Pith. sign in

REVIEW 4 major objections 6 minor 23 references

VR-YOLO: Enhancing PCB Defect Detection with Viewpoint Robustness Based on YOLO

T0 review · 4 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read VR-YOLO, a YOLOv8 enhancement combining shear-rotation augmentation with the SIoU loss and CBAM attention, claims 98.9% mAP on PCB defect images and 94.7% under simulated viewpoint shifts of ±0.06 shear and ±10° rotation.

desk verdict A competent, incremental YOLOv8 combination for PCB defects whose headline viewpoint-robustness number is in-distribution rather than a generalization test. read the letter →

arxiv 2507.02963 v1 pith:UXGQJ6Q5 submitted 2025-06-30 cs.CV eess.IV

Hengyi Zhu , Linye Wei , He Li This is my paper
classification cs.CVeess.IV
keywords PCBdefectdetectionYOLOv8viewpointrobustnessdataaugmentationSIoUlossCBAMattentionsmallobjectaffinetransformation
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

The paper proposes VR-YOLO, an enhancement of the YOLOv8 object detector aimed at printed circuit board (PCB) defect detection. Its central claim is that a diversified scene enhancement (DSE) pipeline—shearing and rotating training images to simulate viewpoint changes, then segmenting them and adding Gaussian blur—combined with a key object focus (KOF) scheme that replaces the default CIoU loss with SIoU and inserts a CBAM attention module, makes the model substantially more tolerant of small affine viewpoint shifts. On the original test images the method reports 98.9% mean average precision (mAP50), and on test images sheared by $\pm0.06$ and rotated by $\pm 10^\circ$ it reports 94.7%, both above the baseline YOLOv8 and several recent PCB-specific YOLO variants. The practical interest is that camera angle and orientation vary in real inspection lines, and the added computation is described as negligible.

What carries the argument

The mechanism that carries the argument is the pairing of two modules. DSE is a data-augmentation stage: a 2D affine shear (horizontal and vertical coefficients up to $\pm0.06$) and a rotation up to $\pm 10^\circ$ simulate a camera viewing the board from a slightly different angle, after which $2\times2$ segmentation shrinks patch size so small defects become relatively larger, and Gaussian blur with radius 1–5 pixels mimics defocus. KOF is a detection-stage scheme: the SIoU loss adds an angle penalty term that lets the predicted box snap to the nearest axis, and the CBAM attention module, inserted into the YOLOv8 backbone, reweights feature maps along channel and spatial dimensions so faint, tiny defect regions are not suppressed. The paper's experiments attribute the gain primarily to the augmentation and loss modules, with CBAM mainly improving recall at higher IoU thresholds.

What would settle it

Run VR-YOLO on a test set with shear coefficients and rotation angles outside the training range (e.g., $\pm0.12$ shear, $\pm 20^\circ$ rotation) or with perspective distortions, and compare mAP50 against the 94.7% reported here; if accuracy drops to near the un-augmented baseline, the claimed viewpoint robustness does not extend beyond the augmentation distribution. Alternatively, test on images of real PCBs captured at physically different camera angles that were never synthetically transformed; if the gain over baseline disappears, the synthetic affine simulation is not transferring to real viewpoint variation.

Watch

Extended reading notes

Core claim

The paper's central claim is that viewpoint robustness for PCB defect detection can be obtained by training the detector on the same affine distortions it will later be tested on, and by making the box-regression loss and the backbone features more orientation-aware. Concretely, DSE applies shear coefficients of $\pm0.06$ and rotations of $\pm 10^\circ$ to every training image, multiplies the dataset threefold, then splits each image into four $640\times640$ crops and blurs some of them. KOF replaces YOLOv8's default CIoU loss with SIoU, whose angular penalty term steers predicted boxes toward the nearest axis, and adds CBAM attention to the backbone to emphasize small defect features in both channel and spatial dimensions. The reported outcome is 98.9% mAP50 on the original PCB test set and 94.7% on the viewpoint-shifted contrast test sets, with precision and recall also higher than the baseline. The authors interpret this as evidence that the method reduces the strict angle, orientation, and clarity requirements of conventional detection algorithms.

Load-bearing premise

The load-bearing premise is that the viewpoint-shifted test set is generated with the same shear and rotation operations used to augment the training set, so the reported 94.7% measures performance inside the training augmentation distribution rather than on genuinely unseen viewpoints.

Editorial extensions

If this is right

  • If the reported numbers hold, PCB inspection systems can tolerate small camera-angle changes without retraining, since the model was trained on those exact distortions.
  • The DSE augmentation component alone accounts for a large jump in mAP50 on shifted test images (from 82.7% without segmentation and blur to 94.7% with them, in the paper's Table II).
  • Swapping CIoU for SIoU speeds up training convergence, reducing both box loss and classification loss across steps.
  • The added CBAM module raises recall at higher IoU thresholds, meaning better localization consistency for small defects, not just higher detection count.
  • Because the extra modules are lightweight, the method can be deployed in near-real-time YOLO pipelines without major hardware upgrades.

Reading between the lines

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

  • The reported robustness is measured on test transforms drawn from the training augmentation family (shear $\pm0.06$, rotation $\pm 10^\circ$), so the 94.7% is best read as in-distribution robustness rather than generalization to unseen viewpoint magnitudes.
  • A stricter evaluation would apply larger rotations, combined shear and rotation at test time, or perspective/out-of-plane distortions not present in training; if accuracy on those falls back to baseline, the method's viewpoint robustness is tied to its augmentation schedule.
  • The $2\times2$ segmentation step trades away large-scale context for small-defect focus; combining it with multi-scale training or test-time augmentation could preserve both, and would be a natural next experiment.
  • The paper's 'negligible computational cost' is qualitative; reporting FLOPs, parameters, and inference FPS would let practitioners judge how the modules affect deployability in real inspection lines.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper proposes VR-YOLO, a YOLOv8-based detector for PCB defect detection, combining a diversified scene enhancement (DSE) pipeline (shear/rotation augmentation, 2x2 segmentation, and Gaussian blur) with a key object focus (KOF) scheme consisting of the SIoU loss and a CBAM attention module. On the PKU PCB defect dataset, the authors report a mAP50 of 98.9% on original test images and 94.7% on viewpoint-shifted test images (shear coefficients ±0.06, rotation angles ±10°), and claim significant improvements over baseline YOLO with negligible additional computational cost. The paper also compares against several recent YOLO variants on the raw test set.

Significance. If the central robustness claim were established, the work would be a useful practical contribution to automated PCB inspection, since viewpoint variation is a real concern in industrial imaging. The paper uses a public dataset, gives a clear description of the DSE and KOF components, and the raw-image results are competitive with published YOLO variants. The main contribution, however, is advertised as viewpoint robustness, and that claim is currently supported only by an in-distribution evaluation on the same shear/rotation transformations used to augment the training set. The ablation study does not isolate the KOF contributions, and the viewpoint comparison is presented as a figure without numeric values. For these reasons, the paper is a reasonable incremental engineering contribution, but the headline robustness result is not yet substantiated by the experimental protocol.

major comments (4)
  1. [§IV-A-3 and §IV-D-4] The viewpoint-shifted test sets (Shear003+Rotate05, Shear006+Rotate05, Shear006+Rotate10) are generated by applying the same shear and rotation operations, with the same coefficient ranges (±0.06 shear, ±10° rotation), that are used to construct the training augmentation set PCB DATASET Shear Rotate. Consequently the reported 94.7% mAP measures in-distribution performance on the training augmentation manifold, not generalization to unseen viewpoints. To support the claim of viewpoint robustness, the authors should evaluate on shear/rotation parameters outside the training ranges (e.g., ±0.12 shear or ±20° rotation) and on transformations not used in training, such as combined perspective changes or anisotropic scaling.
  2. [§IV-D-4 and Fig. 7] The only viewpoint-shift comparison between the baseline YOLOv8 and VR-YOLO is presented in Fig. 7 without numeric values, so the abstract's claim of 'significant improvements compared to the baseline YOLO model' on viewpoint-shifted images is not quantitatively substantiated. Please provide a table with Precision, Recall, mAP50, and mAP50-95 for both models on each of the four contrast test sets, including standard deviations over at least three training runs so that statistical significance can be assessed.
  3. [§IV-D-1 and Table II] The ablation in Table II compares only 'Seg+Gauss' against 'Non Seg+Gauss' and therefore isolates the segmentation and blur components of DSE, but it does not isolate the KOF contributions (SIoU loss and CBAM attention). Since the paper's main algorithmic novelty is KOF, the reader cannot tell how much of the reported improvement is due to the loss function or the attention module. Please add ablations that turn SIoU and CBAM on and off independently, ideally on both the raw and viewpoint-shifted test sets.
  4. [§IV-D-2 and Table III] The comparison in Table III reports numbers taken from prior publications (FDDC-YOLO, CDS-YOLO, DsP-YOLO, Inner-SIoU, SSHP-YOLO) that were likely obtained under different training protocols, input resolutions, augmentation schedules, or hardware. Without a common evaluation protocol and error bars, the claim that VR-YOLO 'demonstrates the best performance' is not strongly supported. Re-running the comparison methods under the same protocol, or at least reporting the variance across runs, would make the comparison fair and informative.
minor comments (6)
  1. [Table I] The header contains a typo: 'Original dateset' should be 'Original dataset'.
  2. [§IV-B] The training is described as 'capped at a maximum of 1000 steps'; it would be clearer to state whether this is epochs or iterations, and to report the batch size and learning rate schedule.
  3. [§IV-A-3] The construction of 'PCB DATASET Contrast' is described as replacing the original test images with transformed ones; please specify how many images each contrast set contains and whether the replacement is applied to the validation set, the test set, or both.
  4. [Fig. 7] The figure presents mAP comparisons without numeric annotations, making it difficult to read exact values; a table would be more appropriate, especially since the article text refers to 'significant improvements'.
  5. [§II-A] The related work section would benefit from a brief discussion of how the proposed shear/rotation augmentation differs from standard data augmentation in object detection, beyond the two biomedical examples cited.
  6. [References] Reference [18] contains an apparent typo in the dataset name 'PKU-Market-PhonePKU-Market-PCB'; please verify the correct dataset URL and name.

Circularity Check

1 steps flagged · score 6.0 of 10

The headline viewpoint-shift mAP is measured on test images generated from the same shear/rotation augmentation distribution used to train the model, so the reported 94.7% is an in-distribution result rather than evidence of generalization to unseen viewpoints.

  1. self definitional [Section IV-A-3 (training dataset construction) and Section IV-D-4 (viewpoint test sets), with the augmentation parameters stated in Section IV-D-1]
    "The training dataset PCB DATASET Shear Rotate performs one shear and one rotation operation on all images in the original dataset. ... To simulate the viewpoint shift, both horizontal and vertical shear transformations with coefficients of ±0.06 and rotation angles of ± 10◦ were applied. ... Shear006+Rotate10 (±0.06 shear coefficients, ± 10◦ rotation)."

    The viewpoint-shift test set is constructed with exactly the same shear and rotation transform family used to augment the training set. Training images are generated by applying one shear and one rotation to every original image, with the parameters given as shear coefficients up to ±0.06 and rotation angles up to ±10°. The test sets in Section IV-D-4 include Shear006+Rotate10, which falls inside that same augmentation range. Therefore the reported 94.7% mAP measures the model's performance on the distribution it was trained on, not its generalization to previously unseen viewpoint changes.

full rationale

The most significant issue is the self-referential viewpoint evaluation. Section IV-A-3 defines the augmented training set as every original image transformed by one shear and one rotation; Section IV-D-1 states those transforms are shear coefficients of ±0.06 and rotation angles of ±10°; Section IV-D-4 then evaluates on test sets including Shear006+Rotate10, i.e., the same transform family and parameter ranges. Thus the headline 94.7% result is an in-distribution measure of how well the model fits the augmentation distribution, not a test of robustness to unseen viewpoints. This is partial circularity of the central claim. The raw-image comparison in Table III is independent and supports the narrower claim that VR-YOLO improves PCB defect detection on the original PKU test set. The self-citation [8] (GS-YOLO) is present but not load-bearing, and the other components (SIoU loss, CBAM) are standard external techniques, so the score is driven by the viewpoint-test construction rather than by citation chains.

Assumptions & free parameters 4 free parameters · 3 assumptions · 0 invented entities

The central claim depends on hand-chosen hyperparameters for augmentation and network architecture, plus three domain assumptions about dataset representativeness, the validity of affine transforms as viewpoint models, and the comparability of published baselines. No new entities are introduced.

free parameters (4)
  • Viewpoint augmentation magnitudes = shear +-0.06, rotation +-10 degrees
    Hand-chosen in Section III-A and used both for training augmentation and test set construction, making the robustness evaluation in-distribution.
  • Gaussian blur radius = 1-5 pixels
    Random blur range applied in DSE (Section III-A-2); chosen to mimic focus variation, no ablation.
  • CBAM configuration = 3 layers, channels 128/256/512, window 7
    Hand-set in Section IV-B; no sensitivity analysis.
  • YOLOv8 scaling factors = depth 0.33, width 0.25, max channels 1024
    Uses YOLOv8n (nano) settings; not tuned for this task.
assumptions (3)
  • domain assumption The PKU PCB dataset is representative of real PCB defect distributions.
    All conclusions are based on 693 synthetic images from one dataset; transfer to real boards is assumed.
  • domain assumption Shear and rotation with the selected coefficients model real viewpoint variation.
    The viewpoint-robustness claim assumes these affine transforms capture practical camera angle changes; no real multi-view data is used.
  • domain assumption Comparison methods' published metrics were measured under equivalent conditions.
    Table III copies mAP from other papers without re-running under the same split, augmentation, and hyperparameters, so the superiority claim is only as valid as that assumption.

how reviews work

0 comments
Cite this review

Pith. "Pith review of VR-YOLO: Enhancing PCB Defect Detection with Viewpoint Robustness Based on YOLO." pith.science (2026). https://pith.science/paper/UXGQJ6Q5

@misc{pith2026250702963,
  author       = {Pith},
  title        = {Pith review of: VR-YOLO: Enhancing PCB Defect Detection with Viewpoint Robustness Based on YOLO},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UXGQJ6Q5}},
  note         = {Machine review of arXiv:2507.02963}
}
abstract

The integration of large-scale circuits and systems emphasizes the importance of automated defect detection of electronic components. The YOLO image detection model has been used to detect PCB defects and it has become a typical AI-assisted case of traditional industrial production. However, conventional detection algorithms have stringent requirements for the angle, orientation, and clarity of target images. In this paper, we propose an enhanced PCB defect detection algorithm, named VR-YOLO, based on the YOLOv8 model. This algorithm aims to improve the model's generalization performance and enhance viewpoint robustness in practical application scenarios. We first propose a diversified scene enhancement (DSE) method by expanding the PCB defect dataset by incorporating diverse scenarios and segmenting samples to improve target diversity. A novel key object focus (KOF) scheme is then presented by considering angular loss and introducing an additional attention mechanism to enhance fine-grained learning of small target features. Experimental results demonstrate that our improved PCB defect detection approach achieves a mean average precision (mAP) of 98.9% for the original test images, and 94.7% for the test images with viewpoint shifts (horizontal and vertical shear coefficients of $\pm 0.06$ and rotation angle of $\pm 10$ degrees), showing significant improvements compared to the baseline YOLO model with negligible additional computational cost.

Figures

Figures reproduced from arXiv: 2507.02963 by the authors.

Figure 2
Figure 2. Network structure of CBAM attention mechanism [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 4
Figure 4. Six types of defects in the dataset TABLE I: Types of images and defects in the original dateset Train Set Val Set Defect Number of Number of Number of Number of type images defects images defects Missing hole 92 398 23 99 Mouse bite 92 392 23 100 Open circuit 92 377 24 105 Short 92 382 24 109 Spur 92 391 23 97 Spurious copper 92 397 24 106 Total 552 2337 141 616 Mouse bite, Open circuit, Short, Spur, Spurious coppe… view at source ↗
Figure 5
Figure 5. Box loss and CLS loss [PITH_FULL_IMAGE:figures/full_fig_p005_5.png] view at source ↗
Figures from the paper (2 more)
Figure 6
Figure 6. Figure 6: Precision and recall curves 2) Comparative Analysis of Loss Function: The YOLOv8 algorithm integrates various metrics for evaluating loss func￾tions. By comparing the network’s convergence rate using the model’s default CIoU and the SIoU employed in this study, the var…
Figure 7
Figure 7. Figure 7: Comparison of mAP50 and mAP50-95 under the two [PITH_FULL_IMAGE:figures/full_fig_p006_7.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

23 extracted references · 19 canonical work pages

  1. [1]

    Challenges and opportunities to enable large-scale computing via heterogeneous chiplets,

    Z. Yang, S. Ji, X. Chen, J. Zhuang, W. Zhang, D. Jani, and P. Zhou, “Challenges and opportunities to enable large-scale computing via heterogeneous chiplets,” in 2024 29th Asia and South Pacific Design Automation Conference (ASP-DAC). IEEE, 2024, pp. 765–770

    work page 2024

  2. [2]

    Chip surface character detection system based on machine vision,

    C. Liao, C. Zhan, W. He, H. Gao, and P. Zhou, “Chip surface character detection system based on machine vision,” in 2024 4th International Conference on Computer Science, Electronic Information Engineering and Intelligent Control Technology (CEI) . IEEE, 2024, pp. 413–418

    work page 2024

  3. [3]

    Deep learning-enhanced defects detection for printed circuit boards,

    X.-T. Kieu, V .-T. Nguyen, D.-T. Chu, X.-H. Van, M. Van, S.-F. Su, and X.-T. Phan, “Deep learning-enhanced defects detection for printed circuit boards,” Results in Engineering , p. 104067, 2025

    work page 2025

  4. [4]

    Defect detection model of printed circuit board components based on the fusion of multi-scale features and efficient channel attention mechanism,

    W. Chen, H. Zhao, and Z. Wang, “Defect detection model of printed circuit board components based on the fusion of multi-scale features and efficient channel attention mechanism,” IEEE Access, 2024

    work page 2024

  5. [5]

    Real-time detection of insulator defects with channel pruning and channel distillation,

    D. Meng, X. Xu, Z. Jiang, and L. Xu, “Real-time detection of insulator defects with channel pruning and channel distillation,” Applied Sciences, vol. 14, no. 19, p. 8587, 2024

    work page 2024

  6. [6]

    MuSAP-GAN: printed circuit board defect detection using multi-level attention-based printed circuit board with generative adver- sarial network,

    N. Patel, “MuSAP-GAN: printed circuit board defect detection using multi-level attention-based printed circuit board with generative adver- sarial network,” Electrical Engineering, pp. 1–18, 2024

    work page 2024

  7. [7]

    Improved printed circuit board defect detection scheme,

    L. Bai and W. H. Xu, “Improved printed circuit board defect detection scheme,” Scientific Reports, vol. 15, no. 1, p. 2389, 2025

    work page 2025

  8. [8]

    GS-YOLO: A lightweight identification model for precision parts,

    H. Zhu, L. Dong, H. Ren, H. Zhuang, and H. Li, “GS-YOLO: A lightweight identification model for precision parts,” Symmetry, vol. 17, no. 2, p. 268, 2025

    work page 2025

Show all 23 references
  1. [9]

    FDDC- YOLO: an efficient detection algorithm for dense small-target solder joint defects in PCB inspection,

    H. Zheng, J. Peng, X. Yu, M. Wu, Q. Huang, and L. Chen, “FDDC- YOLO: an efficient detection algorithm for dense small-target solder joint defects in PCB inspection,” Journal of Real-Time Image Process- ing, vol. 22, no. 2, p. 83, 2025

    2025

  2. [10]

    Early diagnosis of oral cancer using image processing and artificial intelligence

    E. S. Mira, A. M. Saaduddin Sapri, R. F. Aljehanı, B. S. Jambı, T. Bashir, E.-S. M. El-Kenawy, and M. Saber, “Early diagnosis of oral cancer using image processing and artificial intelligence.” Fusion: Practice & Applications, vol. 14, no. 1, 2024

    2024

  3. [11]

    A comparative study of YOLO models for sperm and impurity detection based on proposed augmentation in small dataset,

    M. Nawae, P. Maneelert, C. Choksuchat, T. Phairatana, and J. Jaruenpun- yasak, “A comparative study of YOLO models for sperm and impurity detection based on proposed augmentation in small dataset,” in 2023 15th International Conference on Information Technology and Electrical ...

    2023

  4. [12]

    Distance-IoU loss: Faster and better learning for bounding box regression,

    Z. Zheng, P. Wang, W. Liu, J. Li, R. Ye, and D. Ren, “Distance-IoU loss: Faster and better learning for bounding box regression,” in Proceedings of the AAAI conference on artificial intelligence , vol. 34, no. 07, 2020, pp. 12 993–13 000

    2020

  5. [13]

    Focal and efficient iou loss for accurate bounding box regression,

    Y .-F. Zhang, W. Ren, Z. Zhang, Z. Jia, L. Wang, and T. Tan, “Focal and efficient iou loss for accurate bounding box regression,” Neurocomput- ing, vol. 506, pp. 146–157, 2022

    2022

  6. [14]

    SIoU loss: More powerful learning for bounding box regression,

    Z. Gevorgyan, “SIoU loss: More powerful learning for bounding box regression,” arXiv preprint arXiv:2205.12740 , 2022

    2022 arXiv

  7. [15]

    Show, attend and tell: Neural image caption generation with visual attention,

    K. Xu, J. Ba, R. Kiros, K. Cho, A. Courville, R. Salakhudinov, R. Zemel, and Y . Bengio, “Show, attend and tell: Neural image caption generation with visual attention,” in International conference on machine learning . PMLR, 2015, pp. 2048–2057

    2015

  8. [16]

    Squeeze-and-excitation networks,

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

    2018

  9. [17]

    Cbam: Convolutional block attention module,

    S. Woo, J. Park, J.-Y . Lee, and I. S. Kweon, “Cbam: Convolutional block attention module,” in Proceedings of the European conference on computer vision (ECCV) , 2018, pp. 3–19

    2018

  10. [18]

    P.-H. Lab. (2022) Pku-market-phonepku-market-pcb. [Online]. Available: https://robotics.pkusz.edu.cn/resources/dataset/

    2022

  11. [19]

    Ultralytics yolov8,

    G. Jocher, A. Chaurasia, and J. Qiu, “Ultralytics yolov8,” 2023. [Online]. Available: https://github.com/ultralytics/ultralytics

    2023

  12. [20]

    An enhanced network model for PCB defect detection: CDS-YOLO,

    M. Shao, L. Min, M. Liu, X. Li, X. Li et al. , “An enhanced network model for PCB defect detection: CDS-YOLO,” Journal of Real-Time Image Processing, vol. 21, no. 6, pp. 1–14, 2024

    2024

  13. [21]

    DsP-YOLO: An anchor-free network with DsPAN for small object detection of multiscale defects,

    Y . Zhang, H. Zhang, Q. Huang, Y . Han, and M. Zhao, “DsP-YOLO: An anchor-free network with DsPAN for small object detection of multiscale defects,” Expert Systems with Applications , vol. 241, p. 122669, 2024

    2024

  14. [22]

    Lightweight target detection: An improved YOLOv8 for small target defect detection on printed circuit boards,

    P. Du and X. Song, “Lightweight target detection: An improved YOLOv8 for small target defect detection on printed circuit boards,” in Proceedings of the 2024 International Conference on Generative Artificial Intelligence and Information Security , 2024, pp. 329–334

    2024

  15. [23]

    SSHP-YOLO: A high precision printed circuit board (PCB) defect detection algorithm with a small sample,

    J. Wang, L. Ma, Z. Li, Y . Cao, and H. Zhang, “SSHP-YOLO: A high precision printed circuit board (PCB) defect detection algorithm with a small sample,” Electronics, vol. 14, no. 2, p. 217, 2025

    2025

Pith tools

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