Pith. sign in

REVIEW 3 major objections 5 minor 76 references

Reading a Ruler in the Wild

T0 review · 3 major / 5 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 RulerNet claims that reading a ruler in a photo reduces to detecting centimeter marks and fitting a geometric progression, achieving mean absolute errors of 1.19 pixels/cm on AnyRuler and 1.31 pixels/cm on Rulers2023.

desk verdict A practical, well-engineered ruler-reading system with strong empirical results, but the 'invariance' claim for the geometric-progression model is mathematically overstated and the evaluation metric partly masks perspective distortion. read the letter →

arxiv 2507.07077 v1 pith:F74XKV3C submitted 2025-07-09 cs.CV

Yimu Pan , Manas Mehta , Gwen Sincerbeaux , Jeffery A. Goldstein , Alison D. Gernand , James Z. Wang This is my paper
classification cs.CV
keywords rulerreadingscaleestimationkeypointdetectiongeometricprogressionperspectivedistortionsyntheticdataControlNetmobileinference
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 tries to establish that metric scale can be recovered from an arbitrary photographed ruler by treating it as a geometric keypoint-detection problem rather than a rigid, ruler-specific pipeline. It claims that the image positions of centimeter marks can be modeled with a geometric progression, and that fitting or regressing the progression's parameters yields a perspective-robust pixels-per-centimeter scale. On its AnyRuler test set and the Rulers2023 benchmark, the resulting system, RulerNet with DeepGP, reports mean absolute errors around 1.2 to 1.3 pixels per centimeter, far below digit-detection and mark-frequency baselines. If this holds, the work supplies a general, real-time scale estimator for applications in medicine, forensics, e-commerce, and ecology that currently need manual calibration.

What carries the argument

The central object is the geometric-progression model of ruler marks. It assumes that, after projecting detected keypoints onto the ruler line, the Euclidean distance between successive centimeter marks changes by a constant factor $r$; when $r=1$ it reduces to a linear scale. This model turns scale recovery into a constrained optimization over two initial points and $r$, which the paper solves either by differential evolution (baseline) or by DeepGP, a 1.6M-parameter 1D U-Net trained on 1.23 billion randomly generated noisy point sequences. The Hough transform groups detected marks into individual rulers and filters false positives before the progression is fit or regressed.

What would settle it

Render a synthetic straight ruler with equally spaced marks under known pinhole perspective at several oblique camera angles, run RulerNet on each render, and compare the recovered single pixels-per-centimeter value against the true local scale along the ruler; if the error grows systematically with viewing angle or the GP-fitted interval ratios deviate from the true projected ratios, the constant-ratio assumption is falsified.

Watch

Extended reading notes

Core claim

The central claim is that ruler reading can be reformulated as localizing centimeter marks as keypoints and representing the ruler by a geometric progression: for adjacent detected marks $m_i$, the next mark satisfies $D(m_{i+2}, m_{i+1}) = r\,D(m_{i+1}, m_i)$, with $r$ the common ratio. The paper argues this representation absorbs perspective foreshortening along the ruler's long axis, so a single estimated progression gives the scale across the whole ruler even in oblique views. The empirical discovery is that a heatmap-based dense predictor trained on real plus synthetic rulers, followed by a lightweight feed-forward network (DeepGP) that regresses $m_0$, $m_1$, and $r$ from noisy detections, achieves mean absolute pixel errors of 1.19 pixels/cm on AnyRuler and 1.31 pixels/cm on Rulers2023, outperforming the digit-detection baselines (56.71 and 76.05) and the mark-frequency baseline (62.11 and 14.79) by a wide margin, while running in about 42 ms per image.

Load-bearing premise

The method assumes a straight, rigid ruler whose centimeter marks, as seen in the photo, get closer together by the same multiplicative factor from one gap to the next, and that this constant-ratio description captures perspective foreshortening; if that description is wrong, the recovered scale is biased.

Editorial extensions

If this is right

  • Any straight ruler with visible centimeter marks can supply metric scale without camera calibration, fiducial markers, or ruler-specific templates.
  • Scale can still be recovered when some marks are occluded or spurious, because the geometric progression extrapolates and filters the noisy detections.
  • The DeepGP regressor removes iterative optimization at inference time, bringing scale extraction to roughly 42 ms per image and making mobile or edge deployment feasible.
  • Combined with segmentation, monocular depth, or diagnostic pipelines, the estimated scale can convert pixel measurements into physical dimensions for lesions, wounds, footprints, product images, or fish.
  • The synthetic-data pipeline combining graphics-based rulers with ControlNet-refined images reduces the need for manual ruler annotation and improves cross-dataset generalization.

Reading between the lines

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

  • Under exact pinhole projection, equally spaced world marks do not produce a constant interval ratio along a perspective-distorted ruler, so the geometric-progression assumption is an approximation rather than an invariance; its bias should grow with viewing obliqueness and could be measured by comparing GP-fitted ratios with true projective ratios on synthetic renders.
  • The same 1D point-sequence regression could transfer to other repeating structures with perspective foreshortening (text lines, tiles, fences, or biological scale bars), since the paper's DeepGP is trained purely on synthetic GP point sets.
  • The reported pixels-per-centimeter accuracy presumes the ruler and the measured object are coplanar; for objects offset from the ruler plane the scale error will depend on depth separation, which current benchmarks do not quantify.
  • The approach is tied to metric rulers with centimeter marks, but the same keypoint-plus-progression recipe should extend to inch rulers or custom mark spacings if training data with those marks are provided.
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

3 major / 5 minor

Summary. The paper presents RulerNet, a framework that treats ruler reading as a keypoint-detection problem: a dense predictor localizes centimeter marks, a Hough transform groups them into straight rulers, and scale is recovered from a geometric-progression (GP) model parameterized by two initial points and a common ratio r. The GP parameters are claimed to be invariant to perspective transformations. Scale estimation is performed either by differential-evolution optimization of a Chamfer distance or by DeepGP, a lightweight 1D U-Net trained on 1.23 billion synthetically generated noisy GP sequences. Training data combines 1,416 web images with graphics-based and ControlNet-refined synthetic rulers. On AnyRuler and Rulers2023, RulerNet-DeepGP reports mAPE/cm@768 of 1.19 and 1.31, respectively, far below the digit-detection and mark-frequency baselines, with a reported inference time of 42 ms/sample.

Significance. If the empirical results are representative, RulerNet would be a practically valuable and generalizable ruler-reading system with a concrete deployment story: it is evaluated on a held-out web split and an external benchmark (Rulers2023), the DeepGP module is small and fast, a live demo is provided, and code and data are promised. The synthetic-data pipeline, the ablation study, and the 1.23-billion-sample training of DeepGP are substantive engineering contributions. However, the central theoretical claim of exact perspective invariance is not established: the GP model is only an approximation under projective imaging, and the reported accuracy metric averages over all centimeter segments, which obscures local perspective errors. The significance of the headline numbers for oblique views therefore needs to be demonstrated rather than assumed.

major comments (3)
  1. [III-C, Eq. (3)] The claim that the geometric-progression model is 'inherently invariant to perspective distortions' does not hold for Euclidean distances under a pinhole camera. For equally spaced world marks with interval h, image coordinates along the ruler line are u_i = (a i h + b)/(c i h + d), so adjacent pixel intervals satisfy d_{i+1}/d_i = (c i h + d)/(c(i+2)h + d), which varies with i except in the affine case c=0. Thus Eq. (3) is a 3-parameter approximation to the 4-parameter projective family, and the approximation error grows with obliqueness and with the number of marks covered. This is not a wording issue: DeepGP is trained (Algorithm 1) exclusively on constant-ratio GP sequences, so it never observes the true projective structure, and the optimization in Eq. (4) will fit the wrong model under strong perspective. Please replace the invariance claim with an explicit approximation statement and quantify the induced scale bias on synthetic projective sequences with varying vanishing-point locations.
  2. [IV-A and Fig. 4] The evaluation protocol does not support the perspective-robustness claim. In AnyRuler, ground truth collapses all adjacent cm intervals of a ruler to the median pixels/cm (Section IV-A), and in Rulers2023 each ruler is reduced to a single pixels/cm value, so the reported mAPE/cm measures agreement with an average scale rather than with each mark's local scale. The Fig. 4 caption implicitly concedes this by explaining that the larger error in the third column arises from averaging the pixels/cm across all cm segments in a distorted image, despite the detected marks themselves being highly accurate. To validate perspective handling, report per-segment pixel error or adjacent-interval ratio error on oblique images, or evaluate on annotations that contain individual mark positions.
  3. [III-C, Algorithm 1 and Section V] Section V correctly identifies out-of-plane distortion as an unmodeled limitation, but the same caveat applies along the ruler's longitudinal axis. The GP model in Eq. (3) is not exact under perspective, so the training distribution of DeepGP, which contains only GP sequences with additive Gaussian noise, missing points, and spurious points, does not cover the projective curvature that appears in oblique real images. Consequently, the demonstrated robustness of DeepGP is to point noise, not to perspective deformation. Please add experiments with perspective-distorted synthetic sequences and real oblique images, or train DeepGP with projective simulation, before claiming perspective-robust scale estimation in Section VI.
minor comments (5)
  1. [Section I] The sentence 'Accurate scale estimation is bridges the gap...' contains a grammatical error ('is bridges'), and the bullet 'which enables training on a diverse array of ruler styles and imaging conditions without minimal annotation' should read 'with minimal annotation.'
  2. [Table II and Section IV-B] The metric is defined as mAPE/cm@n in Section IV-B, but Table II uses the header 'MAPE/CM@768'; please use consistent case and notation. Also, the 'ms/s' abbreviation for milliseconds per sample is confusing; 'ms/sample' would be clearer.
  3. [Algorithm 1] Line 2 initializes m2, but the loop beginning at line 4 recomputes m2 using the same formula; the initialization is redundant and can be removed to avoid confusion.
  4. [Table III] The column header 'AnyRuler Rulers2023 Avg. ms/s' does not disambiguate which of the four numeric columns correspond to AnyRuler, Rulers2023, average, and inference time; please reformat the table so that each column has a clear heading.
  5. [III-C Hough transform] The text states 'Assuming minimal camera lens distortion during inference (i.e., non-fisheye views),' which is an additional assumption not listed among the limitations in Section V; please state it explicitly in the limitations section as well.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the central accuracy claim is grounded on a held-out AnyRuler split and the external Rulers2023 benchmark, and the geometric-progression model is an explicit assumption rather than a fitted input renamed as a prediction.

full rationale

RulerNet's main result (mAPE/cm@768 of 1.19 on AnyRuler and 1.31 on Rulers2023 in Table II) is evaluated on images not used for training, and Rulers2023 is an external dataset whose labels are independent of RulerNet's training procedure. DeepGP is trained on synthetic point sequences generated by Algorithm 1, but its usefulness is established only through those external real-image evaluations; synthetic training followed by real-image testing is not a circular reduction. The geometric-progression model (Eq. 3) is introduced as an explicit modeling assumption for straight rulers under perspective, and the optimization in Eq. 4 fits m0, m1, and r to detected marks rather than reading off ground-truth labels. The scale metric compares the resulting estimate with a separately annotated pixels/cm value. Self-citations, notably [15] and [75] in the conclusion, are motivational references to the PlacentaVision project and are not load-bearing for the ruler-reading derivation. The claim that the GP is 'inherently invariant to perspective distortions' is mathematically questionable under a full pinhole model, but that is a correctness concern about the model's assumptions, not a circularity in the derivation chain. No step in the paper reduces a prediction to its own input by construction.

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

The central claim depends mainly on the GP assumption, which is stated as an invariance without derivation; it is an approximation for perspective projection. Other dependencies are standard domain assumptions acknowledged by the authors (straight rigid rulers, coplanarity with the object, visible cm marks). No invented entities are introduced. The evaluation is external, so the ledger is light aside from the GP axiom.

free parameters (4)
  • GP ratio bounds (rmin, rmax) = rmin=1/1.5, rmax=1.5
    Chosen by hand to constrain the optimization in Eq. 4; affects allowed perspective distortion.
  • Heatmap detection threshold tau and smoothing kernel (k, sigma) = not specified
    Algorithm 2 parameters for peak extraction; values not reported in the paper, affecting detected mark sets.
  • Hough transform parameters delta_theta, delta_rho = not specified
    Algorithm 3 grid resolution for line detection; chosen by hand.
  • Loss weights lambda_CE, lambda_DICE = not specified
    Eq. 2 hyperparameters, not reported in the paper.
assumptions (5)
  • domain assumption Ruler marks lie on a common straight line in the image (straight rulers, minimal lens distortion).
    Invoked before Hough grouping in Section III-C: 'Assuming minimal camera lens distortion... and that rulers are straight.'
  • ad hoc to paper Distances between adjacent centimeter marks in the image follow a constant-ratio geometric progression under perspective.
    Eq. 3 in Section III-C; asserted as perspective-invariant without derivation. For an exact pinhole projection of equally spaced world marks, the interval ratio varies with position; constant ratio is only an approximation.
  • domain assumption Centimeter marks are visible and distinguishable from mm and inch marks in typical images.
    Section III-A: the method ignores mm and inch marks and directly localizes cm marks.
  • domain assumption Ruler and target object are coplanar (or nearly so) in the image plane.
    Acknowledged in Section V Limitations: 'preassumes that both ruler and target object lie on (or near) the same physical plane.'
  • domain assumption Gaussians around different mark points do not overlap, so local maxima recovery is valid.
    Section III-C: 'assuming that the Gaussian parameters are chosen such that Gaussians from distinct points do not overlap.'

how reviews work

0 comments
Cite this review

Pith. "Pith review of Reading a Ruler in the Wild." pith.science (2026). https://pith.science/paper/F74XKV3C

@misc{pith2026250707077,
  author       = {Pith},
  title        = {Pith review of: Reading a Ruler in the Wild},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/F74XKV3C}},
  note         = {Machine review of arXiv:2507.07077}
}
read the original abstract

Accurately converting pixel measurements into absolute real-world dimensions remains a fundamental challenge in computer vision and limits progress in key applications such as biomedicine, forensics, nutritional analysis, and e-commerce. We introduce RulerNet, a deep learning framework that robustly infers scale "in the wild" by reformulating ruler reading as a unified keypoint-detection problem and by representing the ruler with geometric-progression parameters that are invariant to perspective transformations. Unlike traditional methods that rely on handcrafted thresholds or rigid, ruler-specific pipelines, RulerNet directly localizes centimeter marks using a distortion-invariant annotation and training strategy, enabling strong generalization across diverse ruler types and imaging conditions while mitigating data scarcity. We also present a scalable synthetic-data pipeline that combines graphics-based ruler generation with ControlNet to add photorealistic context, greatly increasing training diversity and improving performance. To further enhance robustness and efficiency, we propose DeepGP, a lightweight feed-forward network that regresses geometric-progression parameters from noisy marks and eliminates iterative optimization, enabling real-time scale estimation on mobile or edge devices. Experiments show that RulerNet delivers accurate, consistent, and efficient scale estimates under challenging real-world conditions. These results underscore its utility as a generalizable measurement tool and its potential for integration with other vision components for automated, scale-aware analysis in high-impact domains. A live demo is available at https://huggingface.co/spaces/ymp5078/RulerNet-Demo.

Figures

Figures reproduced from arXiv: 2507.07077 by the authors.

Figure 1
Figure 1. Representative real-world applications where accurate ruler-based [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Overview of the RulerNet pipeline. Annotators label centimeter marks (orange “+”) on rulers. A graphics-based generator synthesizes rulers on given [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Example annotations from the AnyRuler and Rulers2023 datasets. [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Qualitative examples from the AnyRuler and Rulers2023 datasets, [PITH_FULL_IMAGE:figures/full_fig_p009_4.png]
Figure 6
Figure 6. Figure 6: Comparison of the prompt words “rectangle” and “ruler” in generated [PITH_FULL_IMAGE:figures/full_fig_p010_6.png]
Figure 5
Figure 5. Figure 5: Examples of failure cases from both the AnyRuler and Rulers2023 [26] [PITH_FULL_IMAGE:figures/full_fig_p010_5.png]
Figure 7
Figure 7. Figure 7: Qualitative comparison of different ruler scale estimation methods. [PITH_FULL_IMAGE:figures/full_fig_p011_7.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

76 extracted references · 66 canonical work pages

  1. [1]

    Trismus... gnashing of the teeth,

    A. Kresha, “Trismus... gnashing of the teeth,” https://www.emdocs.net/trismus-gnashing-of-the-teeth/, Aug 2019

    work page 2019

  2. [2]

    https://www.cancer.org/cancer/types/skin-cancer/skin-cancer-image- gallery.html

    work page

  3. [3]

    https://www.crimemuseum.org/crime-library/forensic- investigation/footprints/, Aug 2023

    work page 2023

  4. [4]

    https://www.ebay.com/itm/112733250366

    work page arXiv

  5. [5]

    https://dep.nj.gov/njfw/fishing/marine/delaware-bay-finfish-trawl- survey/

    work page

  6. [6]

    Image segmentation using deep learning: A survey,

    S. Minaee, Y . Boykov, F. Porikli, A. Plaza, N. Kehtarnavaz, and D. Ter- zopoulos, “Image segmentation using deep learning: A survey,” IEEE Transactions on Pattern Analysis and Machine Intelligence , vol. 44, no. 7, pp. 3523–3542, 2021

    work page 2021

  7. [7]

    Ego4D: Around the world in 3,000 hours of egocentric video,

    K. Grauman, A. Westbury, E. Byrne, Z. Chavis, A. Furnari, R. Girdhar, J. Hamburger, H. Jiang, M. Liu, X. Liu et al., “Ego4D: Around the world in 3,000 hours of egocentric video,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2022, pp. 18 995–19 012

    work page 2022

  8. [8]

    Utility of lesion diameter in the clinical diagnosis of cutaneous melanoma,

    N. R. Abbasi, M. Yancovitz, D. Gutkowicz-Krusin, K. S. Panageas, M. C. Mihm, P. Googe, R. King, V . Prieto, I. Osman, R. J. Friedman et al., “Utility of lesion diameter in the clinical diagnosis of cutaneous melanoma,” Archives of Dermatology , vol. 144, no. 4, pp. 469–474, 2008

    work page 2008

Show all 76 references
  1. [9]

    Excision margins for nonmelanotic skin cancer,

    D. J. Thomas, A. R. King, and B. G. Peat, “Excision margins for nonmelanotic skin cancer,” Plastic and Reconstructive Surgery, vol. 112, no. 1, pp. 57–63, 2003

    2003

  2. [10]

    SizeNet: Weakly super- vised learning of visual size and fit in fashion images,

    N. Karessli, R. Guigour `es, and R. Shirvany, “SizeNet: Weakly super- vised learning of visual size and fit in fashion images,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, 2019, pp. 0–0. IEEE TRANSACTIONS ON PATTERN ANALYSIS ...

    2019

  3. [11]

    PreSizE: predicting size in e-commerce using transformers,

    Y . Eshel, O. Levi, H. Roitman, and A. Nus, “PreSizE: predicting size in e-commerce using transformers,” in Proceedings of the 44th International ACM SIGIR Conference on Research and Development in Information Retrieval , 2021, pp. 255–264

    2021

  4. [12]

    Incorporating customer re- views in size and fit recommendation systems for fashion e-commerce,

    O. Chatterjee, J. R. Tej, and N. V . Dasaraju, “Incorporating customer re- views in size and fit recommendation systems for fashion e-commerce,” arXiv preprint arXiv:2208.06261 , 2022

    2022 arXiv

  5. [13]

    Eliminate the hardware: Mobile terminals-oriented food recognition and weight estimation system,

    Q. Zhang, C. He, W. Qin, D. Liu, J. Yin, Z. Long, H. He, H. C. Sun, and H. Xu, “Eliminate the hardware: Mobile terminals-oriented food recognition and weight estimation system,” Frontiers in Nutrition, vol. 9, p. 965801, 2022

    2022

  6. [14]

    A novel approach to estimate the weight of food items based on features extracted from an image using boosting algorithms,

    F. S. Konstantakopoulos, E. I. Georga, and D. I. Fotiadis, “A novel approach to estimate the weight of food items based on features extracted from an image using boosting algorithms,” Scientific Reports , vol. 13, no. 1, p. 21040, 2023

    2023

  7. [15]

    AI-PLAX: AI-based placental assessment and examination using photos,

    Y . Chen, Z. Zhang, C. Wu, D. Davaasuren, J. A. Goldstein, A. D. Gernand, and J. Z. Wang, “AI-PLAX: AI-based placental assessment and examination using photos,” Computerized Medical Imaging and Graphics, vol. 84, p. 101744, 2020

    2020

  8. [16]

    Measuring surface area of skin lesions with 2D and 3D algorithms,

    H. Mirzaalian Dastjerdi, D. T ¨opfer, S. J. Rupitsch, and A. Maier, “Measuring surface area of skin lesions with 2D and 3D algorithms,” International Journal of Biomedical Imaging , vol. 2019, no. 1, p. 4035148, 2019

    2019

  9. [17]

    Automatic water-level detection using single-camera images with varied poses,

    Y .-T. Lin, Y .-C. Lin, and J.-Y . Han, “Automatic water-level detection using single-camera images with varied poses,” Measurement, vol. 127, pp. 167–174, 2018

    2018

  10. [18]

    A water level ruler recognition method based on deep learning technology,

    J. An, K. Song, D. Wu, and W. He, “A water level ruler recognition method based on deep learning technology,” in Proceedings of the International Conference on Intelligent Computing . Springer, 2023, pp. 40–50

    2023

  11. [19]

    Soil profile ruler detection in YOLOv5,

    X. Yan, Q. Pan, and Y . Zhao, “Soil profile ruler detection in YOLOv5,” in Proceedings of the International Conference on Generative Artificial Intelligence and Information Security , 2024, pp. 324–328

    2024

  12. [20]

    Automatic estimation of a scale resolution in forensic images,

    I. Gertsovich, M. Nilsson, J. Bart ˚unˇek, and I. Claesson, “Automatic estimation of a scale resolution in forensic images,” Forensic Science International, vol. 283, pp. 58–71, 2018

    2018

  13. [21]

    A ruler detection method for auto-adjusting scales of shoeprint images,

    Z. Wen, R. M. Smith, M. Connor, and J. Curran, “A ruler detection method for auto-adjusting scales of shoeprint images,”Science & Justice, vol. 64, no. 5, pp. 498–508, 2024

    2024

  14. [22]

    Gpt-4o system card,

    A. Hurst, A. Lerer, A. P. Goucher, A. Perelman, A. Ramesh, A. Clark, A. Ostrow, A. Welihinda, A. Hayes, A. Radford et al., “Gpt-4o system card,” arXiv preprint arXiv:2410.21276 , 2024

    2024 arXiv

  15. [23]

    Research on image measuring method based on monocular camera and ruler,

    S. Jun, “Research on image measuring method based on monocular camera and ruler,” in Proceedings of the Conference on Informatization in Education, Management and Business . Atlantis Press, 2015, pp. 150–155

    2015

  16. [24]

    Heuristic analysis for in-plane non-contact calibration of rulers using mask R-CNN,

    M. Telahun, D. Sierra-Sossa, and A. S. Elmaghraby, “Heuristic analysis for in-plane non-contact calibration of rulers using mask R-CNN,” Information, vol. 11, no. 5, p. 259, 2020

    2020

  17. [25]

    Method of automatic identification of ruler scales on images to solve the problem of estimating the geometric dimensions of certain objects,

    B. Lukashchuk, “Method of automatic identification of ruler scales on images to solve the problem of estimating the geometric dimensions of certain objects,” in Proceedings of the IEEE 3rd KhPI Week on Advanced Technology. IEEE, 2022, pp. 1–4

    2022

  18. [26]

    Rulers2023: An annotated dataset of synthetic and real images for ruler detection using deep learning,

    D. Matuzevi ˇcius, “Rulers2023: An annotated dataset of synthetic and real images for ruler detection using deep learning,” Electronics, vol. 12, no. 24, p. 4924, 2023

    2023

  19. [27]

    Ruler detection for automatic scaling of fish images,

    D. A. Konovalov, J. A. Domingos, C. Bajema, R. D. White, and D. R. Jerry, “Ruler detection for automatic scaling of fish images,” in Proceedings of the International Conference on Advances in Image Processing, 2017, pp. 90–95

    2017

  20. [28]

    Ruler detection for autoscaling forensic images,

    A. Bhalerao and G. Reynolds, “Ruler detection for autoscaling forensic images,” International Journal of Digital Crime and Forensics , vol. 6, no. 1, pp. 9–27, 2014

    2014

  21. [29]

    Image-based measure- ment of ancient coins,

    M. Herrmann, S. Zambanini, and M. Kampel, “Image-based measure- ment of ancient coins,” in Proceedings of the International Conference on Computer Applications and Quantitative Methods in Archaeology . Oxford: Archaeopress, 2010

    2010

  22. [30]

    In-situ water level measurement using nir-imaging video camera,

    Z. Zhang, Y . Zhou, H. Liu, and H. Gao, “In-situ water level measurement using nir-imaging video camera,” Flow Measurement and Instrumenta- tion, vol. 67, pp. 95–106, 2019

    2019

  23. [31]

    Method on water level ruler reading recognition based on image processing,

    G. Chen, K. Bai, Z. Lin, X. Liao, S. Liu, Z. Lin, Q. Zhang, and X. Jia, “Method on water level ruler reading recognition based on image processing,” Signal, Image and Video Processing , vol. 15, pp. 33–41, 2021

    2021

  24. [32]

    An intelligent water level monitoring method based on SSD algorithm,

    G. Bai, J. Hou, Y . Zhang, B. Li, H. Han, T. Wang, R. Hinkelmann, D. Zhang, and L. Guo, “An intelligent water level monitoring method based on SSD algorithm,” Measurement, vol. 185, p. 110047, 2021

    2021

  25. [33]

    Research on water-level recognition method based on image processing and convolutional neural networks,

    G. Dou, R. Chen, C. Han, Z. Liu, and J. Liu, “Research on water-level recognition method based on image processing and convolutional neural networks,” Water, vol. 14, no. 12, p. 1890, 2022

    2022

  26. [34]

    A two-stage image process for water level recognition via dual-attention CornerNet and CTransformer,

    R. Qiu, Z. Cai, Z. Chang, S. Liu, and G. Tu, “A two-stage image process for water level recognition via dual-attention CornerNet and CTransformer,” The Visual Computer , vol. 39, no. 7, pp. 2933–2952, 2023

    2023

  27. [35]

    A novel depth measurement method for urban flooding based on surveillance video images and a floating ruler,

    S. Liu, W. Zheng, X. Wang, H. Xiong, J. Cheng, C. Yong, W. Zhang, and X. Zou, “A novel depth measurement method for urban flooding based on surveillance video images and a floating ruler,” Natural Hazards, vol. 119, no. 3, pp. 1967–1989, 2023

    1967

  28. [36]

    Ruler-based automatic stitching of spatially overlapping ra- diographs,

    A. Gooßen, M. Schl ¨uter, M. Hensel, T. Pralow, and R.-R. Gri- gat, “Ruler-based automatic stitching of spatially overlapping ra- diographs,” in Bildverarbeitung f ¨ur die Medizin 2008: Algorith- men—Systeme—Anwendungen Proceedings des Workshops vom 6. bis

    2008

  29. [37]

    Springer, 2008, pp

    April 2008 in Berlin . Springer, 2008, pp. 192–196

    2008

  30. [38]

    Ruler based automatic C-arm image stitching without overlapping constraint,

    C. Chen, R. Kojcev, D. Haschtmann, T. Fekete, L. Nolte, and G. Zheng, “Ruler based automatic C-arm image stitching without overlapping constraint,” Journal of Digital Imaging , vol. 28, pp. 474–480, 2015

    2015

  31. [39]

    Extraction of ruler markings for estimating physical size of oral lesions,

    Z. Xue, K. Yu, P. Pearlman, T.-C. Chen, C.-H. Hua, C. J. Kang, C.-Y . Chien, M.-H. Tsai, C.-P. Wang, A. Chaturvedi et al., “Extraction of ruler markings for estimating physical size of oral lesions,” in Proceedings of the International Conference on Pattern Recognition . IEEE,...

    2022

  32. [40]

    Automatic detection of oral lesion measurement ruler toward computer-aided image-based oral cancer screening,

    Z. Xue, K. Yu, P. C. Pearlman, A. Pal, T.-C. Chen, C.-H. Hua, C. J. Kang, C.-Y . Chien, M.-H. Tsai, C.-P. Wang et al., “Automatic detection of oral lesion measurement ruler toward computer-aided image-based oral cancer screening,” in Proceedings of the Annual International Con...

    2022

  33. [41]

    Comprehensive exploration of synthetic data generation: A survey,

    A. Bauer, S. Trapp, M. Stenger, R. Leppich, S. Kounev, M. Leznik, K. Chard, and I. Foster, “Comprehensive exploration of synthetic data generation: A survey,” arXiv preprint arXiv:2401.02524 , 2024

    2024 arXiv

  34. [42]

    The synthia dataset: A large collection of synthetic images for semantic segmentation of urban scenes,

    G. Ros, L. Sellart, J. Materzynska, D. Vazquez, and A. M. Lopez, “The synthia dataset: A large collection of synthetic images for semantic segmentation of urban scenes,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition , 2016, pp. 3234–3243

    2016

  35. [43]

    Virtual worlds as proxy for multi-object tracking analysis,

    A. Gaidon, Q. Wang, Y . Cabon, and E. Vig, “Virtual worlds as proxy for multi-object tracking analysis,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition , 2016, pp. 4340–4349

    2016

  36. [44]

    Playing for data: Ground truth from computer games,

    S. R. Richter, V . Vineet, S. Roth, and V . Koltun, “Playing for data: Ground truth from computer games,” in Proceedings of the European Conference on Computer Vision . Springer, 2016, pp. 102–118

    2016

  37. [45]

    Mathematical models of oversaturated protein spots,

    D. Matuzevi ˇcius, A. Serackis, and D. Navakauskas, “Mathematical models of oversaturated protein spots,” Elektronika ir elektrotechnika , vol. 73, no. 1, pp. 63–68, 2007

    2007

  38. [46]

    Towards a simulation driven stereo vision system,

    M. Peris, S. Martull, A. Maki, Y . Ohkawa, and K. Fukui, “Towards a simulation driven stereo vision system,” in Proceedings of the Interna- tional Conference on Pattern Recognition. IEEE, 2012, pp. 1038–1042

    2012

  39. [47]

    Image-based synthesis and re-synthesis of viewpoints guided by 3D models,

    K. Rematas, T. Ritschel, M. Fritz, and T. Tuytelaars, “Image-based synthesis and re-synthesis of viewpoints guided by 3D models,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2014, pp. 3898–3905

    2014

  40. [48]

    Semantic pose using deep networks trained on synthetic RGB-D,

    J. Papon and M. Schoeler, “Semantic pose using deep networks trained on synthetic RGB-D,” in Proceedings of the IEEE International Con- ference on Computer Vision , 2015, pp. 774–782

    2015

  41. [49]

    Paralleleye pipeline: An effective method to synthesize images for improving the visual intelligence of intelligent vehicles,

    X. Li, K. Wang, X. Gu, F. Deng, and F.-Y . Wang, “Paralleleye pipeline: An effective method to synthesize images for improving the visual intelligence of intelligent vehicles,” IEEE Transactions on Systems, Man, and Cybernetics: Systems , vol. 53, no. 9, pp. 5545–5556, 2023

    2023

  42. [50]

    Data augmentation generative adversarial networks,

    A. Antoniou, “Data augmentation generative adversarial networks,” arXiv preprint arXiv:1711.04340 , 2017

    2017 arXiv

  43. [51]

    Data augmentation in emotion classification using generative adversarial networks,

    X. Zhu, Y . Liu, Z. Qin, and J. Li, “Data augmentation in emotion classification using generative adversarial networks,” arXiv preprint arXiv:1711.00648, 2017

    2017 arXiv

  44. [52]

    GAM augmentation: Augmenting training data using generative adversarial networks,

    C. Bowles, L. Chen, R. Guerrero, P. Bentley, R. Gunn, A. Hammers, D. A. Dickie, M. V . Hern ´andez, J. Wardlaw, and D. Rueckert, “GAM augmentation: Augmenting training data using generative adversarial networks,” arXiv preprint arXiv:1810.10863 , 2018

    arXiv

  45. [53]

    Medical image synthesis for data augmentation and anonymization using generative ad- versarial networks,

    H.-C. Shin, N. A. Tenenholtz, J. K. Rogers, C. G. Schwarz, M. L. Senjem, J. L. Gunter, K. P. Andriole, and M. Michalski, “Medical image synthesis for data augmentation and anonymization using generative ad- versarial networks,” in Proceedings of the Third International Worksho...

    2018

  46. [54]

    Emotion classification with data augmentation using generative adversarial networks,

    X. Zhu, Y . Liu, J. Li, T. Wan, and Z. Qin, “Emotion classification with data augmentation using generative adversarial networks,” in Pro- ceedings of the Advances in Knowledge Discovery and Data Mining . Springer, 2018, pp. 349–360

    2018

  47. [55]

    Red blood cell image generation for data augmentation using conditional generative adversarial networks,

    O. Bailo, D. Ham, and Y . Min Shin, “Red blood cell image generation for data augmentation using conditional generative adversarial networks,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, 2019, pp. 0–0

    2019

  48. [56]

    Data augmentation using generative adversarial networks (CycleGAN) to improve generalizability in ct segmentation tasks,

    V . Sandfort, K. Yan, P. J. Pickhardt, and R. M. Summers, “Data augmentation using generative adversarial networks (CycleGAN) to improve generalizability in ct segmentation tasks,” Scientific Reports , vol. 9, no. 1, p. 16884, 2019

    2019

  49. [57]

    Data augmentation using generative adversarial networks (GANs) for GAN-based detection of pneumonia and covid-19 in chest x-ray images,

    S. Motamed, P. Rogalla, and F. Khalvati, “Data augmentation using generative adversarial networks (GANs) for GAN-based detection of pneumonia and covid-19 in chest x-ray images,” Informatics in Medicine Unlocked, vol. 27, p. 100779, 2021

    2021

  50. [58]

    Ef- fective data augmentation with diffusion models,

    B. Trabucco, K. Doherty, M. Gurinas, and R. Salakhutdinov, “Ef- fective data augmentation with diffusion models,” arXiv preprint arXiv:2302.07944, 2023

    2023 arXiv

  51. [59]

    Diverse data augmentation with diffusions for effective test-time prompt tuning,

    C.-M. Feng, K. Yu, Y . Liu, S. Khan, and W. Zuo, “Diverse data augmentation with diffusions for effective test-time prompt tuning,” in Proceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 2704–2714

    2023

  52. [60]

    Diffusion model-based data augmentation for lung ultrasound classification with limited data

    X. Zhang, A. Gangopadhyay, H.-M. Chang, and R. Soni, “Diffusion model-based data augmentation for lung ultrasound classification with limited data.” in Proceedings of the Machine Learning for Health Workshop at the Conference on Neural Information Processing Systems , 2023, pp...

    2023

  53. [61]

    Synthetic data augmentation by diffusion probabilistic models to enhance weed recognition,

    D. Chen, X. Qi, Y . Zheng, Y . Lu, Y . Huang, and Z. Li, “Synthetic data augmentation by diffusion probabilistic models to enhance weed recognition,” Computers and Electronics in Agriculture , vol. 216, p. 108517, 2024

    2024

  54. [62]

    Data augmentation for object detection via controllable diffusion models,

    H. Fang, B. Han, S. Zhang, S. Zhou, C. Hu, and W.-M. Ye, “Data augmentation for object detection via controllable diffusion models,” in Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, 2024, pp. 1257–1266

    2024

  55. [63]

    DreamDA: Generative data augmentation with diffusion models,

    Y . Fu, C. Chen, Y . Qiao, and Y . Yu, “DreamDA: Generative data augmentation with diffusion models,” arXiv preprint arXiv:2403.12803, 2024

    2024 arXiv

  56. [64]

    Dif- fuseMix: Label-preserving data augmentation with diffusion models,

    K. Islam, M. Z. Zaheer, A. Mahmood, and K. Nandakumar, “Dif- fuseMix: Label-preserving data augmentation with diffusion models,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 27 621–27 630

    2024

  57. [65]

    Advances in diffusion models for image data augmentation: A review of methods, models, evaluation metrics and future research directions,

    P. Alimisis, I. Mademlis, P. Radoglou-Grammatikis, P. Sarigiannidis, and G. T. Papadopoulos, “Advances in diffusion models for image data augmentation: A review of methods, models, evaluation metrics and future research directions,” arXiv preprint arXiv:2407.04103 , 2024

    2024 arXiv

  58. [66]

    Deep high-resolution representation learning for visual recognition,

    J. Wang, K. Sun, T. Cheng, B. Jiang, C. Deng, Y . Zhao, D. Liu, Y . Mu, M. Tan, X. Wang et al. , “Deep high-resolution representation learning for visual recognition,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 43, no. 10, pp. 3349–3364, 2020

    2020

  59. [67]

    ImageNet: A large-scale hierarchical image database,

    J. Deng, W. Dong, R. Socher, L.-J. Li, K. Li, and L. Fei-Fei, “ImageNet: A large-scale hierarchical image database,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition . IEEE, 2009, pp. 248–255

    2009

  60. [68]

    Adding conditional control to text-to-image diffusion models,

    L. Zhang, A. Rao, and M. Agrawala, “Adding conditional control to text-to-image diffusion models,” in Proceedings of the IEEE/CVF International Conference on Computer Vision , 2023, pp. 3836–3847

    2023

  61. [69]

    Near- linear time algorithm for the chamfer distance,

    A. Bakshi, P. Indyk, R. Jayaram, S. Silwal, and E. Waingarten, “Near- linear time algorithm for the chamfer distance,” Advances in Neural Information Processing Systems , vol. 36, pp. 66 833–66 844, 2023

    2023

  62. [70]

    Differential evolution–a simple and efficient heuristic for global optimization over continuous spaces,

    R. Storn and K. Price, “Differential evolution–a simple and efficient heuristic for global optimization over continuous spaces,” Journal of Global Optimization, vol. 11, pp. 341–359, 1997

    1997

  63. [71]

    Mobilenetv4: universal models for the mobile ecosystem,

    D. Qin, C. Leichner, M. Delakis, M. Fornoni, S. Luo, F. Yang, W. Wang, C. Banbury, C. Ye, B. Akin et al., “Mobilenetv4: universal models for the mobile ecosystem,” in Proceedings of the European Conference on Computer Vision. Springer, 2024, pp. 78–96

    2024

  64. [72]

    U-net: Convolutional networks for biomedical image segmentation,

    O. Ronneberger, P. Fischer, and T. Brox, “U-net: Convolutional networks for biomedical image segmentation,” in Proceedings of the Medical Image Computing and Computer-Assisted Intervention. Springer, 2015, pp. 234–241

    2015

  65. [73]

    Understanding hallu- cinations in diffusion models through mode interpolation,

    S. K. Aithal, P. Maini, Z. Lipton, and J. Z. Kolter, “Understanding hallu- cinations in diffusion models through mode interpolation,” Advances in Neural Information Processing Systems , vol. 37, pp. 134 614–134 644, 2025

    2025

  66. [74]

    Camera calibration with lens distortion from low-rank textures,

    Z. Zhang, Y . Matsushita, and Y . Ma, “Camera calibration with lens distortion from low-rank textures,” in CVPR 2011 . IEEE, 2011, pp. 2321–2328

    2011

  67. [75]

    Gyroflow+: Gyroscope-guided unsupervised deep homography and optical flow learning,

    H. Li, K. Luo, B. Zeng, and S. Liu, “Gyroflow+: Gyroscope-guided unsupervised deep homography and optical flow learning,” International Journal of Computer Vision , vol. 132, no. 6, pp. 2331–2349, 2024

    2024

  68. [76]

    Cross-modal contrastive learning for unified placenta analysis using photographs,

    Y . Pan, M. Mehta, J. A. Goldstein, J. Ngonzi, L. M. Bebell, D. J. Roberts, C. K. Carreon, K. Gallagher, R. E. Walker, A. D. Gernand et al. , “Cross-modal contrastive learning for unified placenta analysis using photographs,” Patterns, vol. 5, no. 12, 2024. Yimu Pan is a Ph.D....

    1994

Pith tools

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