VST-Pose: A Velocity-Integrated Spatiotem-poral Attention Network for Human WiFi Pose Estimation

Jingwei Zhang; Shipeng Yu; Xiang Tian; Xinyu Zhang; Zhisheng Liang; Zhonghao Ye

REVIEW 5 major objections 6 minor 37 references

VST-Pose: A Velocity-Integrated Spatiotem-poral Attention Network for Human WiFi Pose Estimation

T0 review · 5 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 The paper claims that explicitly modeling keypoint velocity lets a spatiotemporal attention network estimate human pose from WiFi CSI at 92.2% PCK@50, outperforming prior WiFi pose estimators on a self-collected dataset.

desk verdict The velocity-integrated attention idea is reasonable and the MMFi results are worth a look, but the self-collected 92.2% PCK@50 claim is inflated by a clip-level random split that leaks near-duplicate temporal windows into both train and test. read the letter →

arxiv 2507.09672 v1 pith:2HIFGTC6 submitted 2025-07-13 cs.CV

Xinyu Zhang , Zhonghao Ye , Jingwei Zhang , Xiang Tian , Zhisheng Liang , Shipeng Yu This is my paper · ORCID

classification cs.CV

keywords WiFisensingchannelstateinformationhumanposeestimationspatiotemporalattentionvelocitymodelingtransformerPCKsmarthomemonitoring

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 fine-grained, video-free human pose estimation is achievable from commodity WiFi channel state information, not just coarse activity recognition. It proposes VST-Pose, which reads short sequences of CSI frames and predicts both joint positions and joint velocities through a dual-stream spatiotemporal Transformer. On a self-collected dataset of 15 home-care actions, the method reaches an overall mean PCK@50 of 92.2% and reports gains over three previous WiFi pose estimators across all evaluated metrics. If the result holds, indoor health monitoring and fall detection could run without cameras, preserving privacy while keeping coverage through walls and in low light.

What carries the argument

The central object is the DST-Block inside the ViSTA-Former backbone: it runs two parallel attention streams, one that applies spatial attention then temporal attention and another that applies temporal then spatial, and attaches a local velocity branch to the time-first stream. The velocity branch learns keypoint displacement patterns within a short sliding window, and its features are fused late with the main keypoint features via F_K = 0.5 F_V + F_N, so the decoder sees both static joint structure and short-term motion. The velocity signal is computed as the difference between the last and first frames of the keypoint sequence, making explicit motion dynamics the mechanism that distinguishes subtle limb movements that are hard to see in instantaneous CSI amplitude.

What would settle it

Evaluate VST-Pose with a subject-exclusive split, training on four of the five volunteers and testing on the fifth, and compare the PCK@20 gap between the velocity-on and velocity-off versions. If the 2.5-point gain shrinks to near zero, or if the overall PCK@50 drops well below 92.2%, the paper's core claim that velocity modeling drives fine-grained accuracy would not survive.

Watch

Extended reading notes

Core claim

VST-Pose's central claim is that a dual-stream spatiotemporal attention backbone, with an auxiliary branch that explicitly regresses keypoint velocity, can map short sequences of WiFi CSI amplitude frames into temporally coherent human skeletons. The paper reports an overall mean PCK@50 of 92.2% and PCK@20 of 71.9% on its self-collected dataset of 33,200 clips, stating that this outperforms MetaFi++, Hpeli, and DT-Pose on PCK@50, PCK@20, MPJPE, and PA-MPJPE. The velocity branch is credited with a 2.5-point PCK@20 improvement over the version without velocity modeling, which the authors interpret as evidence that short-term motion dynamics, not just instantaneous spatial structure, carry the fine-grained pose signal. A separate evaluation on the public MMFi dataset shows improvements in PCK@20 and MPJPE for 3D pose estimation, supporting the claim that the mechanism transfers beyond the authors' own hardware and annotation pipeline.

Load-bearing premise

The central assumption is that randomly splitting the 33,200 CSI clips into training and testing sets at a 4:1 ratio produces independent samples, but because clips come from the same 3-second action recordings and the same five volunteers, adjacent clips can appear in both sets and the reported accuracy may reflect memorizing recording conditions rather than generalizing to new people or actions.

Editorial extensions

If this is right

If the 92.2% PCK@50 result survives subject-independent evaluation, WiFi-based pose estimation becomes usable for continuous home-care monitoring without any camera in the room.
The reported 2.5-point PCK@20 gain from velocity modeling suggests that explicit motion features are a general lever for fine-grained keypoint accuracy in radio-frequency pose estimation.
The improvement on MMFi indicates the architecture transfers to 3D pose estimation and to different WiFi hardware than the Intel 5300 setup used for data collection.
Because a three-frame window works best, the method implies that very short-term dynamics suffice for pose regression, which could keep latency low in real-time monitoring systems.

Reading between the lines

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

The paper's performance gap over DT-Pose, especially the 8.75-point PCK@20 difference, would need corroboration on a subject-exclusive split because clips from the same 3-second recordings and same five volunteers may overlap between training and testing, inflating all methods' numbers.
A natural testable extension is to train VST-Pose on the MMFi training set and evaluate on unseen subjects or unseen rooms; the paper reports only random splits, not cross-domain splits.
The velocity branch uses only the difference between the last and first frame of the window; replacing this with per-frame instantaneous velocity might change the reported 2.5-point fine-grained gain.
The ground-truth labels come from a video-based keypoint detector, so the system's ceiling is bounded by that detector's accuracy; evaluating against other annotation sources would clarify the true limit of WiFi-only pose estimation.

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.

Desk Editor's Note

read the letter

The paper's central claim on the self-collected dataset should not be trusted as reported. The evaluation splits 33,200 short clips randomly at 4:1, but each clip comes from a 3-second action segment that was cut into 10 overlapping clips. Clips from the same segment share the same volunteer, room, action, and mostly overlapping CSI content, so the test set contains near-duplicates of training clips. That allows the model to memorize recording conditions rather than learn a generalizable pose mapping. The reported margins over DT-Pose and MetaFi++ on Table II could shrink or disappear under a subject-disjoint or action-segment-disjoint split. The paper's own limitation section acknowledges cross-domain difficulties, which makes the absence of a strict split especially consequential.

What is actually new: a velocity modeling branch added to a dual-stream spatiotemporal attention backbone, applied to WiFi CSI. The architecture is explicitly inherited from DST-Former/MotionBERT, so the novelty is incremental, but the velocity branch is a sensible twist for CSI and the ablation suggests it helps. The self-collected home-care 2D pose dataset is a useful contribution in principle, though the code and data are not actually released despite the abstract's claim. The MMFi experiments use a public benchmark and show a modest but plausible gain (87.37 vs 86.38 PCK@50), which is less exposed to the leakage problem because it follows the official split.

Other soft spots: hyperparameters (window size, depth, velocity branch source, fusion strategy) are selected by comparing on the same test set used for the final reported numbers, so the reported metrics are cherry-picked. There are no error bars or multiple-seed runs. The abstract's "8.3%" improvement does not match Table II, where the absolute PCK@50 margins over the next best method are 4-6 points.

Who this is for: researchers working on WiFi-based pose estimation who want to see a velocity-integrated transformer variant and a home-care dataset description. The MMFi result is worth verifying, and the architecture is easy enough to reproduce once code is out. It deserves a serious referee because the core idea is legitimate and the public-benchmark result is potentially useful, but the self-collected claim needs a strict split and error bars before it can be accepted.

My recommendation: send to peer review with major revision. Ask for a leave-one-subject-out or disjoint action-segment split, multiple seeds, and a corrected abstract.

Referee Report

5 major / 6 minor

Summary. The paper proposes VST-Pose, a WiFi CSI-based human pose estimation method that combines a dual-stream spatiotemporal transformer backbone (ViSTA-Former) with an explicit velocity modeling branch. The authors collect a 2D home-care dataset with five volunteers performing 15 actions, report 92.25% PCK@50 on it, compare against MetaFi++, Hpeli, and DT-Pose, and also evaluate on the public MMFi dataset for 3D pose estimation. The central claims are state-of-the-art accuracy on the self-collected dataset and improved fine-grained motion modeling via velocity supervision.

Significance. If the reported accuracy were obtained under a valid evaluation protocol, the contribution would be useful: explicit velocity modeling for WiFi pose is a sensible idea, the dual-stream attention design builds on established prior work, the new dataset could be a community resource, and the code is publicly released. However, the evaluation protocol as described—random clip-level splitting, test-set-based hyperparameter selection, and single-run results—means the headline numbers cannot currently be interpreted as generalization performance. The MMFi results also do not show uniform superiority over all baselines on all metrics. The architectural contribution is plausible, but the evidence for its claimed advantage is not yet convincing.

major comments (5)

[Section V.A, Table II] The evaluation protocol in Section V.A splits 33,200 CSI clips randomly at a 4:1 ratio after each 3-second action segment is divided into 10 clips. Because clips from the same action segment share the same volunteer, room, action, and contiguous time interval, random clip-level splitting places temporally and environmentally correlated examples in both training and testing. This can inflate the headline 92.25 PCK@50 in Table II and the velocity ablation in Table VI by allowing the model to memorize recording conditions rather than learn generalizable pose mappings. The paper must report results under a subject-disjoint and/or action-segment-disjoint split (e.g., leave-one-subject-out), and should present the same baseline comparisons under that split. The limitation discussion in Section V.F makes the absence of such a strict split especially consequential.
[Section V.E, Tables IV-VIII] Tables IV, V, VII, and VIII select the sliding-window size, DST-block depth, velocity branch source, and fusion strategy by comparing performance on the same test set that later produces the final reported metrics. This is test-set-based model selection, which can systematically overstate the accuracy of the chosen configuration. The authors should use a validation split for these design choices and then report final results on a genuinely held-out test set, or use nested cross-validation.
[Section V.C, Tables I-VIII] All reported numbers appear to come from a single training run with no error bars. Several key comparisons are small; for example, Table VII shows 92.25 PCK@50 for the TS branch versus 92.06 for TS+ST, and Table VIII shows 92.25 with fusion versus 91.68 without. Without multiple seeds and standard deviations, these differences are not statistically interpretable. Please report mean and standard deviation over at least three random seeds, and indicate statistical significance where appropriate.
[Abstract, Table II] The abstract claims that VST-Pose outperforms existing methods by 8.3% in PCK@50, but Table II shows absolute margins of 4.01, 5.88, and 6.65 percentage points over DT-Pose, MetaFi++, and Hpeli, respectively, corresponding to relative improvements of roughly 4.5%, 6.8%, and 7.8%. The paper should specify the baseline and formula used for the 8.3% figure, or correct the claim.
[Section V.D, Table III] On the MMFi dataset, VST-Pose's PA-MPJPE is 105.9 mm, which is worse than Hpeli (105.3) and DT-Pose (105.6). The text in Section V.D says the method gives 'significant improvements in PCK@20 and MPJPE,' and the conclusion states the model 'consistently outperforms existing approaches,' but this is not supported on MMFi across all metrics. Please qualify the MMFi claims or provide additional evidence for robustness.

minor comments (6)

[Section II.A] Typo: 'estimateon' should be 'estimation'.
[Section V.A] Typo: 'diraction' should be 'direction'.
[Section V.A] The dataset description says 3,300 action segments, each divided into 10 clips, which gives 33,000 clips, not 33,200. Please reconcile the count.
[Section V.D, Table II] Reference numbering is inconsistent: MetaFi++ is cited as [37] in Table II but appears as reference [35] in the reference list, where [37] is assigned to a different work (HPE-Li).
[Section IV.D, Eq. (5)] Equation (5) is garbled in the text; please rewrite the loss function and the definition of ground-truth velocity clearly.
[Section V.E] In the Network Depth Selection paragraph, the sentence 'Table IV is composed of stacked DST-Blocks' appears to be a caption duplication; the reference should be to the architecture description or to Table V.

Circularity Check

2 steps flagged · score 6.0 of 10

Reported 92.2% PCK@50 is not an independent prediction: hyperparameters are selected on the same test split used to report final metrics, and the random clip-level split allows train/test overlap from shared action segments.

fitted input called prediction [Section V.E (Ablation Study), Tables IV-VIII; final configuration in Section V.C]
"We compared temporal windows of 3, 4, 5, and 7 frames, with a fixed stride of 2 to reduce temporal redundancy. As shown in Table IV The results indicate that a window size of 3 performs the best. ... Table V shows consistent improvements in PCK and PA-MPJPE as the depth increases up to 5 layers. ... we adopt a 5-layer configuration as the default for our self-collected dataset."

The paper describes a single random 4:1 split of the 33,200 CSI segments into training and testing sets and then uses that same self-collected test set to compare window sizes, DST-Block depths, velocity-branch sources, and fusion strategies. The final configuration is the one that maximizes the reported metrics on the test split, and the same test split is then used to report the 'best' 92.25% PCK@50 / 71.90% PCK@20 numbers. Because the hyperparameters are selected on the evaluation split, the headline results are a selected maximum over the searched configurations, not an independent prediction of generalization. No validation set or nested procedure is described, so the selection and the final measurement are statistically inseparable.
other [Section V.A (Set Up), data partitioning paragraph]
"Each action sequence was divided into 10 data clips to encourage the model to learn the spatial variations of the same action. Each clip contains 9 consecutive CSI frames and is fed into the short-term temporal modeling network using a sliding window mechanism. In total, we obtained 33,200 CSI segments, which were randomly split into training and testing sets with a ratio of 4:1."

By construction, all 10 clips from one 3-second action segment share the same volunteer, room, action, and largely overlapping CSI content (9-frame clips from a continuous sliding window). A random clip-level split therefore places near-duplicate samples in both training and testing, so the test set is not disjoint from training at the action or subject level. The reported PCK/MPJPE values can be inflated by memorizing recording conditions rather than measuring generalization to new people or actions. The paper's own limitation section acknowledges cross-domain difficulty, which makes the absence of a subject- or action-disjoint split especially consequential for the central claim.

full rationale

The paper's architectural derivation (CSI encoder, ViSTA-Former dual-stream attention, velocity branch, decoder) is not circular in the self-citation or ansatz-smuggling sense: the velocity branch is an auxiliary regression trained with a multi-task loss, and the dual-stream design is attributed to an external prior work (MotionBERT/DST-Former, [28]), not to the authors' own unverified theorem. However, the evaluation chain that produces the headline '92.2% PCK@50' is partially circular. The paper selects the window size, network depth, velocity-branch source, and fusion strategy by comparing performance on the same randomly split test set that is later used to report the final state-of-the-art numbers (Tables IV-VIII vs Table II). Since no separate validation split is described, the reported 'best' metrics are fitted to the test set. Additionally, the random split operates on clips rather than action segments or subjects, so training and testing contain near-duplicate CSI clips from the same 3-second recordings; this further makes the test metric an artifact of the split construction. These are evaluation-protocol circularities rather than mathematical tautologies, so the paper retains independent content in its architecture and loss formulation, but the central empirical claim is materially compromised. Score 6 reflects partial circularity: one or more reported predictions reduce, by the test-set selection and split construction, to the inputs used to choose them.

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

The central claims rest on a self-collected dataset, OpenPose-derived labels, and several hyperparameters selected via test-set comparisons. No new physical entities are introduced; the main unverified inputs are the temporal independence of train/test clips and the reliability of OpenPose labels.

free parameters (5)

loss weight alpha = 0.2
Balances position and velocity regression in Eq. (5); chosen by hand with no sensitivity analysis.
sliding window size = 3 frames
Selected as best in Table IV by comparing test-set PCK values.
DST-Block depth N = 5 layers on self-dataset, 1 layer on MMFi
Selected as best in Table V on the test set; the MMFi setting is chosen for model stability.
velocity branch source = TS branch
Selected by comparing ST, TS+ST, and TS variants on the test set (Table VII).
late fusion weight = 0.5
Fixed coefficient in FK = 0.5 FV + FN (Section IV.B); no ablation is reported.

assumptions (5)

domain assumption OpenPose-derived 2D keypoints from synchronized RGB frames are treated as ground truth human pose.
Section III uses OpenPose to extract labels; OpenPose errors become label noise and are not corrected or validated against manual annotations.
domain assumption Single-direction 3x3 MIMO CSI amplitude at 2.4 GHz contains sufficient spatial information to regress 17 keypoints.
Central premise of the system; Section III describes one transmitter-receiver pair in one direction.
domain assumption CSI-video synchronization and 30 Hz keypoint alignment are accurate.
Section III states each CSI frame is temporally aligned with the corresponding RGB frame, but no synchronization error analysis is provided.
domain assumption Randomly splitting clips from the same action sequence into train and test yields independent evaluation samples.
Section V.A describes a segment-level random 4:1 split; adjacent clips from the same recording violate the independence assumption.
standard math Discrete wavelet transform removes noise without removing motion signal.
Section III applies DWT filtering as a standard signal processing step, but the preservation of pose-relevant signal is not validated.

how reviews work

0 comments

Cite this review

Pith. "Pith review of VST-Pose: A Velocity-Integrated Spatiotem-poral Attention Network for Human WiFi Pose Estimation." pith.science (2026). https://pith.science/paper/2HIFGTC6

@misc{pith2026250709672,
  author       = {Pith},
  title        = {Pith review of: VST-Pose: A Velocity-Integrated Spatiotem-poral Attention Network for Human WiFi Pose Estimation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2HIFGTC6}},
  note         = {Machine review of arXiv:2507.09672}
}

read the original abstract

WiFi-based human pose estimation has emerged as a promising non-visual alternative approaches due to its pene-trability and privacy advantages. This paper presents VST-Pose, a novel deep learning framework for accurate and continuous pose estimation using WiFi channel state information. The proposed method introduces ViSTA-Former, a spatiotemporal attention backbone with dual-stream architecture that adopts a dual-stream architecture to separately capture temporal dependencies and structural relationships among body joints. To enhance sensitivity to subtle human motions, a velocity modeling branch is integrated into the framework, which learns short-term keypoint dis-placement patterns and improves fine-grained motion representation. We construct a 2D pose dataset specifically designed for smart home care scenarios and demonstrate that our method achieves 92.2% accuracy on the PCK@50 metric, outperforming existing methods by 8.3% in PCK@50 on the self-collected dataset. Further evaluation on the public MMFi dataset confirms the model's robustness and effectiveness in 3D pose estimation tasks. The proposed system provides a reliable and privacy-aware solution for continuous human motion analysis in indoor environments. Our codes are available in https://github.com/CarmenQing/VST-Pose.

Figures

Figures reproduced from arXiv: 2507.09672 by the authors.

**Figure 1.** 30 different colors represent the 30 subcarriers of a single antenna. (a) Shows the image of a volunteer performing the standing-up action. (b) Displays the corresponding CSI amplitude waveform at the synchronized timestamp, and (c) Shows the CSI after wavelet transform [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

37 extracted references · 34 canonical work pages

[1]

DeepPose: Human pose estimation via deep neural networks,

A. Toshev and C. Szegedy, “DeepPose: Human pose estimation via deep neural networks,” in Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), Columbus, OH, USA, Jun. 2014, pp. 1653–1660

work page 2014
[2]

AlphaPose: Whole-Body Regional Multi-Person Pose Estimation and Tracking in Real-Time

H.-S. Fang et al., “AlphaPose: Whole-body regional multi-person pose estimation and tracking in real-time,” 2022, arXiv: 2211.03375

work page Pith review arXiv 2022
[3]

Cascaded pyramid network for multi-person pose estimation,

Y. Chen, Z. Wang, Y. Peng, Z. Zhang, G. Yu, and J. Sun, “Cascaded pyramid network for multi-person pose estimation,” in Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), Salt Lake City, Jun. 2018, pp. 7103–7112

work page 2018
[4]

OpenPose: Realtime multi-person 2D pose estimation using part affinity fields,

Z. Cao, G. Hidalgo, T. Simon, S. E. Wei, and Y. Sheikh, “OpenPose: Realtime multi-person 2D pose estimation using part affinity fields,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 43, no. 1, pp. 172–186, Jan. 2021

work page 2021
[5]

Human pose estimation in extremely low-light condi- tions,

S. Lee et al., “Human pose estimation in extremely low-light condi- tions,” in Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), Jun. 2023, pp. 704–714

work page 2023
[6]

3D Human pose estimation in the wild by adversarial learning,

W. Yang, W. Ouyang, X. Wang, J. Ren, H. Li, and X. Wang, “3D Human pose estimation in the wild by adversarial learning,” in Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), Jun. 2018, pp. 5255–5264

work page 2018
[7]

WiFi sensing with channel state in- formation: A survey,

Y. Ma, G. Zhou, and S. Wang, “WiFi sensing with channel state in- formation: A survey,” ACM Comput. Surv., vol. 52, no. 3, pp. 1–36, Jul. 2020

work page 2020
[8]

Inferring motion direction using commodity Wi-Fi for interactive exergames,

K. Qian, C. Wu, Z. Zhou, Y. Zheng, Z. Yang, and Y. Liu, “Inferring motion direction using commodity Wi-Fi for interactive exergames,” in Proc. CHI Conf. Human Factors Comput. Syst. (CHI), Denver, CO, USA, May 2017, pp. 1961–1972. TABLE VI THE PERFORMANCE OF VELOCITY MODELING Velocity Method PCK@50 PCK@20 MPJPE PA-MPJPE Without Velocity Modeling 91.54 69.4...

work page 2017

Show all 37 references

[9]

E-eyes: Device-free location-oriented activity identification using fine-grained Wi-Fi signatures,

Y. Wang, J. Liu, Y. Chen, M. Gruteser, J. Yang, and H. Liu, “E-eyes: Device-free location-oriented activity identification using fine-grained Wi-Fi signatures, ” in Proc. 20th Annu. Int. Conf. Mobile Comput. Netw. (MobiCom), Maui, HI, USA, Sep. 2014, pp. 617–628

2014
[10]

Widar3.0: Zero-effort cross-domain gesture recogni- tion with Wi-Fi,

Y. Zhang et al., “Widar3.0: Zero-effort cross-domain gesture recogni- tion with Wi-Fi,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 44, no. 11, pp. 8671–8688, Nov. 2022

2022
[11]

Wi-Breath: A WiFi-based contactless and real-time respiration monitoring scheme for remote healthcare,

N. Bao et al., “Wi-Breath: A WiFi-based contactless and real-time respiration monitoring scheme for remote healthcare,” IEEE J. Biomed. Health Inform., vol. 27, no. 5, pp. 2276–2285, May 2023

2023
[12]

A Survey on behavior recognition Using WiFi Channel State Information,

S. Yousefi, H. Narui, S. Dayal, S. Ermon, and S. Valaee, “A Survey on behavior recognition Using WiFi Channel State Information,” IEEE Commun. Mag., vol. 55, no. 10, pp. 98–104, Oct. 2017

2017
[13]

EfficientFi: Toward Large-scale lightweight WiFi Sensing via CSI Compression,

J. Yang, X. Chen, H. Zou, D. Wang, Q. Xu, and L. Xie, “EfficientFi: Toward Large-scale lightweight WiFi Sensing via CSI Compression,” IEEE Internet Things J., vol. 9, no. 15, pp. 13086–13095, Aug. 2022

2022
[14]

Can WiFi estimate person pose?

F. Wang, S. Panev, Z. Dai, J. Han, and D. Huang, “Can WiFi estimate person pose?” 2019, arXiv:1904.00277

2019 arXiv
[15]

From signal to image: Capturing fine-grained human poses With Commodity Wi-Fi,

L. Guo, Z. Lu, X. Wen, S. Zhou, and Z. Han, “From signal to image: Capturing fine-grained human poses With Commodity Wi-Fi,” IEEE Commun. Lett., vol. 24, no. 4, pp. 802–806, Apr. 2020

2020
[16]

PerUnet: Deep signal channel attention in unet for WiFi-based human pose estimation,

Y. Zhou, A. Zhu, C. Xu, F. Hu, and Y. Li, “PerUnet: Deep signal channel attention in unet for WiFi-based human pose estimation,” IEEE Sens. J., vol. 22, no. 20, pp. 19750–19760, Oct. 2022

2022
[17]

Towards 3D human pose construction using WiFi,

W. Jiang et al., “Towards 3D human pose construction using WiFi,” in Proc. 26th Annu. Int. Conf. Mobile Comput. Netw., 2020, pp. 1–14

2020
[18]

From point to space: 3D moving human pose estimation using commodity WiFi,

Y. Wang, L. Guo, Z. Lu, X. Wen, S. Zhou, and W. Meng, “From point to space: 3D moving human pose estimation using commodity WiFi,” IEEE Commun. Lett., vol. 25, no. 7, pp. 2235–2239, Jul. 2021

2021
[19]

GoPose: 3D human pose estimation using WiFi,

Y. Ren, Z. Wang, Y. Wang, S. Tan, Y. Chen, and J. Yang, “GoPose: 3D human pose estimation using WiFi,” Proc. ACM Interact. Mob. Wearable Ubiquitous Technol., vol. 6, no. 2, pp. 1–25, Jul. 2022

2022
[20]

Person-in-WiFi: Fine-grained person perception using WiFi,

F. Wang, S. Zhou, S. Panev, J. Han, and D. Huang, “Person-in-WiFi: Fine-grained person perception using WiFi,” in Proc. IEEE/CVF Int. Conf. Comput. Vis., Oct. 2019, pp. 5452–5461

2019
[21]

Per- son-in-WiFi 3D: End-to-end multi-person 3D pose estimation with Wi-Fi,

K. Yan, F. Wang, B. Qian, H. Ding, J. Han, and X. Wei, “Per- son-in-WiFi 3D: End-to-end multi-person 3D pose estimation with Wi-Fi,” in Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), Seattle, WA, USA, Jun. 2024, pp. 969–978

2024
[22]

DensePose from WiFi,

Geng J., Huang D., and Torre F. D., “DensePose from WiFi,” 2022, arXiv: 2301.00250

2022 arXiv
[23]

Towards robust and Rrealistic human pose estimation via WiFi signals,

Y. Chen, J. Guo, S. Guo, J. Zhou, and D. Tao, “Towards robust and Rrealistic human pose estimation via WiFi signals,” 2025, arXiv: 2501.09411

2025 arXiv
[24]

AdaPose: Toward cross-site device-free human pose estimation with commodity Wi-Fi,

Y. Zhou, J. Yang, H. Huang, and L. Xie, “AdaPose: Toward cross-site device-free human pose estimation with commodity Wi-Fi, ”IEEE In- ternet Things J., vol. 11, no. 24, pp. 40255–40267, Dec. 2024

2024
[25]

Winect: 3D human pose tracking for free-form activity using commodity WiFi,

Y. Ren, Z. Wang, S. Tan, Y. Chen, and J. Yang, “Winect: 3D human pose tracking for free-form activity using commodity WiFi, ” Proc. ACM Interact. Mob. Wearable Ubiquitous Technol., vol. 5, no. 4, pp. 1–29, Dec. 2021

2021
[26]

Robust multi-user pose estimation based on spatial and temporal features from Wi-Fi CSI,

T.-W. Hsu and H.-Y. Hsieh, “Robust multi-user pose estimation based on spatial and temporal features from Wi-Fi CSI, ” in Proc. IEEE Int. Conf. Commun. (ICC), Denver, CO, USA, Jun. 2024, pp. 1600–1605

2024
[27]

Attention is all you need,

A. Vaswani et al., “Attention is all you need,” 2017, arXiv:1706.03762

2017 arXiv
[28]

MotionBERT: A unified perspective on learning human motion representations,

W. Zhu, X. Ma, Z. Liu, L. Liu, W. Wu, and Y. Wang, “MotionBERT: A unified perspective on learning human motion representations, ”in Proc. IEEE Int. Conf. Comput. Vis. (ICCV), Oct. 2023, pp. 15039–15053

2023
[29]

MM-Fi: Multi-modal non-intrusive 4D human dataset for versatile wireless sensing,

J. Yang et al., “MM-Fi: Multi-modal non-intrusive 4D human dataset for versatile wireless sensing,” 2023, arXiv: 2305.10345

2023 arXiv
[30]

Wearable sensors for moni- toring human motion: A review on mechanisms, materials, and chal- lenges,

S. Z. Homayounfar and T. L. Andrew, “Wearable sensors for moni- toring human motion: A review on mechanisms, materials, and chal- lenges,” SLAS Technol., vol. 25, no. 1, pp. 9–24, Feb. 2020

2020
[31]

LiDARCap: Long-range markerless 3D human motion capture with LiDAR point clouds,

J. Li et al., “LiDARCap: Long-range markerless 3D human motion capture with LiDAR point clouds,” in Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), New Orleans, LA, USA, Jun. 2022, pp. 20470–20480

2022
[32]

Through-wall human pose estimation using radio signals,

M. Zhao et al., “Through-wall human pose estimation using radio signals,” in Proc. IEEE Conf. Comput. Vis. Pattern Recognit., 2018, pp. 7356–7365

2018
[33]

RF-based 3-D skeleton,

M. Zhao et al. , “RF-based 3-D skeleton,”in Proc. ACM SIGCOMM Conf. Data Commun. (SIGCOMM), Aug. 2018, pp. 267–281

2018
[34]

CSI-Former: Pay more attention to pose estimation with WiFi,

Y. Zhou, C. Xu, L. Zhao, A. Zhu, F. Hu, and Y. Li, “CSI-Former: Pay more attention to pose estimation with WiFi,” Entropy, vol. 25, no. 1, p. 20, Dec. 2022

2022
[35]

MetaFi++: WiFi-enabled transformer-based human pose estimation for metaverse avatar simulation,

Y. Zhou, H. Huang, S. Yuan, H. Zou, L. Xie, and J. Yang, “MetaFi++: WiFi-enabled transformer-based human pose estimation for metaverse avatar simulation,” IEEE Internet Things J., vol. 10, no. 16, pp. 14128– 14136, Aug. 2023

2023
[36]

Tool Release: Gathering 802.11n Traces With Channel State Information,

D. Halperin, W. Hu, A. Sheth, and D. Wetherall, “Tool Release: Gathering 802.11n Traces With Channel State Information, ” ACM SIGCOMM Comput. Commun. Rev., vol. 41, no. 1, p. 53, Jan. 2011

2011
[37]

HPE-Li: Wi-Fi-enabled lightweight dual selective kernel convolution for human pose estimation,

T. D. Gian, T. D. Lai, T. V. Luong, K.-S. Wong, and V.-D. Nguyen, “HPE-Li: Wi-Fi-enabled lightweight dual selective kernel convolution for human pose estimation, ” in Proc. Eur. Conf. Comput. Vis. (ECCV), vol. 15089. Cham, Switzerland, Jan. 2025, pp. 93–111

2025

Pith tools

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

[1] [1]

DeepPose: Human pose estimation via deep neural networks,

A. Toshev and C. Szegedy, “DeepPose: Human pose estimation via deep neural networks,” in Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), Columbus, OH, USA, Jun. 2014, pp. 1653–1660

work page 2014

[2] [2]

AlphaPose: Whole-Body Regional Multi-Person Pose Estimation and Tracking in Real-Time

H.-S. Fang et al., “AlphaPose: Whole-body regional multi-person pose estimation and tracking in real-time,” 2022, arXiv: 2211.03375

work page Pith review arXiv 2022

[3] [3]

Cascaded pyramid network for multi-person pose estimation,

Y. Chen, Z. Wang, Y. Peng, Z. Zhang, G. Yu, and J. Sun, “Cascaded pyramid network for multi-person pose estimation,” in Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), Salt Lake City, Jun. 2018, pp. 7103–7112

work page 2018

[4] [4]

OpenPose: Realtime multi-person 2D pose estimation using part affinity fields,

Z. Cao, G. Hidalgo, T. Simon, S. E. Wei, and Y. Sheikh, “OpenPose: Realtime multi-person 2D pose estimation using part affinity fields,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 43, no. 1, pp. 172–186, Jan. 2021

work page 2021

[5] [5]

Human pose estimation in extremely low-light condi- tions,

S. Lee et al., “Human pose estimation in extremely low-light condi- tions,” in Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), Jun. 2023, pp. 704–714

work page 2023

[6] [6]

3D Human pose estimation in the wild by adversarial learning,

W. Yang, W. Ouyang, X. Wang, J. Ren, H. Li, and X. Wang, “3D Human pose estimation in the wild by adversarial learning,” in Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), Jun. 2018, pp. 5255–5264

work page 2018

[7] [7]

WiFi sensing with channel state in- formation: A survey,

Y. Ma, G. Zhou, and S. Wang, “WiFi sensing with channel state in- formation: A survey,” ACM Comput. Surv., vol. 52, no. 3, pp. 1–36, Jul. 2020

work page 2020

[8] [8]

Inferring motion direction using commodity Wi-Fi for interactive exergames,

K. Qian, C. Wu, Z. Zhou, Y. Zheng, Z. Yang, and Y. Liu, “Inferring motion direction using commodity Wi-Fi for interactive exergames,” in Proc. CHI Conf. Human Factors Comput. Syst. (CHI), Denver, CO, USA, May 2017, pp. 1961–1972. TABLE VI THE PERFORMANCE OF VELOCITY MODELING Velocity Method PCK@50 PCK@20 MPJPE PA-MPJPE Without Velocity Modeling 91.54 69.4...

work page 2017