Pith. sign in

REVIEW 4 major objections 4 minor 24 references

Applying Adversarial Auto-encoder for Estimating Human Walking Gait Abnormality Index

T0 review · 4 major / 4 minor · reviewed 2026-08-14 · deepseek-v4-flash

Pith's one-line read An adversarial auto-encoder trained only on normal walking can score gait abnormality from 3D point clouds, reaching a 4.6% equal error rate in leave-one-out evaluation.

desk verdict A solid, modest gait-index paper whose headline comparison is undermined by test-set-tuned hyperparameters and a missing AE baseline, but it deserves a careful review rather than a desk reject. read the letter →

arxiv 1908.06188 v1 pith:3LDL2SCP submitted 2019-08-16 cs.CV

classification cs.CV
keywords adversarialauto-encodergaitqualityindexanomalydetectionone-classclassification3Dpointcloudcylindricalhistogramdepthcamerawalking
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper tries to show that an adversarial auto-encoder (AAE) can serve as a one-class gait quality index when trained only on normal walking. The input is a 3D point cloud of a person on a treadmill, captured by one time-of-flight depth camera with two mirrors and converted into a 16x16 cylindrical histogram. The AAE produces three partial scores: reconstruction error, likelihood under a prior, and discriminator output, and the paper argues that their weighted combination is a better abnormality index than any single score. On nearly 100,000 point clouds, the combination reaches equal error rates of 0.063 for a 5:4 subject split and 0.046 for leave-one-out cross-validation in whole-sequence assessment, improving on reimplemented methods that use silhouettes, skeletons, or depth maps. If correct, this makes GAN-family models useful not only as generators but as unsupervised anomaly scorers.

What carries the argument

The central object is the cylindrical histogram: each 3D point cloud is wrapped in a cylinder split into 16x16 equal-volume sectors, flattened to a 2D array, scaled to [0,1], and quantized to 256 levels. This fixed-size representation feeds a three-part AAE: an encoder, a decoder, and a discriminator trained with a GAN-style minimax loss. Abnormality is detected as deviation from normal training postures: high reconstruction error, low prior density, or low discriminator output. The final index is a weighted sum of the three normalized measures, where each weight is inversely proportional to the average training value of that measure, and the prior-density term is raised to a small exponent (u = 1/8) to avoid degrading the combination.

What would settle it

Collect a dataset where every single frame of an abnormal walk is posturally identical to a normal frame but the sequence has an abnormal rhythm, such as asymmetric step timing; if the AAE's per-frame scores, averaged over segments, cannot separate it from normal gait, the per-frame scoring assumption fails.

Watch

Extended reading notes

Core claim

The central claim is that the reconstruction-oriented parts of an AAE, rather than its generation ability, carry a usable signal of walking gait abnormality. Trained only on normal gait postures, the AAE yields three measures: the reconstruction error between input and output, the normalized prior density of the latent code, and the discriminator's probability that the latent code is normal. Combining these as a weighted sum with weights derived from training values gives a per-frame abnormality index, and averaging over frames gives segment and sequence indices. The best combination, using all three measures, has an EER of 0.063 in the 5:4 split and 0.046 in leave-one-out for whole-sequence assessment, the lowest among the compared approaches.

Load-bearing premise

The paper scores each individual posture point cloud independently and then averages those scores, so it assumes gait abnormality shows up in individual postures rather than only in the timing or rhythm of the walk.

Editorial extensions

If this is right

  • A markerless, single-camera gait monitor becomes feasible: the depth-plus-mirrors setup avoids body-worn sensors and camera synchronization.
  • GAN-family models can be repurposed from sample generators to input-oriented anomaly scorers by combining partial-network outputs.
  • Longer observation windows improve the index: the reported curves show EER decreasing as segment length grows.
  • The method generalizes across subjects under leave-one-out evaluation, with whole-sequence EER dropping to 0.046.

Reading between the lines

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

  • Because the method is unsupervised and needs only normal samples, the same cylinder-histogram-plus-AAE scoring could transfer to other one-class posture assessments such as fall-risk screening or rehabilitation monitoring; the paper does not test these, but nothing in the pipeline is gait-specific.
  • The temporal dimension is deliberately left out of the model; an AAE consuming short stacks of histograms might capture cadence and symmetry abnormalities that the current averaging cannot, but that would be an extension beyond the paper.
  • The heuristic weight rule and the manually chosen exponent for the prior-density term suggest that a validation-based calibration of the combination could change the relative ranking of the three measures; the paper reports only the ad hoc choice.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 4 minor

Summary. The paper proposes an unsupervised method for estimating a human walking gait quality index from depth point clouds using an adversarial auto-encoder (AAE). Each point cloud is converted into a 16×16 cylindrical histogram, which is fed to an AAE trained only on normal gait postures. Three measures are extracted: reconstruction error (Υ_AE), the probability that the latent code is drawn from the prior (Υ_P), and the discriminator output (Υ_D). These are combined via a weighted sum with weights computed from the training set. The method is evaluated on a dataset of 9 subjects and 9 gait types (normal plus artificially induced abnormalities), using a 5:4 subject split and leave-one-out cross-validation. The authors report EERs for per-frame, per-segment, and per-sequence assessment, and compare with reimplemented silhouette-, skeleton-, and depth-based methods. They claim the AAE-based indices, especially the combination Υ_AE+Υ_P+Υ_D, achieve the lowest EER.

Significance. If the results are validated, the paper would make a useful contribution: a simple, one-class, unsupervised system that produces a usable gait quality index from 3D point clouds, without requiring manual labels or multi-camera calibration. The strengths include the large dataset (nearly 100,000 point clouds), the use of a standard reconstruction-error anomaly score, the clear pipeline from raw depth to cylindrical histogram, and the consistent observation that averaging over longer segments improves performance. The paper also honestly notes several limitations, such as the arbitrary histogram size and the lack of a principled stopping criterion. However, the central comparative claim - that the proposed combination outperforms related approaches - is weakened by post-hoc hyperparameter selection on the test set and by the absence of a plain auto-encoder baseline.

major comments (4)
  1. [Section 3.3, Eq. (5), Fig. 6, Table 2] The exponent u=1/8 in Eq. (5) was selected 'after considering some small values' directly on the 5:4 test split whose EERs are reported in Table 2. Because the improvement of Υ_AE+Υ_P+Υ_D over Υ_AE alone is small (0.081 to 0.063 per-sequence in the 5:4 split and 0.049 to 0.046 in leave-one-out), the reported margin could plausibly be an artifact of test-set-driven model selection rather than a genuine property of the combination. Please re-evaluate with u chosen on a validation fold, or report EER as a function of u on a truly held-out test set.
  2. [Section 3.2, Fig. 4] The training-stopping criterion is also selected post hoc: the text states that the losses were unstable after epoch 370 and 'a range of 200-300 was thus selected,' and each of the nine leave-one-out models uses a different stability-based epoch range. This means the reported EERs are averages over epoch ranges chosen after inspecting the same data used for evaluation. In addition, no variance or error bars are reported for Fig. 5-7 or Table 2, so it is impossible to tell whether the small EER differences between combinations are statistically meaningful. Please report standard deviations over training runs and over folds, and justify the epoch-selection procedure without reference to test performance.
  3. [Table 2] No plain auto-encoder baseline is reported. Since Υ_AE alone already achieves low EER (0.081 and 0.049 per-sequence) and the adversarial losses are known to be unstable, the comparison in Table 2 cannot establish that the adversarial components (Υ_P, Υ_D, and the GAN training) provide any benefit over a standard AE trained with the same architecture and reconstruction loss. Please include an AE-only baseline trained with the same encoder/decoder but without the discriminator or prior loss.
  4. [Section 2 (Gait Index Estimation)] The proposed index is measured per point-cloud frame and then averaged over segments, so the method only captures posture-based deviations from the training distribution. Temporal gait characteristics such as cadence, left-right asymmetry in swing phase, and step-length variability are not modeled. Since the paper claims a general 'walking gait quality index' but evaluates only on abnormalities that alter instantaneous posture (sole thickness and ankle weight), the generalization of the method to temporal gait disorders remains unsubstantiated. Please state this limitation explicitly and, ideally, evaluate on a dataset with temporal abnormalities.
minor comments (4)
  1. [Abstract and Section 1] The terms 'Gait Abnormality Index' and 'Gait Quality Index' are used interchangeably across the abstract and the main text; please align the terminology for consistency.
  2. [Eq. (5)] The notation Υ_X is used for the combined index, but the individual terms Υ_AE, Υ_P, and Υ_D are defined in prose rather than in the equation; please add explicit definitions or a notational glossary.
  3. [Table 2] The per-frame entries for HMM-based methods are shown as dashes without explanation; a footnote clarifying why those methods cannot be evaluated per frame would help the reader compare fairly.
  4. [Figures 5-7] Adding error bars or shaded confidence intervals would greatly improve interpretability, since the differences between combinations in Fig. 6 are small and no measure of variance is provided.

Circularity Check

1 steps flagged · score 6.0 of 10

Headline EERs partly reflect test-set selection: the 3-measure combination and u=1/8 were chosen after seeing Fig. 6 on the same 5:4 split, then reported as the predicted winner in Table 2.

  1. fitted input called prediction [Section 3.3 'Experimental Results', paragraph after Fig. 6; Table 2 (5:4 separation rows)]
    "According to Fig. 6 (where u = 1/8 after considering some small values), improving ΥAE by both ΥD and ΥP is recommended since its results were the best compared with the other combinations."

    The paper first evaluates all combinations of ΥAE, ΥP, ΥD on the 5:4 test split (Fig. 6) and, after seeing those test-based EER/AUC curves, fixes the exponent u=1/8 and decides that adding ΥP and ΥD is 'recommended.' The same test split then produces the headline per-sequence EERs in Table 2 (ΥAE+ΥP+ΥD: 0.063). Thus the reported 'best' result is the argmin of the same test EERs that were used to select the configuration; the claimed prediction is selected on the data used to verify it, so the headline comparison is not an out-of-sample prediction. The gain over ΥAE alone (0.081 to 0.063) is small, making the selection responsible for the margin.

full rationale

The core index is not circular: ΥAE is a standard reconstruction RMSE from an AAE trained only on normal gait point clouds, and the per-subject split evaluation is an honest test of an unsupervised anomaly score. The weights in Eq. (5) are computed from training statistics, not labels. The self-citations ([12], [14]) supply dataset, reconstruction method, and a combination weighting idea, but the formula is restated in the paper and is not load-bearing. However, the central comparative claim -- that ΥAE+ΥP+ΥD is the best index -- is partially circular: the exponent u=1/8 and the decision to include both ΥP and ΥD were made after inspecting EER curves computed on the same 5:4 test set that is later used to report the winning EERs in Table 2. Because the improvement over ΥAE alone is small, test-set-driven model selection could account for the reported advantage. The leave-one-out results inherit u=1/8 from the 5:4 selection, so they are not fully independent either, although they are less directly circular. This is a partial circularity/selection artifact affecting the headline result, not a fully circular derivation; score 6.

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

All free parameters are representation or architecture choices; the most consequential is the exponent u, which was tuned on test labels. The axioms are the usual one-class anomaly assumptions for autoencoders plus the specifics of the cylindrical histogram and the Gaussian prior. No invented entities are needed. The net effect is that the method is a plausibly engineered extension of a standard AAE, with the evaluation partially optimized in-sample.

free parameters (4)
  • Cylindrical histogram grid size = 16 x 16 (256 bins)
    Paper states 'the selected size of cylindrical histogram is 16x16... this is just an arbitrary choice, not necessarily the optimal one' (Section 2.1). The representation and all derived scores depend on this resolution.
  • Hidden layer width = 96 units
    Paper states 'a similar hidden layer of (experimentally selected) 96 units' (Section 2.2). No ablation justifies this width.
  • Latent dimension = 16
    The AAE latent space z has 16 dimensions (Section 2.2); the reconstruction error, prior probability, and discriminator score all operate on this bottleneck. No systematic search is reported.
  • Exponent u on the prior-probability measure = 1/8
    Paper: 'where u = 1/8 after considering some small values' (Section 3.3). This exponent was chosen after evaluating test-set EERs for several small values, so the reported improvement of the combined index is partly a fitted result.
assumptions (4)
  • ad hoc to paper The 16x16 cylindrical histogram is a sufficient representation of gait posture for abnormality detection.
    Introduced in Section 2.1. The histogram discards exact point positions and shape details; whether it preserves the subtle asymmetries in pathological gaits is not tested. If it does not, the whole pipeline cannot detect those abnormalities.
  • domain assumption An AAE trained only on normal gaits produces larger reconstruction errors, lower prior probabilities, and lower discriminator outputs for abnormal gaits.
    Section 2.3 relies on this one-class assumption. It is plausible for autoencoders generally, but the paper does not analyze cases where an abnormal posture (e.g., with a moderate 5 cm sole) still lies within the normal latent manifold.
  • domain assumption The prior distribution P is a multivariate normal with zero mean and scalar covariance, and the normalized density f_s is a meaningful anomaly score.
    Section 2.2, Eq. (2)-(5). The choice of prior is standard for AAEs, but the paper does not test sensitivity to the prior's variance, which directly affects Υ_P.
  • domain assumption The dataset labels are correct, the 8 abnormal conditions are representative of gait abnormality, and the training split contains only normal gaits.
    Section 3.1. The one-class training and the EER evaluation both depend on label quality. The paper does not discuss possible label noise or whether the induced impairments reflect clinical gait abnormalities.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Applying Adversarial Auto-encoder for Estimating Human Walking Gait Abnormality Index." pith.science (2026). https://pith.science/paper/3LDL2SCP

@misc{pith2026190806188,
  author       = {Pith},
  title        = {Pith review of: Applying Adversarial Auto-encoder for Estimating Human Walking Gait Abnormality Index},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3LDL2SCP}},
  note         = {Machine review of arXiv:1908.06188}
}
read the original abstract

This paper proposes an approach that estimates human walking gait quality index using an adversarial auto-encoder (AAE), i.e. a combination of auto-encoder and generative adversarial network (GAN). Since most GAN-based models have been employed as data generators, our work introduces another perspective of their application. This method directly works on a sequence of 3D point clouds representing the walking postures of a subject. By fitting a cylinder onto each point cloud and feeding obtained histograms to an appropriate AAE, our system is able to provide different measures that may be used as gait quality indices. The combinations of such quantities are also investigated to obtain improved indicators. The ability of our method is demonstrated by experimenting on a large dataset of nearly 100 thousands point clouds and the results outperform related approaches that employ different input data types.

Figures

Figures reproduced from arXiv: 1908.06188 by the authors.

Figure 1
Figure 1. Data acquisition of our system: (a) a depth map showing our setup that includes a tread [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. The remainder of this paper is organized as follows: Section 2 describes the processing flow of our approach; the experiments on a large dataset and a comparison with related methods are given in Section 3; Section 4 presents the conclusion together with possible extensions that may improve the current work. 2 PROPOSED METHOD As presented in [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 2
Figure 2. A typical AAE where X and Xb are respectively an input and its reconstruction result provided by the AE, z is the representation of X in latent space, P is a predefined prior distribution that draws samples z˜, l + and l − respectively indicate the assigning of positive and negative labels, and p is the probability that an input is real, i.e. its label is positive (l +). The operation ∪ represents the union of label… view at source ↗
Figures from the paper (5 more)
Figure 3
Figure 3. Figure 3: Illustration of estimating a cylindrical histogram: (a) a cylinder, that contains 16 equal [PITH_FULL_IMAGE:figures/full_fig_p003_3.png]
Figure 4
Figure 4. Figure 4: The change of AAE losses during first 500 training epochs. The training set includes [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: The average AUCs and EERs of the three partial measures estimated on segments of [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: The average AUCs and EERs of ΥAE’s possible combinations estimated with different segment lengths. The AAE was evaluated according to the suggested 5:4 separation. In order to enhance ΥAE using the other measures, we attempted to perform some combinations. We observed …
Figure 7
Figure 7. Figure 7: The average AUCs and EERs estimated in the leave-one-out evaluation stage. The dis [PITH_FULL_IMAGE:figures/full_fig_p009_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

24 extracted references · 8 canonical work pages

  1. [1]

    Auvinet, J

    E. Auvinet, J. Meunier, and F. Multon. Multiple depth cameras calibration and body vol- ume reconstruction for gait analysis. In Information Science, Signal Processing and their Applications (ISSPA), 2012 11th International Conference on , pp. 478–483, July 2012. doi: 10.1109/ISSPA.2012.6310598

  2. [2]

    Automatic detection of abnormal gait

    Christian Bauckhage, John K Tsotsos, and Frank E Bunn. Automatic detection of abnormal gait. Image and Vision Computing, 27(1):108–115, 2009

  3. [3]

    Amini Maghsoud Bigy, K

    A. Amini Maghsoud Bigy, K. Banitsas, A. Badii, and J. Cosmas. Recognition of postures and freezing of gait in parkinson’s disease patients using microsoft kinect sensor. In 2015 7th International IEEE/EMBS Conference on Neural Engineering (NER) , pp. 731–734, April

  4. [4]

    Hierarchical motion history images for recognizing human motion

    James W Davis. Hierarchical motion history images for recognizing human motion. In Detec- tion and Recognition of Events in Video, 2001. Proceedings. IEEE Workshop on , pp. 39–46. IEEE, 2001

  5. [5]

    Generative adversarial nets

    Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. Generative adversarial nets. In Z. Ghahramani, M. Welling, C. Cortes, N. D. Lawrence, and K. Q. Weinberger (eds.), Advances in Neural In- formation Processing Systems 27, pp. 2672–2680. Curran Associates, Inc., 2014. URLhttp: //...

  6. [6]

    Individual recognition using gait energy image

    Ju Han and Bir Bhanu. Individual recognition using gait energy image. IEEE Transactions on Pattern Analysis and Machine Intelligence , 28(2):316–322, Feb 2006. ISSN 0162-8828. doi: 10.1109/TPAMI.2006.38

  7. [7]

    Real Time Gait Recognition System Based on Kinect Skeleton Feature, pp

    Shuming Jiang, Yufei Wang, Yuanyuan Zhang, and Jiande Sun. Real Time Gait Recognition System Based on Kinect Skeleton Feature, pp. 46–57. Springer International Publishing, Cham,

  8. [8]

    Adam: A method for stochastic optimization

    Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980, 2014

Show all 24 references
  1. [9]

    doi: 10.1007/978-3-319-16628-5 4

    ISBN 978-3-319-16628-5. doi: 10.1007/978-3-319-16628-5 4. 11 This is a pre-print of an article published in Pattern Analysis and Applications. The final authenticated version is available online at: https://doi.org/10.1007/s10044-019-00790-7

  2. [10]

    Adversarial au- toencoders

    Alireza Makhzani, Jonathon Shlens, Navdeep Jaitly, and Ian Goodfellow. Adversarial au- toencoders. In International Conference on Learning Representations , 2016. URL http: //arxiv.org/abs/1511.05644

  3. [11]

    L ´opez-Fern´andez, F.J

    D. L ´opez-Fern´andez, F.J. Madrid-Cuevas, A. Carmona-Poyato, R. Mu noz Salinas, and R. Medina-Carnicer. A new approach for multi-view gait recognition on unconstrained paths. Journal of Visual Communication and Image Representation , 38:396 – 406, 2016. ISSN 1047-3203. doi: h...

  4. [12]

    T. N. Nguyen, H. H. Huynh, and J. Meunier. Assessment of gait normality using a depth camera and mirrors. In 2018 IEEE EMBS International Conference on Biomedical Health Informatics (BHI), pp. 37–41, Las Vegas, NV , USA, March 2018. doi: 10.1109/BHI.2018.8333364

  5. [13]

    Electric power system anomaly detection using neural networks

    Marco Martinelli, Enrico Tronci, Giovanni Dipoppa, and Claudio Balducelli. Electric power system anomaly detection using neural networks. In Mircea Gh. Negoita, Robert J. Howlett, and Lakhmi C. Jain (eds.), Knowledge-Based Intelligent Information and Engineering Sys- tems, pp....

  6. [14]

    Walking gait dataset: point clouds, skeletons and silhouettes

    Trong-Nguyen Nguyen and Jean Meunier. Walking gait dataset: point clouds, skeletons and silhouettes. Technical Report 1379, DIRO, University of Montreal, April 2018. URL http: //www.iro.umontreal.ca/˜labimage/GaitDataset/dataset.pdf

  7. [15]

    T. N. Nguyen, H. H. Huynh, and J. Meunier. 3d reconstruction with time-of-flight depth camera and multiple mirrors. IEEE Access , 6:38106–38114, 2018. ISSN 2169-3536. doi: 10.1109/ACCESS.2018.2854262

  8. [16]

    Skeleton-based abnormal gait detection

    Trong-Nguyen Nguyen, Huu-Hung Huynh, and Jean Meunier. Skeleton-based abnormal gait detection. Sensors, 16(11):1792, 2016. ISSN 1424-8220. doi: 10.3390/s16111792. URL http://www.mdpi.com/1424-8220/16/11/1792

  9. [17]

    Extracting silhouette-based characteristics for human gait analysis using one camera

    Trong-Nguyen Nguyen, Huu-Hung Huynh, and Jean Meunier. Extracting silhouette-based characteristics for human gait analysis using one camera. In Proceedings of the Fifth Sym- posium on Information and Communication Technology , SoICT ’14, pp. 171–177, New York, NY , USA, 2014. ...

  10. [18]

    Stabilizing train- ing of generative adversarial networks through regularization

    Kevin Roth, Aurelien Lucchi, Sebastian Nowozin, and Thomas Hofmann. Stabilizing train- ing of generative adversarial networks through regularization. In I. Guyon, U. V . Luxburg, S. Bengio, H. Wallach, R. Fergus, S. Vishwanathan, and R. Garnett (eds.),Advances in Neural Inform...

  11. [19]

    Using tof camera and two mirrors for 3d reconstruction of dynamic objects

    Trong-Nguyen Nguyen, Huu-Hung Huynh, and Jean Meunier. Using tof camera and two mirrors for 3d reconstruction of dynamic objects. Technical Report 1380, DIRO, Univer- sity of Montreal, April 2018. URL http://www.iro.umontreal.ca/˜labimage/ GaitDataset/reconstruct3D.pdf

  12. [20]

    Shotton, A

    J. Shotton, A. Fitzgibbon, M. Cook, T. Sharp, M. Finocchio, R. Moore, A. Kipman, and A. Blake. Real-time human pose recognition in parts from single depth images. In CVPR 2011, pp. 1297–1304, June 2011. doi: 10.1109/CVPR.2011.5995316

  13. [21]

    Anomaly detection using autoencoders with nonlinear dimensionality reduction

    Mayu Sakurada and Takehisa Yairi. Anomaly detection using autoencoders with nonlinear dimensionality reduction. In Proceedings of the MLSDA 2014 2Nd Workshop on Machine Learning for Sensory Data Analysis , MLSDA’14, pp. 4:4–4:11, New York, NY , USA, 2014. ACM. ISBN 978-1-4503-...

  14. [22]

    Real-time action recognition by spatiotem- poral semantic and structural forest

    Tsz-Ho Yu, Tae-Kyun Kim, and Roberto Cipolla. Real-time action recognition by spatiotem- poral semantic and structural forest. In Proceedings of the British Machine Vision Conference, pp. 52.1–52.12. BMV A Press, 2010. ISBN 1-901725-40-5. doi:10.5244/C.24.52. 12

  15. [23]

    Shotton, R

    J. Shotton, R. Girshick, A. Fitzgibbon, T. Sharp, M. Cook, M. Finocchio, R. Moore, P. Kohli, A. Criminisi, A. Kipman, and A. Blake. Efficient human pose estimation from single depth images. IEEE Transactions on Pattern Analysis and Machine Intelligence, 35(12):2821–2840, Dec 20...

  16. [2015]

    doi: 10.1109/NER.2015.7146727

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.