Pith. sign in

REVIEW 4 major objections 5 minor 32 references

Efficient Leaf Disease Classification and Segmentation using Midpoint Normalization Technique and Attention Mechanism

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read Attention and a tanh normalization push leaf disease accuracy to 93%

desk verdict A routine normalization-plus-SE application on a small dataset whose central segmentation claim is confounded by the paper's own ablation tables. read the letter →

arxiv 2505.21316 v1 pith:WVLHIDHK submitted 2025-05-27 cs.CV eess.IV

Enam Ahmed Taufik , Antara Firoz Parsa , Seraj Al Mahmud Mostafa This is my paper
classification cs.CVeess.IV
keywords leafdiseaseclassificationimagesegmentationMid-PointNormalizationSqueeze-and-ExcitationattentionmechanismU-NetBeteldatasetpreprocessing
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 argues that a two-stage recipe solves a practical problem: rescaling leaf images with a tanh-based 'mid-point normalization' (MPN) and adding Squeeze-and-Excitation (SE) attention blocks lets small models match or beat heavier baselines on a 1,000-image Betel leaf dataset. For classification, the proposed SE-ConvNet reaches 93% accuracy, the highest in the paper's comparisons, while remaining smaller than MobileNetV2 and VGG16. For segmentation, U-Net with the same SE blocks reaches 72.44% Dice and 58.54% IoU, up from 70.79% and 55.75% for plain U-Net. The authors take this as evidence that dataset-specific preprocessing plus channel-wise feature recalibration is a workable route to resource-constrained agricultural disease detection.

What carries the argument

The Squeeze-and-Excitation block is the load-bearing component. It takes a feature tensor X of shape H x W x C, compresses each channel to a single descriptor via global average pooling (the squeeze), passes the descriptor vector through two fully connected layers with a bottleneck and sigmoid to produce per-channel weights (the excitation), and scales the original feature maps by those weights (the recalibration). The paper inserts this block after convolutional blocks in both a custom CNN (forming SE-ConvNet) and in U-Net, so the network can emphasize disease-relevant channels. MPN supplies the normalized input that stabilizes training.

What would settle it

Run the U-Net with SE segmentation pipeline on the same data with masks re-verified by an independent annotator and with a fixed, pre-registered train/test split; if the Dice and IoU gains over plain U-Net shrink to the noise range or reverse, the claim that MPN plus SE attention improves segmentation is not supported.

Watch

Extended reading notes

Core claim

The paper's central claim is that Mid-Point Normalization (MPN), which maps pixel values to [-1,1] via tanh((resized_image/127.5)-1), and Squeeze-and-Excitation (SE) blocks are a complementary pair: MPN stabilizes the input distribution, and SE blocks let the network dynamically reweight feature channels. On the Betel leaf dataset, SE-ConvNet achieves 93% classification accuracy with per-class F1 scores of 0.89-1.00, and U-Net integrated with SE blocks achieves 72.44% Dice and 58.54% IoU, outperforming the evaluated baselines in both tasks. The authors claim this shows that lightweight attention-based models, combined with MPN preprocessing, can deliver strong disease detection without large model sizes.

Load-bearing premise

The paper assumes its 1,000-image Betel leaf dataset comes with accurate pixel-level segmentation labels and that those labels were split correctly for training and evaluation, but it never describes how the masks were created or validated.

Editorial extensions

If this is right

  • A practitioner using the same dataset and preprocessing can expect SE-ConvNet to classify Betel leaf images with about 93% accuracy, with near-perfect separation of dried and fungal classes.
  • Adding SE blocks to a plain U-Net yields a measurable segmentation gain on this dataset (about +1.7 Dice and +2.8 IoU) at similar computational cost.
  • MobileNetV2 and VGG16 trained on MPN-preprocessed images underperform the custom CNN and SE-ConvNet, suggesting that generic lightweight backbones do not automatically benefit from this normalization.
  • The combined pipeline is compact enough for deployment in resource-constrained settings: the authors report SE-ConvNet is 7.5% smaller than MobileNetV2 and 56.5% smaller than VGG16.

Reading between the lines

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

  • Because the paper never describes how segmentation masks were created or split, the reported Dice and IoU numbers should be treated as conditional on mask quality and on a proper held-out split; re-measuring with verified masks and a fixed split would sharpen the claim.
  • The same MPN plus SE recipe may transfer to other small agricultural datasets, but the paper itself shows MPN hurts MobileNetV2 and VGG16, so the benefit is architecture-dependent rather than universal.
  • An immediate testable extension is to replace the single SE head with multi-head attention, which the authors flag as a limitation, and check whether the Healthy-versus-Bacterial confusion (8 and 9 misclassifications) shrinks.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The manuscript proposes Mid-Point Normalization (MPN), a tanh-based pixel-intensity preprocessing, and combines it with Squeeze-and-Excitation (SE) blocks in two pipelines: a custom CNN (SE-ConvNet) for classification of Betel leaf diseases and a modified U-Net for segmentation of diseased regions. It reports 93% classification accuracy for SE-ConvNet and 72.44% Dice / 58.54% IoU for U-Net with SE blocks, claiming these results outperform the evaluated baselines. The paper also claims computational efficiency, explaining that SE-ConvNet is smaller than MobileNetV2 and VGG16. The central empirical claim is that MPN preprocessing plus channel-wise attention improves disease detection, but the reported experiments do not yet isolate the contributions of MPN and attention, and several dataset-protocol details are missing.

Significance. If the reported results were fully supported, the paper would offer a simple and inexpensive preprocessing-plus-attention recipe for small-scale agricultural image datasets, with lightweight models suitable for resource-constrained deployment. The topic is relevant and the architecture descriptions in Eqs. (1)-(4) are clear. However, the paper currently provides no code, no train/test split description, no error bars, no segmentation-mask provenance, and the central segmentation ablation is confounded. The strength lies in the clearly stated formulas and reproducible architecture sketch, but the empirical evidence needs substantial strengthening before the claimed attribution of gains to MPN and SE blocks is credible.

major comments (4)
  1. [Section 5, Tables 3 and 5] The segmentation ablation does not isolate the contribution of the attention mechanism. The improved model differs from the baseline in two ways: it uses MPN preprocessing and SE blocks, yet Table 5 is labeled 'U-Net' versus 'U-Net with MPN and Attention.' No U-Net+MPN-without-SE condition is reported, so the +2.78 IoU and +1.65 Dice improvements cannot be attributed to SE blocks rather than to MPN. Table 3 compounds the problem by labeling the same comparison as 'U-Net' vs. 'U-Net + Attention,' which invites exactly the causal reading that the experiments do not support. Please provide a 2x2 ablation (raw input, MPN only, SE only, MPN+SE) with per-condition IoU and Dice, and report the test protocol used for each condition.
  2. [Section 5, Table 1] The classification comparison lacks a stated train/test split, number of repetitions, and error bars, and the reported MobileNetV2 result with MPN (0.239 accuracy) is far below both its CLAHE result (0.76) and the chance level for four balanced classes (0.25). This large and unexplained drop suggests that the pretrained transfer-learning baselines are evaluated under an input distribution (tanh-scaled MPN) that is outside the ImageNet normalization they were trained for, which would make the comparison unfavorable by construction rather than by architectural merit. Please report the exact data split, per-model training hyperparameters (epochs, learning rate schedules, early-stopping criteria), and results over multiple random seeds, and consider a matched protocol in which all models are either trained from scratch or fine-tuned with the same normalization.
  3. [Section 2 and Section 4] The manuscript never describes how the pixel-level segmentation masks for the 1,000 Betel leaf images were created, validated, or split relative to the classification labels. Since segmentation is one of the two central contributions, the absence of mask provenance is load-bearing: without knowing whether the masks are manually annotated, automatically generated, noisy, or divided into held-out sets, the reported Dice and IoU values in Tables 3 and 5 are unsupported. Please specify the annotation procedure, any quality-control measures, and the exact train/validation/test partition used for the segmentation experiments, including per-class results on the held-out test set.
  4. [Section 5, Table 4] Table 4 is ambiguous: the caption reads 'Performance Metrics of Models without MPN-based Image Pre-processing,' but the table lists class-wise precision, recall, and F1 values without identifying which model produced them. As a result, the ablation narrative in the text ('MPN boosted Bacterial Leaf Disease precision from 0.76 to 0.90') cannot be verified against the table. Please label each table with the exact model, preprocessing condition, and split, so the ablation comparison is unambiguous.
minor comments (5)
  1. [Abstract and Introduction] The abstract uses the phrase 'transformative two-stage methodology' and later calls the approach 'remarkable'; these promotional terms are out of place in a technical report and should be replaced with neutral descriptions of the proposed method.
  2. [Equation (1)] The MPN formula prints as tanh(image resized / 127.5 - 1.0), which is ambiguous without parentheses; please write tanh((image/127.5) - 1.0) or the equivalent to clarify the order of operations.
  3. [Section 2 vs. Section 4] Section 2 states that images are resized to 225x225 pixels, while Section 4 states that training inputs are resized to 224x224 pixels; these numbers should be reconciled and made consistent throughout the paper.
  4. [Table 1] The table reports accuracies as decimal fractions (0.37, 0.239, 0.93) while the text refers to percentages (93%, 17%, 10%); please choose one consistent notation and apply it throughout.
  5. [Section 5, Table 2] The abstract claims a 'perfect F1 score attained for our target class,' but the target class is not defined anywhere in the paper; Table 2 shows the Dried Leaf class with F1 = 1.00, which should be stated explicitly if that is the intended reference.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: MPN and SE blocks are standard definitions and the reported metrics are direct measurements; the two self-citations are peripheral and non-load-bearing.

full rationale

The paper's central claims are empirical rather than derivational. MPN is defined by the fixed formula tanh(resized/127.5 - 1.0), which is a standardization rule and not fitted to labels or to the target accuracies. The SE block equations (2)-(4) reproduce the standard Squeeze-and-Excitation formulation from Hu et al.; no parameter is fit to the reported classification or segmentation metrics and then renamed as a prediction. The classification table and segmentation tables report held-out measurements on a fixed dataset with stated training configurations, so the headline 93% accuracy, 72.44% Dice, and 58.54% IoU do not reduce to the paper's inputs by construction. The two self-citations are not load-bearing: reference [13] is cited only as inspiration for using a Laplacian filter, and reference [25] is co-cited with the original SE paper [8] for the standard SE block; neither citation supplies the uniqueness or validity of the central experimental result. The skeptical concern that the U-Net ablation compares a baseline without MPN to a model with both MPN and SE blocks is an experimental-confound issue, not circularity: it does not show that any reported number is equivalent to an input by definition or that any fit is disguised as a prediction. Overall, the derivation chain is self-contained in the sense that the methods are standard and the results are externally checkable measurements, so no significant circularity is present.

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

The central claim relies on standard mathematical operations and domain assumptions about the dataset quality. The main unstated premises are the existence and correctness of segmentation masks and the representativeness of the reported metrics. The model hyperparameters are hand-chosen but conventional, so they add uncertainty rather than circular dependency.

free parameters (3)
  • SE bottleneck reduction ratio r
    Defined in eq. (3) but its value is never stated; it controls the capacity of the attention block and was presumably chosen by hand.
  • Learning rates = 1e-3 (classification), 1e-4 (segmentation)
    Chosen by hand, typical values, no search reported.
  • L2 regularization and dropout coefficients
    Mentioned as mitigations for overfitting but values are not reported.
assumptions (3)
  • standard math Standard definitions of tanh, global average pooling, sigmoid, and ReLU are used without proof.
    Invoked in equations (1)-(4) and Algorithm 1.
  • domain assumption The Betel leaf dataset has accurate class labels and pixel-level segmentation masks.
    The paper never describes how masks were created or verified; this is required for the segmentation claim.
  • domain assumption The reported accuracy and Dice values are representative of model performance.
    No cross-validation, seeds, or confidence intervals are provided.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Efficient Leaf Disease Classification and Segmentation using Midpoint Normalization Technique and Attention Mechanism." pith.science (2026). https://pith.science/paper/WVLHIDHK

@misc{pith2026250521316,
  author       = {Pith},
  title        = {Pith review of: Efficient Leaf Disease Classification and Segmentation using Midpoint Normalization Technique and Attention Mechanism},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WVLHIDHK}},
  note         = {Machine review of arXiv:2505.21316}
}
read the original abstract

Enhancing plant disease detection from leaf imagery remains a persistent challenge due to scarce labeled data and complex contextual factors. We introduce a transformative two-stage methodology, Mid Point Normalization (MPN) for intelligent image preprocessing, coupled with sophisticated attention mechanisms that dynamically recalibrate feature representations. Our classification pipeline, merging MPN with Squeeze-and-Excitation (SE) blocks, achieves remarkable 93% accuracy while maintaining exceptional class-wise balance. The perfect F1 score attained for our target class exemplifies attention's power in adaptive feature refinement. For segmentation tasks, we seamlessly integrate identical attention blocks within U-Net architecture using MPN-enhanced inputs, delivering compelling performance gains with 72.44% Dice score and 58.54% IoU, substantially outperforming baseline implementations. Beyond superior accuracy metrics, our approach yields computationally efficient, lightweight architectures perfectly suited for real-world computer vision applications.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

32 extracted references · 30 canonical work pages

  1. [13]

    Very deep convo- lutional networks for large-scale image recognition,

    K. Simonyan and A. Zisserman, “Very deep convo- lutional networks for large-scale image recognition,” arXiv preprint arXiv:1409.1556, 2014

    arXiv 2014

  2. [25]

    Arm-unet: attention residual path modified unet model to segment the fungal pathogen diseases in potato leaves,

    D. N. K. Pandiri, R. Murugan, and T. Goel, “Arm-unet: attention residual path modified unet model to segment the fungal pathogen diseases in potato leaves,”Signal, Image and Video Processing (SIViP), vol. 19, p. 80, 2025

    work page 2025

  3. [1]

    INTRODUCTION Image processing plays a crucial role in disease detec- tion [1–4], particularly in identifying and localizing dis- eased regions. In agriculture, effective preprocessing tech- niques are essential for enhancing data quality, improving feature extraction, and ultimately boosting the performance of classification and segmentation models. Despi...

    work page Pith review arXiv 2025

  4. [2]

    IMAGE PREPROCESSING The dataset used in this work comprises images of Betel leaves [12], categorized into four conditions: healthy, dried, bacterial leaf disease, and fungal brown spot disease (Fig. 1). It contains 1,000 images evenly distributed across these cat- egories. For image processing, we primarily employ MPN alongside preprocessing techniques su...

    work page

  5. [3]

    Femi and Mukunthan

    RELA TED WORKS Several studies have explored image classification and seg- mentation for plant disease detection. Femi and Mukunthan

    work page

  6. [4]

    We emphasize accuracy, computational efficiency, and interpretability throughout our image processing and model evaluation pipeline

    METHODOLOGY This research proposes a systematic approach to betel leaf disease classification and segmentation using deep learning models. We emphasize accuracy, computational efficiency, and interpretability throughout our image processing and model evaluation pipeline. The complete process is illus- trated in Fig. 3. For classification, we employ lightw...

    work page

  7. [5]

    RESULTS AND DISCUSSIONS In our study, we conducted experiments on classification and segmentation tasks. For classification, we evaluated Mo- bileNetV2, VGG16, CNN, and SE-ConvNet under different preprocessing techniques: Edge Detection, CLAHE, and MPN comparing results with raw data to assess their impact on accuracy. Models Resized Edge CLAHE MPN Size (...

    work page

  8. [6]

    CONCLUSION This study demonstrates the effectiveness of integrating Mid- Point Normalization (MPN) with attention mechanisms for plant disease detection. Our SE-ConvNet achieved 93% classification accuracy, outperforming traditional lightweight models by significant margins while maintaining computa- tional efficiency with a compact architecture. The inte...

    work page

Show all 32 references
  1. [7]

    Contrast limited adaptive histogram equalization,

    K. Zuiderveld, “Contrast limited adaptive histogram equalization,” inGraphics gems IV. Academic Press Professional, Inc., 1994, pp. 474–485

    1994

  2. [8]

    Re- cent advances in image processing techniques for auto- mated leaf pest and disease recognition–a review,

    L. C. Ngugi, M. Abelwahab, and M. Abo-Zahhad, “Re- cent advances in image processing techniques for auto- mated leaf pest and disease recognition–a review,”In- formation processing in agriculture, vol. 8, no. 1, pp. 27–51, 2021

    2021

  3. [9]

    A robust deep learning ap- proach for tomato plant leaf disease localization and classification,

    M. Nawaz, T. Nazir, A. Javed, M. Masood, J. Rashid, J. Kim, and A. Hussain, “A robust deep learning ap- proach for tomato plant leaf disease localization and classification,”Scientific reports, vol. 12, no. 1, p. 18568, 2022

    2022

  4. [10]

    Identification and localization of grape diseased leaf images captured by uav based on cnn,

    W. Li, X. Yu, C. Chen, and Q. Gong, “Identification and localization of grape diseased leaf images captured by uav based on cnn,”Computers and Electronics in Agri- culture, vol. 214, p. 108277, 2023

    2023

  5. [11]

    Detection of rice leaf diseases using image processing,

    M. E. Pothen and M. L. Pai, “Detection of rice leaf diseases using image processing,” in2020 fourth inter- national conference on computing methodologies and communication (ICCMC). IEEE, 2020, pp. 424–430

    2020

  6. [12]

    Mobilenetv2: Inverted residuals and lin- ear bottlenecks,

    M. Sandler, A. Howard, M. Zhu, A. Zhmoginov, and L.-C. Chen, “Mobilenetv2: Inverted residuals and lin- ear bottlenecks,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2018, pp. 4510–4520

    2018

  7. [14]

    Kumar et al

    employed machine learning for betel leaf disease clas- sification using region-of-interest algorithms for segmenta- tion and GLCM for feature extraction, achieving 97% accu- racy with an Extreme Learning Machine classifier. Kumar et al. [15] combined machine learning and deep ...

  8. [15]

    Squeeze-and-excitation networks,

    J. Hu, L. Shen, and G. Sun, “Squeeze-and-excitation networks,” inProceedings of the IEEE conference on computer vision and pattern recognition, 2018, pp. 7132–7141

    2018

  9. [16]

    Deep learning-based seg- mentation for disease identification,

    O. Mzoughi and I. Yahiaoui, “Deep learning-based seg- mentation for disease identification,”Ecological Infor- matics, vol. 75, p. 102000, 2023

    2023

  10. [17]

    A two-stage deep-learning based segmentation model for crop dis- ease quantification based on corn field imagery,

    L. Divyanth, A. Ahmad, and D. Saraswat, “A two-stage deep-learning based segmentation model for crop dis- ease quantification based on corn field imagery,”Smart Agricultural Technology, vol. 3, p. 100108, 2023

    2023

  11. [18]

    U-net: Convo- lutional networks for biomedical image segmentation,

    O. Ronneberger, P. Fischer, and T. Brox, “U-net: Convo- lutional networks for biomedical image segmentation,” inInternational Conference on Medical Image Com- puting and Computer-Assisted Intervention (MICCAI). Springer, 2015, pp. 234–241

    2015

  12. [19]

    Betel leaf image dataset from bangladesh,

    M. R. A. Rashid, M. M. Hossain, J. Biswas, and H. Majumder, “Betel leaf image dataset from bangladesh,” 2024. [Online]. Available: https://data.mendeley.com/datasets/g7fpgj57wc/2

    2024

  13. [20]

    gWaveNet: Classification of gravity waves from noisy satellite data using custom kernel integrated deep learning method,

    S. A. M. Mostafa, O. Faruque, C. Wang, J. Yue, S. Pu- rushotham, and J. Wang, “gWaveNet: Classification of gravity waves from noisy satellite data using custom kernel integrated deep learning method,” inInterna- tional Conference on Pattern Recognition. Springer, 2025, pp. 164–180

    2025

  14. [21]

    Betel leaf diseases classification using machine learning algorithm: A fea- sible approach,

    F. David and M. A. Mukunthan, “Betel leaf diseases classification using machine learning algorithm: A fea- sible approach,”Journal of Advanced Research in Ap- plied Sciences and Engineering Technology, vol. 40, no. 1, p. 7486, 2024

    2024

  15. [22]

    An inte- grated tea leaf diseases identification and retrieval model using machine learning and deep learn- ing approach,

    G. Kumar, P. Sharma, and A. Gupta, “An inte- grated tea leaf diseases identification and retrieval model using machine learning and deep learn- ing approach,” March 2024. [Online]. Available: http://dx.doi.org/10.2139/ssrn.4757484

    2024 doi

  16. [23]

    Joint plant- spraypoint detector with convnext modules and hist- match normalization,

    J. Ford, E. Sadgrove, and D. Paul, “Joint plant- spraypoint detector with convnext modules and hist- match normalization,”Precision Agriculture, vol. 26, p. 24, 2025

    2025

  17. [24]

    Recognition of early blight and late blight diseases on potato leaves based on graph cut segmenta- tion,

    C. Hou, J. Zhuang, Y . Tang, Y . He, A. Miao, H. Huang, and S. Luo, “Recognition of early blight and late blight diseases on potato leaves based on graph cut segmenta- tion,”Journal of Agriculture and Food Research, vol. 5, p. 100154, 2021

    2021

  18. [26]

    Atten-adapter: A unified attention-based adapter for ef- ficient tuning,

    K. Li, W. Gu, M. Xue, J. Xiao, D. Shi, and X. Wei, “Atten-adapter: A unified attention-based adapter for ef- ficient tuning,” in2023 IEEE International Conference on Image Processing (ICIP). IEEE, 2023

    2023

  19. [27]

    Attention toward neighbors: A context aware framework for high resolution image segmentation,

    F. F. Niloy, M. A. Amin, A. A. Ali, and A. M. Rah- man, “Attention toward neighbors: A context aware framework for high resolution image segmentation,” in 2021 IEEE International Conference on Image Process- ing (ICIP). IEEE, 2021, pp. 2279–2283

    2021

  20. [28]

    Convolutional atten- tion model for restaurant recommendation with multi- view visual features,

    H. Luo, X. Zhang, and G. Guoy, “Convolutional atten- tion model for restaurant recommendation with multi- view visual features,” in2020 IEEE International Con- ference on Image Processing (ICIP). IEEE, 2020

    2020

  21. [29]

    High-accuracy gesture recognition using mm- wave radar based on convolutional block attention mod- ule,

    Y . Song, L. Wu, Y . Zhao, P. Liu, R. Lv, and H. Ul- lah, “High-accuracy gesture recognition using mm- wave radar based on convolutional block attention mod- ule,” in2023 IEEE International Conference on Image Processing (ICIP). IEEE, 2023, pp. 1485–1489

    2023

  22. [30]

    Attention unet++: A nested attention-aware u-net for liver ct image segmentation,

    C. Li, Y . Tan, W. Chen, X. Luo, Y . Gao, X. Jia, and Z. Wang, “Attention unet++: A nested attention-aware u-net for liver ct image segmentation,” in2020 IEEE international conference on image processing (ICIP). IEEE, 2020, pp. 345–349

    2020

  23. [31]

    Fibonet: A light-weight and efficient neural network for image segmentation,

    R. Wu, X. Xiao, G. Hu, H. Zhao, H. Zhang, and Y . Peng, “Fibonet: A light-weight and efficient neural network for image segmentation,” in2023 IEEE International Conference on Image Processing (ICIP). IEEE, 2023, pp. 1345–1349

    2023

  24. [32]

    Enhancing Satellite Object Localization with Dilated Convolutions and Attention-aided Spatial Pooling,

    S. A. M. Mostafa, C. Wang, J. Yue, Y . Hozumi, and J. Wang, “Enhancing Satellite Object Localization with Dilated Convolutions and Attention-aided Spatial Pooling,”arXiv preprint arXiv:2505.05599, 2025. [Online]. Available: https://arxiv.org/abs/2505.05599

    2025 arXiv

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.