scale Invariant Feature Transform “Fly” with CUDA." International Journal of Engineering and Management Research 14.3 (2024): 38-45

    2024

  6. [14]

    Lidar and Monocular Sensor Fusion Depth Estimation

    He, Shuyao, et al. "Lidar and Monocular Sensor Fusion Depth Estimation." Applied Science and Engineering Journal for Advanced Research 3.3 (2024): 20-26

    2024

  7. [15]

    Unraveling large language models: From evolution to ethical implications-introduction to large language models

    Liu, Jihang, et al. "Unraveling large language models: From evolution to ethical implications-introduction to large language models." World Scientific Research Journal 10.5 (2024): 97-102

    2024

  8. [16]

    Mo, Yuhong & Zhang, Yuchen & Li, Hanzhe & Wang, Han & Yan, Xu. (2024). Prediction of heart failure patients based on multiple machine learning algorithms. Applied and Computational Engineering. 75. 1-7. 10.54254/2755-2721/75/20240498

    2024 doi

  9. [17]

    Research on Heterogeneous Computation Resource Allocation based on Data-driven Method[J]

    Tang X, Wang Z, Cai X, et al. Research on Heterogeneous Computation Resource Allocation based on Data-driven Method[J]. arXiv preprint arXiv:2408.05671, 2024

    2024 arXiv

  10. [18]

    Deep Learning Applications in the Medical Image Recognition

    Yukun, Song. "Deep Learning Applications in the Medical Image Recognition." American Journal of Computer Science and Technology 9.1 (2019): 22-26

    2019

  11. [19]

    Going Blank Comfortably: Positioning Monocular Head-Worn Displays When They are Inactive

    Song, Yukun, et al. "Going Blank Comfortably: Positioning Monocular Head-Worn Displays When They are Inactive." Proceedings of the 2023 ACM International Symposium on Wearable Computers. 2023

    2023

  12. [20]

    Looking From a Different Angle: Placing Head-Worn Displays Near the Nose

    Song, Yukun, et al. "Looking From a Different Angle: Placing Head-Worn Displays Near the Nose." Proceedings of the Augmented Humans International Conference 2024. 2024

    2024

  13. [21]

    Twitter Sentiment analysis of covid vaccines[C]//2021 5th International Conference on Artificial Intelligence and Virtual Reality (AIVR)

    Zhu W, Hu T. Twitter Sentiment analysis of covid vaccines[C]//2021 5th International Conference on Artificial Intelligence and Virtual Reality (AIVR). 2021: 118-122

    2021

  14. [22]

    Artificial intelligence aspect of transportation analysis using large scale systems[C]//Proceedings of the 2023 6th Artificial Intelligence and Cloud Computing Conference

    Hu T, Zhu W, Yan Y. Artificial intelligence aspect of transportation analysis using large scale systems[C]//Proceedings of the 2023 6th Artificial Intelligence and Cloud Computing Conference. 2023: 54-59

    2023

  15. [23]

    Zhu W. Optimizing distributed networking with big data scheduling and cloud computing[C]//International Conference on Cloud Computing, Internet of Things, and Computer Applications (CICA 2022). SPIE, 2022, 12303: 23-28

    2022

  16. [24]

    Yan Y. (2022). Influencing Factors of Housing Price in New York-analysis: Based on Excel Multi-regression Model. In Proceedings of the International Conference on Big Data Economy and Digital Management - Volume 1: BDEDM, ISBN 978-989-758-593-7, pages 1005-1009. DOI: 10.5220/0...

    2022 doi

  17. [25]

    Predicting 30-Day Hospital Readmission in Medicare Patients: Insights from an LSTM Deep Learning Model

    Li, Xintao, and Sibei Liu. "Predicting 30-Day Hospital Readmission in Medicare Patients: Insights from an LSTM Deep Learning Model." medRxiv (2024): 2024-09

    2024

  18. [26]

    Research on image generation optimization based deep learning

    Yan, Hao, et al. "Research on image generation optimization based deep learning." Proceedings of the International Conference on Machine Learning, Pattern Recognition and Automation Engineering. 2024

    2024

  19. [27]

    The application of augmented reality (ar) in remote work and education[J]

    Li K, Xirui P, Song J, et al. The application of augmented reality (ar) in remote work and education[J]. arXiv preprint arXiv:2404.10579, 2024

    2024 arXiv

  20. [28]

    Utilizing deep learning to optimize software development processes[J]

    Li K, Zhu A, Zhou W, et al. Utilizing deep learning to optimize software development processes[J]. arXiv preprint arXiv:2404.13630, 2024

    2024 arXiv

  21. [29]

    DRAL: Deep Reinforcement Adaptive Learning for Multi-UAVs Navigation in Unknown Indoor Environment

    Mo, Kangtong, et al. "DRAL: Deep Reinforcement Adaptive Learning for Multi-UAVs Navigation in Unknown Indoor Environment." arXiv preprint arXiv:2409.03930 (2024)

    2024 arXiv

  22. [30]

    Robust Domain Generalization for Multi-modal Object Recognition

    Qiao, Yuxin, et al. "Robust Domain Generalization for Multi-modal Object Recognition." arXiv preprint arXiv:2408.05831 (2024)

    2024 arXiv

  23. [31]

    Optimizing Automated Picking Systems in Warehouse Robots Using Machine Learning

    Li, Keqin, et al. "Optimizing Automated Picking Systems in Warehouse Robots Using Machine Learning." arXiv preprint arXiv:2408.16633 (2024)

    2024

  24. [32]

    Research on adaptive algorithm recommendation system based on parallel data mining platform

    Lu, Qingyi, et al. "Research on adaptive algorithm recommendation system based on parallel data mining platform." Advances in Computer, Signals and Systems 8.5 (2024): 23-33

    2024

  25. [33]

    Research on Prediction Recommendation System Based on Improved Markov Model

    Wu, Zhizhong, et al. "Research on Prediction Recommendation System Based on Improved Markov Model." Advances in Computer, Signals and Systems 8.5 (2024): 87-97

    2024

  26. [34]

    Advances in Computer, Signals and Systems (2024) Vol

    Chunyan Mao, Shuaishuai Huang, Mingxiu Sui, Haowei Yang, Xueshe Wang, Analysis and Design of a Personalized Recommendation System Based on a Dynamic User Interest Model. Advances in Computer, Signals and Systems (2024) Vol. 8: 109-118. DOI: http://dx.doi.org/10.23977/acss.2024.080513

    2024

  27. [35]

    Deep Adaptive Interest Network: Personalized Recommendation with Context-Aware Learning

    Huang, Shuaishuai, et al. "Deep Adaptive Interest Network: Personalized Recommendation with Context-Aware Learning." arXiv preprint arXiv:2409.02425 (2024)

    2024 arXiv

  28. [36]

    Nonlinear Energy Harvesting with Tools from Machine Learning

    Wang, Xueshe. Nonlinear Energy Harvesting with Tools from Machine Learning. Diss. Duke University, 2020

    2020

  29. [37]

    A model-free sampling method for basins of attraction using hybrid active learning (HAL)

    Wang, Xue-She, et al. "A model-free sampling method for basins of attraction using hybrid active learning (HAL)." Communications in Nonlinear Science and Numerical Simulation 112 (2022): 106551

    2022

  30. [38]

    Constrained attractor selection using deep reinforcement learning

    Wang, Xue-She, James D. Turner, and Brian P. Mann. "Constrained attractor selection using deep reinforcement learning." Journal of Vibration and Control 27.5-6 (2021): 502-514

    2021

  31. [39]

    Attractor selection in nonlinear energy harvesting using deep reinforcement learning

    Wang, Xue-She, and Brian P. Mann. "Attractor selection in nonlinear energy harvesting using deep reinforcement learning." arXiv preprint arXiv:2010.01255 (2020)

    2020 arXiv

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.