Pith. sign in

REVIEW 4 major objections 6 minor 21 references

Dual LiDAR-Based Traffic Movement Count Estimation at a Signalized Intersection: Deployment, Data Collection, and Preliminary Analysis

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 A dual-LiDAR setup can estimate intersection traffic movement counts by direction, maneuver, and vehicle class.

desk verdict A useful but under-validated field deployment of dual LiDARs for intersection turning-movement counts; the ground-truth labeling mismatch needs fixing before the accuracy claims can be trusted. read the letter →

arxiv 2507.13073 v1 pith:PWOBUQPR submitted 2025-07-17 eess.SY cs.CVcs.SY

Saswat Priyadarshi Nayak , Guoyuan Wu , Kanok Boriboonsomsin , Matthew Barth This is my paper · ORCID
classification eess.SYcs.CVcs.SY
keywords dualLiDARtrafficmovementcountsignalizedintersectionPointPillarszone-basedcountingvehicleclassificationroadsideturningcounts
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 reports a field deployment of two roadside LiDAR sensors at a signalized intersection and shows that a detection pipeline followed by zone-based counting produces traffic movement counts—per approach, per maneuver, and per vehicle class—that are broadly consistent with manual counts from drone footage. If the result holds, intersections could obtain turning-movement counts without cameras, manual counting, or intrusive pavement sensors, including in conditions where cameras struggle. The authors are careful to frame this as a preliminary analysis: they identify the main error sources as limited detection range, missed detections, fixed time thresholds, and a mismatch between visual class labels and length-based class labels. The practical value is that LiDAR-based TMC could feed signal timing optimization, congestion analysis, and eco-driving applications.

What carries the argument

The method is zone-based TMC estimation. Rectangular ingress zones are defined on the approach side of each movement and egress zones on the departure side; a vehicle is counted when its 3D bounding-box centroid enters a zone while the corresponding signal phase is permissible, with right-turn zones always permissible. Consecutive detections are clustered using minimum time thresholds (2 seconds for right turns, 1.2 seconds for others) to avoid counting one slow vehicle multiple times. The detections come from PointPillars, which converts each LiDAR point cloud into a pseudo-image of pillars and outputs 3D bounding boxes with position, size, and heading; the boxes are transformed from each LiDAR's sensor frame to a common local NED frame via ground-control-point georeferencing before hitting the zones.

What would settle it

Re-label the drone footage using length-based criteria (e.g., measure vehicle lengths in the video or from known model dimensions) and recompute per-class TMC errors; if the class-4 error remains around 30% instead of dropping toward the class-3 level, the LiDAR pipeline itself—not the human annotator—is the dominant error source.

Watch

Extended reading notes

Core claim

The central claim is that two infrastructure-mounted LiDARs, one at the northwest and one at the southeast corner of the intersection, combined with a PointPillars 3D detection model and a zone-based counting logic, can estimate traffic movement counts that are 'satisfactory'—close enough to manually derived drone ground truth to be used for analyzing flow, volume, and movement patterns. The paper demonstrates this for a 19-minute validation window at Baseline and Cedar in Rialto, California, reporting per-direction, per-movement, and per-class counts and their errors. The largest errors are attributed not to LiDAR detection itself but to the ground-truth annotation: medium and large SUVs whose bounding-box lengths exceeded 5 meters were visually labeled as class 3, inflating the reported error for class 4. The authors also document overcounting of heavy vehicles, which dwell in the detection zone long enough to trigger multiple counts, and higher errors on eastbound/westbound through movements, where the LiDARs' limited range and occlusion by adjacent vehicles degrade detection.

Load-bearing premise

The drone-based ground truth is treated as error-free, and visual class labels are compared directly with length-based LiDAR class labels even though the two classification schemes do not match.

Editorial extensions

If this is right

  • Signal timing optimization could use per-movement counts directly from roadside LiDAR, removing the need for temporary manual counts or camera installation at each intersection.
  • Without cameras, vehicle counting raises fewer privacy concerns, which may ease public acceptance of infrastructure-based traffic monitoring.
  • The same dual-LiDAR geometry could be extended to trajectory extraction and tracking-by-detection, yielding not just counts but speed, queue, and conflict metrics.
  • The reported error pattern—higher errors on far-side through movements and for long vehicles—shows where engineering effort should focus: extending detection range and improving temporal clustering.
  • If the ground-truth class-label mismatch is corrected, per-class accuracy for SUVs and pickup trucks may already be near the level of the dominant sedan class.

Reading between the lines

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

  • A corrected ground truth based on measured vehicle length rather than visual appearance would probably reduce the reported class-4 error, meaning the LiDAR pipeline itself may be more accurate than the headline numbers suggest.
  • The fixed 1.2–2 s thresholds are a weak point under congested or low-speed conditions; an adaptive threshold based on speed or tracking would likely cut both overcounts of heavy vehicles and missed counts of close-following vehicles.
  • The 40 m detection range and 3–5 Hz processing rate look like the real operational limits; multi-threading or a tracking model would let the system exploit the sensors' native 10 Hz and cover the intersection corners.
  • One could test the weather-resilience claim directly by running the same pipeline in rain or fog, where the paper's motivation says LiDAR should beat cameras.
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 reports a field deployment of two Ouster OS1-128 LiDARs at a signalized intersection in Rialto, California, for traffic movement count (TMC) estimation. Using a PointPillars-based detection pipeline, the authors transform detections into a common NED coordinate frame and apply a zone-based counting method with fixed time thresholds to estimate per-direction, per-movement, and per-class counts. They validate against manual counts from drone footage over a 19-minute dataset, report selected error percentages and qualitative trends, and conclude that the LiDAR-based TMC results are satisfactory. The paper also documents deployment constraints and proposes future improvements such as tracking-by-detection.

Significance. If the validation were quantitatively robust, the paper would provide a practical demonstration that infrastructure LiDAR can produce intersection turning-movement counts without cameras or manual counting, while also offering vehicle-class information and privacy advantages. The deployment details (pole heights, Ethernet length limits, dual-sensor coverage, edge-computer architecture) are genuinely useful for practitioners. However, the current evidence does not establish the central claim: no overall accuracy or error table is provided, the ground-truth labeling is internally inconsistent, and the counting thresholds appear to be tuned to the same validation dataset. These issues are addressable in a revision, which makes the underlying contribution potentially valuable rather than disqualifying.

major comments (4)
  1. [Section V.A and Section V.B.4] The ground-truth definition and the explanation of class-4 error contradict each other. Section V.A states that the drone-based manual counts are considered ground truth 'assuming that human error in visual inspection is negligible,' but Section V.B.4 attributes the 30.21% class-4 error to the human annotator mislabeling length>5 m SUVs as class 3 based on drone footage visuals. Since all per-class errors in Fig. 8b are computed against those same manual labels, the reported errors conflate LiDAR/counting errors with ground-truth labeling errors. This is load-bearing because the paper's class-level TMC claim depends on those error numbers, and class 4 represents 14.72% of the observed traffic. The ground-truth labels need independent verification or a stated correction procedure before class-level TMC errors can be interpreted.
  2. [Section IV.C and Section V] The minimum time thresholds of 2 seconds for right turns and 1.2 seconds for other movements, as well as the 8 m ingress-zone length, are introduced as empirical values and applied to the same 19-minute dataset from which they were derived. The manuscript reports no sensitivity analysis, no validation on a separate time interval, and no evidence that these thresholds generalize to other traffic conditions. Because the zone-based counting method relies critically on these thresholds to avoid double counting, the reported error rates may reflect overfitting to this particular dataset rather than the performance of a generalizable method.
  3. [Section VI opening and Section V.B] The central claim that 'The estimated TMC results using LiDARs are satisfactory' is not supported by any summary accuracy statistic, error table, or confidence interval. The paper reports only selected percentages (10.1% for class 3, 30.21% for class 4) and provides Fig. 8 plots without a tabulated per-direction, per-movement, and per-class comparison. Section V.B.5 explicitly defers quantification of errors to future work, which is reasonable for a preliminary study, but then the opening of Section VI overstates the support for the claim. A quantitative summary with overall counts, error rates, and uncertainty bounds is needed to substantiate 'satisfactory.'
  4. [Table I and Section V.A] There is a structural mismatch between the LiDAR-based class definitions and the ground-truth class definitions. LiDAR classes are assigned by bounding-box length (Table I), while ground-truth classes are assigned visually by vehicle type (Section V.A). This means that even a perfectly accurate human annotator would not necessarily agree with the length-based taxonomy for boundary vehicles such as medium and large SUVs, whose lengths straddle the 5 m threshold. The manuscript attributes the class-4 discrepancy to annotator error, but part of the discrepancy is an unavoidable artifact of comparing a continuous length measure with categorical visual labels. The authors should either align the taxonomies, for example by measuring vehicle lengths from the drone footage, or report class-agnostic counts as the primary validation metric while treating class-level results as illustrative.
minor comments (6)
  1. [Table I] The abbreviation 'FHW A' should be 'FHWA' (Federal Highway Administration).
  2. [Section IV.B] Equation (1) uses 'TEL' to denote the transformation from the LiDAR frame to ECEF, which is nonstandard notation since the subscript order usually indicates the inverse mapping; consider clarifying the notation to avoid confusion.
  3. [Section IV.C] The paper states the detection frequency is 3-5 Hz and then says consecutive frames are approximately 250 ms apart; 3-5 Hz corresponds to 200-333 ms, so the 250 ms value is only a nominal midpoint. Please state the actual mean and range of the detection interval.
  4. [Section V.A and Table II] Table II shows an excerpt for 'Northbound ...' but the ellipsis layout makes it hard to infer the full structure; consider presenting the full ground-truth table in an appendix or as a supplementary file for reproducibility.
  5. [Figure 8] The figure caption and subplot labels (a) and (b) are not described in the text beyond a generic reference; please add explicit pointers in Section V.B to the specific subplot being discussed, for example when citing the class-4 error.
  6. [Section II] Reference [9] is cited as an SSRN preprint; if a peer-reviewed version exists, it should be cited instead to strengthen the related-work discussion.

Circularity Check

0 steps flagged · score 2.0 of 10

No load-bearing circularity: LiDAR TMC counts are evaluated against independent manual drone counts; self-citations are background only.

full rationale

The paper's central claim is an empirical evaluation: dual-LiDAR detections are passed through a zone-based counting module and the resulting TMC is compared with manual counts from independent drone footage. The counting inputs (bounding-box positions, lengths, and fixed timing/zone thresholds) are not derived from the ground-truth counts, and no fitted parameter is relabeled as a prediction. The 2 s and 1.2 s time thresholds and 8 m zone length are justified from free-flow speed and observed headways, not fitted to the validation outcome. The self-citations ([9], [10], [20]) support motivation, localization/georeferencing, and the choice of PointPillars, but the paper's own detection output and separate drone reference carry the conclusion; the cited prior work is not the source of the reported counts. Section V.B.4's admission that the human annotator mislabeled large SUVs conflicts with Section V.A's assumption of negligible human error, but that is a ground-truth quality/internal-consistency issue, not a circular derivation. Accordingly, no circular step meets the quoting standard; the score reflects only minor non-load-bearing self-citation.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The central claim rests on the reliability of the pre-trained PointPillars detector, manual drone ground truth, and hand-set counting thresholds. No new physical or algorithmic entities are introduced.

free parameters (3)
  • minimum time threshold for right-turn clustering = 2 seconds
    Chosen from empirical headway observations to merge repeated detections of one vehicle; not validated on a holdout dataset.
  • minimum time threshold for other movement clustering = 1.2 seconds
    Chosen empirically from the same 19-minute dataset; affects whether a later trigger is counted as a new vehicle.
  • ingress detection zone length = 8 meters
    Set so a 20 m/s vehicle remains in the zone across 3-5 Hz frames; a design choice that directly influences count completeness.
assumptions (5)
  • domain assumption A pre-trained PointPillars detector transfers to roadside LiDAR with adequate detection accuracy.
    Section IV.A selects MMDetection3D PointPillars without reporting per-class detection metrics for this deployment.
  • domain assumption Manual visual inspection of drone footage yields negligible error and is an appropriate ground truth.
    Section V.A states 'assuming that human error in visual inspection is negligible.'
  • domain assumption Bounding box centroid and length are sufficient to infer zone crossing and vehicle class.
    Section IV.C uses only centroid-in-zone for counts, and Section V uses length bins from Table I for class labels.
  • domain assumption Signal phase and timing data used to gate zones are correct.
    Section IV.C: 'The SPaT information is used to determine permissible zones.'
  • ad hoc to paper Minimum vehicle time headway is at least 1.5 seconds in the validation data.
    Used to set the 1.2 s and 2 s clustering thresholds; stated as an empirical observation in Section IV.C, not measured or reported.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Dual LiDAR-Based Traffic Movement Count Estimation at a Signalized Intersection: Deployment, Data Collection, and Preliminary Analysis." pith.science (2026). https://pith.science/paper/PWOBUQPR

@misc{pith2026250713073,
  author       = {Pith},
  title        = {Pith review of: Dual LiDAR-Based Traffic Movement Count Estimation at a Signalized Intersection: Deployment, Data Collection, and Preliminary Analysis},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PWOBUQPR}},
  note         = {Machine review of arXiv:2507.13073}
}
read the original abstract

Traffic Movement Count (TMC) at intersections is crucial for optimizing signal timings, assessing the performance of existing traffic control measures, and proposing efficient lane configurations to minimize delays, reduce congestion, and promote safety. Traditionally, methods such as manual counting, loop detectors, pneumatic road tubes, and camera-based recognition have been used for TMC estimation. Although generally reliable, camera-based TMC estimation is prone to inaccuracies under poor lighting conditions during harsh weather and nighttime. In contrast, Light Detection and Ranging (LiDAR) technology is gaining popularity in recent times due to reduced costs and its expanding use in 3D object detection, tracking, and related applications. This paper presents the authors' endeavor to develop, deploy and evaluate a dual-LiDAR system at an intersection in the city of Rialto, California, for TMC estimation. The 3D bounding box detections from the two LiDARs are used to classify vehicle counts based on traffic directions, vehicle movements, and vehicle classes. This work discusses the estimated TMC results and provides insights into the observed trends and irregularities. Potential improvements are also discussed that could enhance not only TMC estimation, but also trajectory forecasting and intent prediction at intersections.

Figures

Figures reproduced from arXiv: 2507.13073 by the authors.

Figure 1
Figure 1. Detection results from a single LiDAR located at the [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 3
Figure 3. Dual-LiDAR deployment at Baseline and Cedar in [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 2
Figure 2. LiDAR deployment configuration and corresponding [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figures from the paper (4 more)
Figure 6
Figure 6. Figure 6: TMC calculation based on ingresses and egresses at [PITH_FULL_IMAGE:figures/full_fig_p004_6.png]
Figure 5
Figure 5. Figure 5: Sample representation of global (ECEF), local (NED), [PITH_FULL_IMAGE:figures/full_fig_p004_5.png]
Figure 8
Figure 8. Figure 8: Upon careful observation of the TMC plots, the following trends and irregularities were noted: 1) Thru movement appears to be the dominant movement across various travel directions, with the majority of traffic moving in the north-south direction. 2) The eastbound and …
Figure 8
Figure 8. Figure 8: TMC comparison: (a) Travel Directions vs. Movements, (b) Vehicle Classes vs. Travel Directions [PITH_FULL_IMAGE:figures/full_fig_p006_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

21 extracted references · 20 canonical work pages

  1. [1]

    Predicting the adoption of connected au- tonomous vehicles: A new approach based on the theory of diffusion of innovations,

    A. Talebian and S. Mishra, “Predicting the adoption of connected au- tonomous vehicles: A new approach based on the theory of diffusion of innovations,” Transportation Research Part C: Emerging Technologies, vol. 95, pp. 363–380, 10 2018

    work page 2018

  2. [2]

    Traffic monitoring and accident detection at intersections,

    S. Kamijo, Y . Matsushita, K. Ikeuchi, and M. Sakauchi, “Traffic monitoring and accident detection at intersections,” IEEE Transactions on Intelligent Transportation Systems , vol. 1, pp. 108–118, 01 2000

    work page 2000

  3. [3]

    Accuracy of pneumatic road tube counters,

    P. T. McGowen and M. A. Sanderson, “Accuracy of pneumatic road tube counters,” 2011. [Online]. Available: https://api.semanticscholar. org/CorpusID:5930786

    work page 2011

  4. [4]

    Application of cement-based piezo- electric sensors for monitoring traffic flows,

    Z.-X. Li, X.-M. Yang, and Z. Li, “Application of cement-based piezo- electric sensors for monitoring traffic flows,”Journal of Transportation Engineering-asce - J TRANSP ENG-ASCE , vol. 132, 07 2006

    work page 2006

  5. [5]

    A vehicle classification based on inductive loop detectors,

    J. Gajda, R. Sroka, M. Stencel, A. Wajda, and T. Zeglen, “A vehicle classification based on inductive loop detectors,” in IMTC 2001. Proceedings of the 18th IEEE Instrumentation and Measurement Technology Conference. Rediscovering Measurement in the Age of Informatics (Cat. No.01CH 37188) , vol. 1, 2001, pp. 460–464 vol.1

    work page 2001

  6. [6]

    Real-Time Estimation of Vehicle Counts on Signalized Intersection Approaches Using Probe Vehicle Data,

    M. A. Aljamal, H. M. Abdelghaffar, and H. A. Rakha, “Real-Time Estimation of Vehicle Counts on Signalized Intersection Approaches Using Probe Vehicle Data,” IEEE Transactions on Intelligent Trans- portation Systems , vol. 22, no. 5, pp. 2719–2729, 5 2021

    work page 2021

  7. [7]

    Com- prehensive review on vehicle Detection, classification and counting on highways,

    P. Premaratne, I. Jawad Kadhim, R. Blacklidge, and M. Lee, “Com- prehensive review on vehicle Detection, classification and counting on highways,” Neurocomputing, vol. 556, p. 126627, 11 2023

    work page 2023

  8. [8]

    Enhanced Vehicle Movement Counting at Intersections via a Self-Learning Fisheye Camera System,

    M. Adl, R. Ahmed, C. Vidal, and A. Emadi, “Enhanced Vehicle Movement Counting at Intersections via a Self-Learning Fisheye Camera System,” IEEE Access , vol. 12, pp. 77 947–77 958, 2024

    work page 2024

Show all 21 references
  1. [9]

    Cyber Mobility Mirror: A Deep Learning-based Real-World Object Perception Platform Using Roadside LiDAR,

    Z. Bai, S. Member, S. P. Nayak, X. Zhao, G. Wu, S. Member, M. J. Barth, X. Qi, Y . Liu, E. Akin Sisbot, and K. Oguchi, “Cyber Mobility Mirror: A Deep Learning-based Real-World Object Perception Platform Using Roadside LiDAR,” SSRN Electronic Journal , 2 2022. [Online]. Availab...

    2022 arXiv

  2. [10]

    Infrastructure-Assisted Cooperative State Estimation of Ego-Vehicle via Augmentation of Asynchronous Kinematic Measurements,

    S. P. Nayak, G. Wu, M. Barth, Y . Liu, E. A. Sisbot, and K. Oguchi, “Infrastructure-Assisted Cooperative State Estimation of Ego-Vehicle via Augmentation of Asynchronous Kinematic Measurements,” in Proceedings of the International Technical Meeting of The Institute of Navigati...

    2024

  3. [11]

    Pointpillars: Fast encoders for object detection from point clouds,

    A. H. Lang, S. V ora, H. Caesar, L. Zhou, J. Yang, and O. Beijbom, “Pointpillars: Fast encoders for object detection from point clouds,”

  4. [12]

    A review on traffic monitoring system techniques,

    N. K. Jain, R. K. Saini, and P. Mittal, “A review on traffic monitoring system techniques,” Advances in Intelligent Systems and Computing , vol. 742, pp. 569–577, 2019. [Online]. Available: https://link.springer.com/chapter/10.1007/978-981-13-0589-4 53

    2019 doi

  5. [13]

    Robust LIDAR localization using multiresolution Gaussian mixture maps for autonomous driving:,

    R. W. Wolcott and R. M. Eustice, “Robust LIDAR localization using multiresolution Gaussian mixture maps for autonomous driving:,” http://dx.doi.org/10.1177/0278364917696568, vol. 36, no. 3, pp. 292–319, 4 2017. [Online]. Available: https://journals.sagepub.com/ doi/full/10.117...

    2017 doi

  6. [14]

    Development and Evaluation of Traffic Count Sensor with Low-Cost Light-Detection and Ranging and Continuous Wavelet Transform: Initial Results,

    R. Jagirdar, J. Lee, K. Kim, and M. W. Kang, “Development and Evaluation of Traffic Count Sensor with Low-Cost Light-Detection and Ranging and Continuous Wavelet Transform: Initial Results,” Transportation Research Record , vol. 2673, no. 11, pp. 209–219, 11 2019. [Online]. Av...

    2019

  7. [15]

    Development and Evaluation of Intersection-Based Turning Movement Counts Frame- work Using Two Channel LiDAR Sensors,

    R. Jagirdar, J. Lee, D. Besenski, M.-W. Kang, C. Pathak, R. Jagirdar, J. Lee, D. Besenski, M.-W. Kang, and C. Pathak, “Development and Evaluation of Intersection-Based Turning Movement Counts Frame- work Using Two Channel LiDAR Sensors,” Journal of Transportation Technologies,...

    2023

  8. [16]

    Vehicle Counting and Maneuver Classifica- tion with Support Vector Machines Using Low-Density Flash Lidar,

    Z. Xie and R. Rajamani, “Vehicle Counting and Maneuver Classifica- tion with Support Vector Machines Using Low-Density Flash Lidar,” IEEE Transactions on V ehicular Technology, vol. 71, no. 1, pp. 86–97, 1 2022

    2022

  9. [17]

    A Multi- Sensor Video/LiDAR System for Analyzing Intersection Safety,

    A. Wu, T. Banerjee, K. Chen, A. Rangarajan, and S. Ranka, “A Multi- Sensor Video/LiDAR System for Analyzing Intersection Safety,”IEEE Conference on Intelligent Transportation Systems, Proceedings, ITSC , pp. 1158–1165, 2023

    2023

  10. [18]

    Evaluation of Roadside LiDAR-Based and Vision-Based Multi-Model All-Traffic Trajectory Data,

    F. Guan, H. Xu, and Y . Tian, “Evaluation of Roadside LiDAR-Based and Vision-Based Multi-Model All-Traffic Trajectory Data,” Sensors (Basel, Switzerland) , vol. 23, no. 12, p. 5377, 6 2023. [Online]. Available: https://pmc.ncbi.nlm.nih.gov/articles/PMC10300730/

    2023

  11. [19]

    Vehicle and Pedestrian Traffic Signal Performance Measures Using LiDAR-Derived Trajectory Data,

    E. D. Saldivar-Carranza, J. Desai, A. Thompson, M. Taylor, J. Sturdevant, and D. M. Bullock, “Vehicle and Pedestrian Traffic Signal Performance Measures Using LiDAR-Derived Trajectory Data,” Sensors 2024, V ol. 24, Page 6410 , vol. 24, no. 19, p. 6410, 10 2024. [Online]. Avail...

    2024

  12. [20]

    Evaluation of Infrastructure-Assisted Cooperative Tracking of Ve- hicles Using Various Motion Models,

    S. P. Nayak, G. Wu, M. Barth, Y . Liu, E. A. Sisbot, and K. Oguchi, “Evaluation of Infrastructure-Assisted Cooperative Tracking of Ve- hicles Using Various Motion Models,” 2023 IEEE/ION Position, Location and Navigation Symposium, PLANS 2023 , pp. 243–253, 2023

    2023

  13. [2019]

    Available: https://arxiv.org/abs/1812.05784

    [Online]. Available: https://arxiv.org/abs/1812.05784

    arXiv

Pith tools

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