Pith. sign in

REVIEW 2 major objections 2 minor 14 references

Subject-Level Unknown-Identity Identification from Leap Motion Controller 2 Hand Landmarks

T0 review · 2 major / 2 minor · reviewed 2026-06-26 · grok-4.3

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

Pith's one-line read Extra Trees with enriched hand landmarks outperforms neural models at identifying known subjects and rejecting unknowns under a nested leave-one-subject-out protocol.

desk verdict The paper adds geometric hand descriptors and a nested LOSO protocol for unknown-subject rejection on Leap Motion data, but the inner threshold step may not proxy the true unknown well and no numbers are shown. read the letter →

arxiv 2606.22986 v1 pith:LZDWNXV3 submitted 2026-06-22 cs.CV cs.LG

Bahar Moharrer , Susanna Cifani , Marco Raoul Marini , Luigi Cinque , Maria De Marsico This is my paper · ORCID
classification cs.CVcs.LG
keywords handlandmarksunknownidentityidentificationopen-setrecognitionExtraTreesleave-one-subject-outgeometricdescriptorscontactlessbiometrics
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 evaluates contactless subject recognition from Leap Motion hand landmark data under a protocol that holds out one subject entirely as unknown at test time. An inner validation loop withholds a different enrolled subject only to set the rejection threshold, keeping the true unknown unseen during tuning. Geometric landmarks are augmented with fingertip-to-palm distances and palm-normalized inter-finger angles before feeding tree ensembles, centroid embeddings, and an MLP with OpenMax. Extra Trees yields the strongest combined identification and rejection performance, showing that separating unknown probes from known ones is harder than discriminating among enrolled subjects alone. The results indicate that compact, interpretable landmark descriptors can support unknown-subject rejection and identification on small-cohort datasets.

What carries the argument

The nested leave-one-subject-out protocol that uses an inner validation subject solely for threshold selection to avoid bias from the true outer unknown.

What would settle it

Applying the inner-validation threshold to the outer unknown subject produces markedly lower rejection rates than an oracle threshold tuned directly on that unknown subject.

Watch

Extended reading notes

Core claim

Under the leave-one-subject-out protocol with inner validation for threshold estimation, an Extra Trees classifier on the enriched landmark descriptors achieves the best overall balance of known-subject identification accuracy and unknown-probe rejection, outperforming both a learned embedding baseline and an MLP+OpenMax open-set model; this indicates that the dominant difficulty lies in robust score separation between known and unknown probes rather than in enrolled-subject discrimination.

Load-bearing premise

The rejection threshold estimated from an inner enrolled subject generalizes without bias to the true unknown subject held out in the outer test fold.

Editorial extensions

If this is right

  • Enriched geometric descriptors suffice for the task without requiring large-scale neural training.
  • Score separation between known and unknown probes remains the primary performance bottleneck.
  • Tree ensembles can deliver stronger unknown rejection than embedding or OpenMax alternatives under this protocol.
  • Inner validation enables threshold selection that preserves the unknown-subject guarantee.
  • Compact landmark methods become practical for contactless unknown-identity scenarios on small cohorts.

Reading between the lines

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

  • The same protocol could be applied to other landmark-based hand trackers to test device independence.
  • Adding temporal consistency checks across multiple frames might further improve rejection of unknowns.
  • Demographic variation in hand geometry could be measured by repeating the protocol on more diverse subject groups.
  • Fusion with additional modalities such as depth or color could be tested as a direct extension of the current landmark pipeline.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 2 minor

Summary. The paper studies subject-level unknown-identity identification from Leap Motion Controller 2 hand landmark data on the ML2HP dataset. It enriches geometric landmarks with fingertip-to-palm distances and palm-normalized inter-finger angles, then evaluates under a nested Leave-One-Subject-Out protocol: an outer LOSO fold holds out one subject as unknown, while an inner validation step withholds one enrolled subject solely to select the unknown-rejection threshold. Three methods are compared (Extra Trees on the descriptors, a learned embedding with centroid matching and cosine rejection, and MLP+OpenMax); the abstract states that Extra Trees is strongest overall, implying the key challenge is robust known/unknown score separation rather than enrolled discrimination alone.

Significance. If the empirical comparison holds under a validated protocol, the work demonstrates that compact, interpretable landmark descriptors can support contactless open-set hand identification on small-cohort data. The nested validation protocol is a strength, as it explicitly avoids direct tuning on the true outer unknown. Reproducible evaluation on a public dataset with multiple baselines is also a positive.

major comments (2)
  1. [Abstract / Evaluation Protocol] Abstract and Evaluation Protocol: the central claim that 'Extra Trees remains the strongest overall method' is presented without any quantitative results, error bars, dataset statistics, or per-method metrics in the abstract; the full results section must supply these numbers (including the inner-validation threshold values) to make the comparison verifiable and to confirm that the reported superiority is not an artifact of the specific operating point.
  2. [Evaluation Protocol] Evaluation Protocol (inner validation step): the threshold is chosen by withholding one enrolled subject and treating its gallery similarity scores as a proxy for unknown rejection. No analysis or diagnostic is provided showing that the impostor-score distribution of this inner withheld subject matches the distribution of the true outer unknown subject (different hand geometry and motion patterns). Because the operating point directly determines all reported metrics, any systematic mismatch undermines the claim that Extra Trees outperforms the embedding and OpenMax baselines on robust score separation.
minor comments (2)
  1. [Abstract] The abstract states the protocol 'avoids tuning on the true outer unknown' but does not name the concrete threshold-selection criterion (e.g., maximum F1 on the inner fold, fixed false-positive rate); this should be stated explicitly.
  2. [Dataset] Dataset statistics (number of subjects, samples per subject, acquisition conditions) are referenced but not quantified; these belong in a dedicated 'Dataset' subsection or table for reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. We respond point-by-point to the major comments below.

read point-by-point responses

  1. Referee: [Abstract / Evaluation Protocol] Abstract and Evaluation Protocol: the central claim that 'Extra Trees remains the strongest overall method' is presented without any quantitative results, error bars, dataset statistics, or per-method metrics in the abstract; the full results section must supply these numbers (including the inner-validation threshold values) to make the comparison verifiable and to confirm that the reported superiority is not an artifact of the specific operating point.

    Authors: We agree that the abstract states the overall finding without supporting numbers. The results section already reports the full per-method metrics under the nested LOSO protocol. We will revise the manuscript to explicitly list the inner-validation threshold values selected for each method (Extra Trees, embedding baseline, and MLP+OpenMax) and ensure all quantitative results, error bars, and dataset statistics are clearly tabulated for verifiability. revision: yes

  2. Referee: [Evaluation Protocol] Evaluation Protocol (inner validation step): the threshold is chosen by withholding one enrolled subject and treating its gallery similarity scores as a proxy for unknown rejection. No analysis or diagnostic is provided showing that the impostor-score distribution of this inner withheld subject matches the distribution of the true outer unknown subject (different hand geometry and motion patterns). Because the operating point directly determines all reported metrics, any systematic mismatch undermines the claim that Extra Trees outperforms the embedding and OpenMax baselines on robust score separation.

    Authors: The nested protocol deliberately avoids any exposure to the true outer unknown during threshold selection. We acknowledge that the original submission contains no explicit diagnostic comparing impostor-score distributions between the inner proxy and outer unknowns. We will add this analysis (e.g., distribution plots and summary statistics across folds) in the revision to allow readers to evaluate the proxy's suitability. revision: yes

Circularity Check

0 steps flagged · score 0.0 of 10

Empirical evaluation protocol contains no derivation chain or self-referential reductions

full rationale

The manuscript reports an experimental comparison of classifiers (Extra Trees, embedding baseline, MLP+OpenMax) on the ML2HP dataset under a LOSO protocol with an inner validation step for threshold selection. No equations, first-principles derivations, or parameter-fitting steps are present that could reduce any claimed prediction to its own inputs by construction. The protocol is described purely in terms of data partitioning and held-out evaluation; threshold selection uses a withheld enrolled subject as a proxy without any self-definitional or fitted-input circularity. Self-citations are absent from the provided text. This is a standard empirical ML benchmark study whose central claims rest on direct performance numbers rather than any analytic reduction.

Assumptions & free parameters 1 free parameters · 2 assumptions · 0 invented entities

The central claim rests on the assumption that the ML2HP dataset and LOSO protocol adequately represent real-world unknown-subject scenarios, plus standard supervised learning assumptions of representative sampling.

free parameters (1)
  • unknown-rejection threshold
    Selected via inner validation on withheld enrolled subject; value not reported in abstract.
assumptions (2)
  • domain assumption Inner validation subject is representative for setting rejection threshold that will apply to true unknown subjects
    Invoked to justify the nested protocol avoiding direct use of the outer unknown.
  • domain assumption Enriched geometric descriptors (fingertip-to-palm distances and normalized angles) capture identity-relevant information beyond raw landmarks
    Basis for feature engineering step described in abstract.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Subject-Level Unknown-Identity Identification from Leap Motion Controller 2 Hand Landmarks." pith.science (2026). https://pith.science/paper/LZDWNXV3

@misc{pith2026260622986,
  author       = {Pith},
  title        = {Pith review of: Subject-Level Unknown-Identity Identification from Leap Motion Controller 2 Hand Landmarks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LZDWNXV3}},
  note         = {Machine review of arXiv:2606.22986}
}
read the original abstract

This work studies subject recognition from Leap Motion Controller 2 (LMC2) hand landmark data under a subject-level unknown-identity identification protocol on the Multi View Leap2 Hand Pose (ML2HP) dataset. Using only the landmark modality, we retain the original geometric representation and enrich it with fingertip-to-palm distances and palm-normalized inter-finger angular descriptors. Evaluation is performed under a Leave-One-Subject-Out (LOSO) protocol in which, for each outer fold, one subject is excluded from the enrolled set and treated as unknown at test time. To avoid tuning on the true outer unknown subject, the unknown-rejection threshold is selected in an inner validation step by temporarily withholding one enrolled subject from the inner gallery and using it only for threshold estimation. We compare a tree ensemble baseline with two neural alternatives: a learned embedding baseline based on centroid matching and cosine-similarity-based rejection, and an MLP+OpenMax model, which represents a more established open-set recognition approach. Under this evaluation setup, Extra Trees remains the strongest overall method, indicating that the main challenge on this benchmark is not enrolled-subject discrimination alone, but robust score separation between known and unknown probes. The results support the feasibility of compact, interpretable landmark-based descriptors for contactless hand-based unknown-subject rejection and identification on a small-cohort dataset.

Figures

Figures reproduced from arXiv: 2606.22986 by the authors.

Figure 1
Figure 1. Overall architecture of the pipeline: Preprocessing, Feature Engineering, Classification. [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Unknown-score distributions for known and unknown probes under the LOSO open-set protocol: Extra Trees, using [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

14 extracted references · 4 canonical work pages

  1. [1]

    Multi-view Leap2 Hand Pose Dataset,

    M. Gil Mart ´ın, M. R. Marini, R. San Segundo Hern ´andez, and L. Cinque, “Multi-view Leap2 Hand Pose Dataset,” 2024. [Online]. Available: https://doi.org/10.21950/ZKWE6Z

    work page doi:10.21950/zkwe6z 2024

  2. [2]

    Dual leap motion controller 2: A robust dataset for multi-view hand pose recognition,

    M. Gil-Mart ´ın, M. R. Marini, R. San-Segundo, and L. Cinque, “Dual leap motion controller 2: A robust dataset for multi-view hand pose recognition,”Scientific Data, vol. 11, 10 2024

    2024

  3. [3]

    Recent advances in open set recognition: A survey,

    C. Geng, S.-J. Huang, and S. Chen, “Recent advances in open set recognition: A survey,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 43, pp. 3614–3631, 2018. [Online]. Available: https://api.semanticscholar.org/CorpusID:53734567

    2018

  4. [4]

    A survey on open-set image recognition,

    J. Sun and Q. Dong, “A survey on open-set image recognition,” 2023. [Online]. Available: https://arxiv.org/abs/2312.15571

    work page arXiv 2023

  5. [5]

    Towards open set deep networks,

    A. Bendale and T. E. Boult, “Towards open set deep networks,”2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 1563–1572, 2016. [Online]. Available: https://api.semanticscholar. org/CorpusID:14240373

    2016

  6. [6]

    Open-set recognition: A good closed-set classifier is all you need,

    S. Vaze, K. Han, A. Vedaldi, and A. Zisserman, “Open-set recognition: A good closed-set classifier is all you need,” inInternational Conference on Learning Representations, 2022. [Online]. Available: https://openreview.net/forum?id=5hLP5JY9S2d

    2022

  7. [7]

    On calibration of modern neural networks,

    C. Guo, G. Pleiss, Y . Sun, and K. Q. Weinberger, “On calibration of modern neural networks,” inProceedings of the 34th International Con- ference on Machine Learning - Volume 70, ser. ICML’17. JMLR.org, 2017, p. 1321–1330

    2017

  8. [8]

    Open-set biometrics: Beyond good closed-set models,

    Y . Su, M. Kim, F. Liu, A. Jain, and X. Liu, “Open-set biometrics: Beyond good closed-set models,” inComputer Vision – ECCV 2024, A. Leonardis, E. Ricci, S. Roth, O. Russakovsky, T. Sattler, and G. Varol, Eds. Springer Nature Switzerland, 2025, pp. 243–261

    2024

Show all 14 references
  1. [9]

    Leap motion controller for authentication via hand geometry and gestures,

    A. Chan, T. Halevi, and N. Memon, “Leap motion controller for authentication via hand geometry and gestures,” inHuman Aspects of Information Security, Privacy, and Trust, T. Tryfonas and I. Askoxylakis, Eds. Cham: Springer International Publishing, 2015, pp. 13–22

    2015

  2. [10]

    User authentication using leap motion,

    K. Maruyama, J. Shin, C. M. Kim, and C.-L. Chen, “User authentication using leap motion,” inProceedings of the International Conference on Research in Adaptive and Convergent Systems, ser. RACS ’17. New York, NY , USA: Association for Computing Machinery, 2017, p. 213–216. [On...

    2017 doi

  3. [11]

    Hand gesture authentication using optimal feature selection and dynamic time warping based k-nearest neighbor,

    J. Shin, M. A. M. Hasan, and M. Maniruzzaman, “Hand gesture authentication using optimal feature selection and dynamic time warping based k-nearest neighbor,” inProceedings of the 2022 5th International Conference on Electronics, Communications and Control Engineering, ser. IC...

    2022 doi

  4. [12]

    Dynamic hand gesture-based person identification using leap motion and machine learning approaches,

    J. Shin, M. A. M. Hasan, M. Maniruzzaman, T. Watanabe, and I. Jozume, “Dynamic hand gesture-based person identification using leap motion and machine learning approaches,”Computers, Materials and Continua, vol. 79, no. 1, pp. 1205–1222, 2024. [Online]. Available: https://www.s...

    2024

  5. [13]

    Hand gesture recognition using mediapipe landmarks and deep learning networks,

    M. Gil-Mart ´ın, M. R. Marini, I. Mart ´ın-Fern´andez, S. Esteban-Romero, and L. Cinque, “Hand gesture recognition using mediapipe landmarks and deep learning networks,” inProceedings of the 17th International Conference on Agents and Artificial Intelligence - Volume 3: ICAART...

    2025

  6. [14]

    To- wards multi-view hand pose recognition using a fusion of image em- beddings and leap 2 landmarks,

    S. Esteban-Romero, R. Lanzino, M. Marini, and M. Gil-Mart ´ın, “To- wards multi-view hand pose recognition using a fusion of image em- beddings and leap 2 landmarks,” inProceedings of the 17th International Conference on Agents and Artificial Intelligence - Volume 3: ICAART, I...

    2025

Pith tools

Reviewed June 26, 2026 · model on record in the stance chip above.