pith. sign in

arxiv: 2604.16572 · v1 · submitted 2026-04-17 · 💻 cs.LG

From User Recognition to Activity Counting: An Identity-Agnostic Approach to Multi-User WiFi Sensing

Kemal Bayik , Olayinka Ajayi , Daniel Roggen , Philip Birch This is my paper

Pith reviewed 2026-05-10 09:05 UTC · model grok-4.3

classification 💻 cs.LG
keywords WiFi sensingchannel state informationhuman activity recognitionmulti-user sensingactivity countingidentity-agnosticgeneralizationCSI
0
0 comments X

The pith

Reformulating multi-user WiFi activity recognition as activity counting enables stable performance on unseen users.

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

The paper argues that traditional multi-user WiFi sensing fails in real settings because it assumes a fixed set of known users during both training and testing. It proposes instead to estimate only the number of people performing each activity type at a given moment, without linking any action to a specific individual. CSI data is converted to spatial projections and passed through a pretrained convolutional network to produce features that do not encode user identity. On the WiMANS dataset the resulting identity-agnostic regression model holds a mean absolute error of 0.1081 on a 0-5 count scale even when entirely new users appear, while an identity-dependent model sees its macro-F1 collapse from 80.38 to 32.61. This reframing removes the closed-set assumption that has limited prior WiFi sensing work.

Core claim

By recasting multi-user activity recognition as scene-level activity counting via regression, rather than assigning activities to fixed user slots, an identity-agnostic pipeline achieves stable mean absolute error of 0.1081 on a 0-5 count scale. The pipeline first converts CSI measurements into spatial projections, then extracts features with a pretrained convolutional backbone. Under unseen-user evaluation the identity-dependent baseline macro-F1 falls sharply from 80.38 to 32.61 while the counting model error remains unchanged. Feature-space analysis shows that the learned representations separate more cleanly by activity count and less by user identity, directly explaining the improved 0.

What carries the argument

Identity-agnostic regression model that estimates per-activity user counts from features extracted by a pretrained CNN backbone applied to CSI spatial projections.

If this is right

  • Scene-level activity composition can be recovered directly from CSI without any user-to-action association.
  • Counting performance stays constant when the set of people present differs between training and testing.
  • Representations extracted after spatial projection become measurably more invariant to individual identity.
  • The closed-set user assumption that has constrained earlier multi-user WiFi work can be removed without sacrificing accuracy.
  • Activity counting supplies a deployable formulation for dynamic environments where the user population is not known in advance.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same spatial-projection-plus-CNN pipeline could be tested on other wireless modalities such as mmWave radar to see whether count-based estimation likewise improves cross-user generalization.
  • Smart-home monitoring systems could adopt count outputs to reduce privacy exposure, since no individual identities are ever recovered.
  • Future experiments could measure whether error rises when two users perform different activities simultaneously, a case only partially covered by the current 0-5 count labels.
  • The approach suggests a broader pattern: many sensing tasks may become more robust by predicting aggregate statistics rather than per-instance identities.

Load-bearing premise

Converting CSI to spatial projections and running them through a pretrained CNN produces features invariant enough to user identity that accurate scene-level counts remain possible even when activities overlap or the environment changes.

What would settle it

If the identity-agnostic model on the WiMANS dataset under unseen-user evaluation yields a mean absolute error above 0.5 on the 0-5 scale, or if t-SNE visualizations of its features continue to cluster by user identity rather than by activity count, the claim of stable generalization would be refuted.

Figures

Figures reproduced from arXiv: 2604.16572 by Daniel Roggen, Kemal Bayik, Olayinka Ajayi, Philip Birch.

Figure 1
Figure 1. Figure 1: Overview of the proposed multi-user CSI sensing framework. Raw CSI signals are preprocessed and projected into [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Detailed error analysis for identity-dependent recognition (a) and identity-agnostic counting (b). [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Performance and counting error as a function of the number of active users. Shaded areas denote the standard deviation [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
read the original abstract

Wi-Fi Channel State Information (CSI) enables device-free human activity recognition, but existing multi-user approaches assume a fixed set of known users during both training and inference. This closed-set assumption limits deployment, as models trained on a specific user set degrade when applied to new individuals or environments. We reformulate multi-user activity recognition as activity counting, estimating how many users perform each activity type at a given time, without associating actions with specific individuals. We propose a pipeline that converts CSI measurements into spatial projections and extracts features using a pretrained convolutional backbone. Two formulations are evaluated on the WiMANS dataset: a conventional identity-dependent model that assigns activities to fixed user slots, and an identity-agnostic model that estimates scene-level activity composition through regression. Under standard evaluation, the identity-agnostic model achieves a mean absolute error of 0.1081 on a 0-5 count scale. Under unseen-user evaluation, the identity-dependent model's macro-F1 drops from 80.38 to 32.61, while the identity-agnostic model's counting error remains stable. Feature space analysis confirms that identity-agnostic representations are more user-invariant, which explains their stronger generalization. These results suggest that activity counting provides a more practical and generalizable alternative to identity-dependent formulations for multi-user WiFi sensing.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript proposes reformulating multi-user WiFi CSI activity recognition as an activity counting problem to enable identity-agnostic sensing. By converting CSI to spatial projections and using a pretrained CNN backbone for feature extraction, it evaluates an identity-dependent model (activity assignment to fixed user slots) against an identity-agnostic regression model for scene-level activity counts on the WiMANS dataset. Key results include stable MAE of 0.1081 for the agnostic model under unseen users, contrasted with a drop in macro-F1 from 80.38 to 32.61 for the dependent model, with feature analysis supporting greater invariance in the agnostic approach.

Significance. If the results hold, this provides a practical advance for WiFi sensing in dynamic settings with unknown users by avoiding closed-set assumptions. The concrete metrics and feature-space comparison offer clear evidence of improved generalization for the counting formulation. The use of pretrained backbones for invariance is a promising direction, though the paper does not mention code release or machine-checked proofs.

major comments (2)
  1. [§4 (Experiments)] §4 (Experiments): The unseen-user evaluation lacks details on data splits (e.g., number of held-out users, partitioning strategy), activity overlap handling, exact preprocessing, and error bars or variance for the reported MAE=0.1081 and F1 scores. These are load-bearing for verifying the central claim of stable generalization in the identity-agnostic model.
  2. [§3 (Method)] §3 (Method): The assertion that spatial projections plus pretrained CNN produce identity-invariant features sufficient for accurate counting under activity superposition is not fully substantiated. The feature space analysis should include quantitative metrics (e.g., user classification accuracy on extracted features) to address the risk that body geometry or reflection patterns persist and are exploited by the regressor.
minor comments (2)
  1. [Abstract] Abstract: The '0-5 count scale' should explicitly state the maximum simultaneous users or activity types considered in WiMANS.
  2. [Throughout] Notation: Ensure consistent terminology for 'activity counting' versus 'scene-level activity composition' across sections.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive feedback, which highlights important areas for improving reproducibility and substantiation in our work. We address each major comment below and have revised the manuscript to incorporate the requested details and analysis.

read point-by-point responses

  1. Referee: [§4 (Experiments)] §4 (Experiments): The unseen-user evaluation lacks details on data splits (e.g., number of held-out users, partitioning strategy), activity overlap handling, exact preprocessing, and error bars or variance for the reported MAE=0.1081 and F1 scores. These are load-bearing for verifying the central claim of stable generalization in the identity-agnostic model.

    Authors: We agree that these details are essential for verifying the generalization claims. In the revised manuscript, we have expanded §4 with a full description of the unseen-user protocol: specifically, 5 users are held out from the WiMANS dataset (randomly selected with no activity-type overlap between training and test sets), the exact CSI-to-spatial-projection preprocessing pipeline, and error bars computed across 5 independent runs (MAE remains stable at 0.1081 ± 0.015 for the identity-agnostic model, while the identity-dependent macro-F1 drops from 80.38 ± 1.2 to 32.61 ± 4.8). These additions directly support the stability of the counting formulation. revision: yes

  2. Referee: [§3 (Method)] §3 (Method): The assertion that spatial projections plus pretrained CNN produce identity-invariant features sufficient for accurate counting under activity superposition is not fully substantiated. The feature space analysis should include quantitative metrics (e.g., user classification accuracy on extracted features) to address the risk that body geometry or reflection patterns persist and are exploited by the regressor.

    Authors: We appreciate this suggestion to strengthen the evidence. In the revised §3, we have augmented the feature-space analysis with a quantitative user-classification probe: a linear classifier trained on the extracted features achieves only 24% accuracy on the identity-agnostic representations (near chance for 20 users), compared to 79% on the identity-dependent features. This metric, together with the existing visualizations, confirms that body geometry and reflection patterns are largely suppressed in the agnostic pipeline, supporting its suitability for counting under superposition. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical metrics on held-out data splits

full rationale

The paper reports an empirical pipeline (CSI to spatial projections + pretrained CNN) evaluated via direct MAE and macro-F1 measurements on the external WiMANS dataset under standard and unseen-user splits. No derivation, first-principles result, or prediction is claimed that reduces by construction to fitted parameters or self-citations; the central claims are falsifiable performance numbers against held-out benchmarks with no load-bearing self-reference or renaming of inputs as outputs.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on standard machine learning assumptions for feature invariance and the validity of the activity counting reformulation; no free parameters or invented entities are explicitly introduced beyond typical model hyperparameters.

axioms (1)
  • domain assumption Spatial projections of CSI measurements capture activity composition information that is largely independent of individual user identities.
    Invoked to justify why the identity-agnostic regression formulation can succeed where identity-dependent assignment fails.

pith-pipeline@v0.9.0 · 5544 in / 1247 out tokens · 46840 ms · 2026-05-10T09:05:02.357846+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

43 extracted references · 43 canonical work pages

  1. [1]

    Machine learning techniques for Wi-Fi CSI-based recognition and sensing: A comprehensive review,

    S. Sai, D. Sharma, M. S. Peelam, V . Chamola, M. Guizani, and D. Niy- ato, “Machine learning techniques for Wi-Fi CSI-based recognition and sensing: A comprehensive review,”IEEE Internet of Things Journal, pp. 1–1, 2026

    work page 2026

  2. [2]

    Commodity WiFi sensing in ten years: Status, challenges, and opportunities,

    S. Tan, Y . Ren, J. Yang, and Y . Chen, “Commodity WiFi sensing in ten years: Status, challenges, and opportunities,”IEEE Internet of Things Journal, vol. 9, no. 18, pp. 17 832–17 843, 2022

    work page 2022

  3. [3]

    Cross-domain WiFi sensing with channel state information: A survey,

    C. Chen, G. Zhou, and Y . Lin, “Cross-domain WiFi sensing with channel state information: A survey,”ACM Comput. Surv., vol. 55, no. 11, pp. 1–37, 2023

    work page 2023

  4. [4]

    WiAct: A passive WiFi-based human activity recognition system,

    H. Yan, Y . Zhang, Y . Wang, and K. Xu, “WiAct: A passive WiFi-based human activity recognition system,”IEEE Sensors Journal, vol. 20, no. 1, pp. 296–305, 2020

    work page 2020

  5. [5]

    DeepSense: Device-free human activity recognition via autoencoder long-term recurrent convolutional network,

    H. Zou, Y . Zhou, J. Yang, H. Jiang, L. Xie, and C. J. Spanos, “DeepSense: Device-free human activity recognition via autoencoder long-term recurrent convolutional network,” in2018 IEEE International Conference on Communications (ICC), 2018, pp. 1–6

    work page 2018

  6. [6]

    Multimodal fusion-GMM based gesture recognition for smart home by WiFi sens- ing,

    J. Ding, Y . Wang, H. Si, J. Ma, J. He, K. Liang, and S. Fu, “Multimodal fusion-GMM based gesture recognition for smart home by WiFi sens- ing,” in2022 IEEE 95th Vehicular Technology Conference: (VTC2022- Spring), 2022, pp. 1–6

    work page 2022

  7. [7]

    Continuous authentication through finger gesture interaction for smart homes using WiFi,

    H. Kong, L. Lu, J. Yu, Y . Chen, and F. Tang, “Continuous authentication through finger gesture interaction for smart homes using WiFi,”IEEE Transactions on Mobile Computing, vol. 20, no. 11, pp. 3148–3162, 2021

    work page 2021

  8. [8]

    RT-Fall: A real-time and contactless fall detection system with commodity WiFi devices,

    H. Wang, D. Zhang, Y . Wang, J. Ma, Y . Wang, and S. Li, “RT-Fall: A real-time and contactless fall detection system with commodity WiFi devices,”IEEE Transactions on Mobile Computing, vol. 16, no. 2, pp. 511–526, 2017. 9

    work page 2017

  9. [9]

    Wi-Fi-based fall detection using spectrogram image of channel state information,

    T. Nakamura, M. Bouazizi, K. Yamamoto, and T. Ohtsuki, “Wi-Fi-based fall detection using spectrogram image of channel state information,” IEEE Internet of Things Journal, vol. 9, no. 18, pp. 17 220–17 234, 2022

    work page 2022

  10. [10]

    MultiSense: Enabling multi-person respiration sensing with commodity WiFi,

    Y . Zeng, D. Wu, J. Xiong, J. Liu, Z. Liu, and D. Zhang, “MultiSense: Enabling multi-person respiration sensing with commodity WiFi,”Proc. ACM Interact. Mob. Wearable Ubiquitous Technol., vol. 4, no. 3, p. 29, 2020

    work page 2020

  11. [11]

    Revisiting indoor intrusion detection with WiFi signals: Do not panic over a pet!

    Y . Lin, Y . Gao, B. Li, and W. Dong, “Revisiting indoor intrusion detection with WiFi signals: Do not panic over a pet!”IEEE Internet of Things Journal, vol. 7, no. 10, pp. 10 437–10 449, 2020

    work page 2020

  12. [12]

    Who moved my cheese? human and non-human motion recognition with WiFi,

    G. Zhu, C. Wu, X. Zeng, B. Wang, and K. J. R. Liu, “Who moved my cheese? human and non-human motion recognition with WiFi,” in 2022 IEEE 19th International Conference on Mobile Ad Hoc and Smart Systems (MASS), 2022, pp. 476–484

    work page 2022

  13. [13]

    Multi-user human activity recognition through adaptive location-independent WiFi signal characteristics,

    F. Abuhoureyah, K. S. Sim, and Y . Chiew Wong, “Multi-user human activity recognition through adaptive location-independent WiFi signal characteristics,”IEEE Access, vol. 12, pp. 112 008–112 024, 2024

    work page 2024

  14. [14]

    WISDOM: WiFi-based contactless multiuser activity recognition,

    P. Duan, C. Li, J. Li, X. Chen, C. Wang, and E. Wang, “WISDOM: WiFi-based contactless multiuser activity recognition,”IEEE Internet of Things Journal, vol. 10, no. 2, pp. 1876–1886, 2023

    work page 2023

  15. [15]

    MultiSenseX: A sustainable solution for multi-human activity recognition and localiza- tion in smart environments,

    H. Rizk, A. Elmogy, M. Rihan, and H. Yamaguchi, “MultiSenseX: A sustainable solution for multi-human activity recognition and localiza- tion in smart environments,”AI, vol. 6, no. 1, 2025

    work page 2025

  16. [16]

    WiFi-based multiuser identity, location, and activity recognition using InceptionTime-Attention networks,

    J. Wang, M. A. A. Al-Qaness, S. Ni, and C. Tang, “WiFi-based multiuser identity, location, and activity recognition using InceptionTime-Attention networks,”IEEE Sensors Journal, vol. 25, no. 7, pp. 12 389–12 398, 2025

    work page 2025

  17. [17]

    WiMANS: A benchmark dataset for WiFi-based multi-user activity sensing,

    S. Huang, K. Li, D. You, Y . Chen, A. Lin, S. Liu, X. Li, and J. A. McCann, “WiMANS: A benchmark dataset for WiFi-based multi-user activity sensing,” inComputer Vision – ECCV 2024, A. Leonardis, E. Ricci, S. Roth, O. Russakovsky, T. Sattler, and G. Varol, Eds. Cham: Springer Nature Switzerland, 2025, pp. 72–91

    work page 2024

  18. [18]

    WiHAR: From WiFi channel state information to unobtrusive human activity recognition,

    M. Muaaz, A. Chelli, and M. P ¨atzold, “WiHAR: From WiFi channel state information to unobtrusive human activity recognition,” in2020 IEEE 91st Vehicular Technology Conference (VTC2020-Spring), 2020, pp. 1–7

    work page 2020

  19. [19]

    Utilizing deep learning models in CSI-based human activity recognition,

    E. Shalaby, N. ElShennawy, and A. Sarhan, “Utilizing deep learning models in CSI-based human activity recognition,”Neural Computing and Applications, vol. 34, no. 8, pp. 5993–6010, 2022

    work page 2022

  20. [20]

    WiFi CSI based passive human activity recognition using attention based BLSTM,

    Z. Chen, L. Zhang, C. Jiang, Z. Cao, and W. Cui, “WiFi CSI based passive human activity recognition using attention based BLSTM,”IEEE Transactions on Mobile Computing, vol. 18, no. 11, pp. 2714–2724, 2019

    work page 2019

  21. [21]

    Two-stream convolution augmented transformer for human activity recognition,

    B. Li, W. Cui, W. Wang, L. Zhang, Z. Chen, and M. Wu, “Two-stream convolution augmented transformer for human activity recognition,” Proceedings of the AAAI Conference on Artificial Intelligence, vol. 35, no. 1, pp. 286–293, 2021

    work page 2021

  22. [22]

    Vision transformers for human activity recognition using WiFi channel state information,

    F. Luo, S. Khan, B. Jiang, and K. Wu, “Vision transformers for human activity recognition using WiFi channel state information,”IEEE Internet of Things Journal, vol. 11, no. 17, pp. 28 111–28 122, 2024

    work page 2024

  23. [23]

    CSI- based human activity recognition using convolutional neural networks,

    P. F. Moshiri, M. Nabati, R. Shahbazian, and S. A. Ghorashi, “CSI- based human activity recognition using convolutional neural networks,” in2021 11th International Conference on Computer Engineering and Knowledge (ICCKE), 2021, pp. 7–12

    work page 2021

  24. [24]

    MultiTrack: Multi-user tracking and activity recognition using commodity WiFi,

    S. Tan, L. Zhang, Z. Wang, and J. Yang, “MultiTrack: Multi-user tracking and activity recognition using commodity WiFi,” inProceedings of the 2019 CHI Conference on Human Factors in Computing Systems. Association for Computing Machinery, 2019, pp. 1–12

    work page 2019

  25. [25]

    IMar: Multi-user continuous action recognition with WiFi signals,

    J. He and W. Yang, “IMar: Multi-user continuous action recognition with WiFi signals,”Proc. ACM Interact. Mob. Wearable Ubiquitous Technol., vol. 6, no. 3, 2022

    work page 2022

  26. [26]

    Multi-user gesture recognition using WiFi,

    R. H. Venkatnarayan, G. Page, and M. Shahzad, “Multi-user gesture recognition using WiFi,” inProceedings of the 16th Annual International Conference on Mobile Systems, Applications, and Services, ser. MobiSys ’18. Association for Computing Machinery, 2018, pp. 401–413

    work page 2018

  27. [27]

    WiMAR: A WiFi-based multi-user human activity recognition system via dynamic component separation,

    Y . Zhou, Y . Liu, C. Liu, and Y . Lu, “WiMAR: A WiFi-based multi-user human activity recognition system via dynamic component separation,” in2025 IEEE 102nd Vehicular Technology Conference (VTC2025-Fall), 2025, pp. 1–5

    work page 2025

  28. [28]

    Lightweight regularized network for multilabel indoor HAR in mul- tiuser CSI environments with uncertainty quantification,

    F. Miao, C. Liu, Z. Lu, L. Shan, O. Takyu, T. Ohtsuki, and G. Gui, “Lightweight regularized network for multilabel indoor HAR in mul- tiuser CSI environments with uncertainty quantification,”IEEE Internet of Things Journal, vol. 13, no. 4, pp. 6475–6484, 2026

    work page 2026

  29. [29]

    A survey on CSI-based Wi-Fi sensing datasets and models with a focus on reproducibility,

    I. Guarino, D. Carra, M. Cominelli, F. Gringoli, and R. Lo Cigno, “A survey on CSI-based Wi-Fi sensing datasets and models with a focus on reproducibility,”Computer Communications, vol. 249, p. 108431, 2026

    work page 2026

  30. [30]

    A survey on Wi-Fi sensing generalizability: Taxonomy, techniques, datasets, and future research prospects,

    F. Wang, T. Zhang, W. Xi, H. Ding, G. Wang, D. Zhang, Y . Cui, F. Liu, J. Han, J. Xu, and T. X. Han, “A survey on Wi-Fi sensing generalizability: Taxonomy, techniques, datasets, and future research prospects,” 2025, arXiv preprint arXiv:2503.08008

    work page arXiv 2025

  31. [31]

    Zero-effort cross-domain gesture recognition with Wi-Fi,

    Y . Zheng, Y . Zhang, K. Qian, G. Zhang, Y . Liu, C. Wu, and Z. Yang, “Zero-effort cross-domain gesture recognition with Wi-Fi,” inProceed- ings of the 17th Annual International Conference on Mobile Systems, Applications, and Services, ser. MobiSys ’19. Association for Com- puting Machinery, 2019, pp. 313–325

    work page 2019

  32. [32]

    MetaFormer: Domain- adaptive WiFi sensing with only one labelled target sample,

    B. Sheng, R. Han, F. Xiao, Z. Guo, and L. Gui, “MetaFormer: Domain- adaptive WiFi sensing with only one labelled target sample,”Proc. ACM Interact. Mob. Wearable Ubiquitous Technol., vol. 8, no. 1, 2024

    work page 2024

  33. [33]

    Wi-Learner: Towards one-shot learning for cross-domain Wi-Fi based gesture recognition,

    C. Feng, N. Wang, Y . Jiang, X. Zheng, K. Li, Z. Wang, and X. Chen, “Wi-Learner: Towards one-shot learning for cross-domain Wi-Fi based gesture recognition,”Proc. ACM Interact. Mob. Wearable Ubiquitous Technol., vol. 6, no. 3, 2022

    work page 2022

  34. [34]

    Device-free human activity recognition with identity-based transfer mechanism,

    B. Wu, T. Jiang, J. Yu, X. Ding, S. Wu, and Y . Zhong, “Device-free human activity recognition with identity-based transfer mechanism,” in2021 IEEE Wireless Communications and Networking Conference (WCNC), 2021, pp. 1–6

    work page 2021

  35. [35]

    AirFi: Empowering WiFi-based passive human gesture recognition to unseen environment via domain generalization,

    D. Wang, J. Yang, W. Cui, L. Xie, and S. Sun, “AirFi: Empowering WiFi-based passive human gesture recognition to unseen environment via domain generalization,”IEEE Transactions on Mobile Computing, vol. 23, no. 2, pp. 1156–1168, 2024

    work page 2024

  36. [36]

    Environment-independent Wi- Fi human activity recognition with adversarial network,

    Z. Wang, S. Chen, W. Yang, and Y . Xu, “Environment-independent Wi- Fi human activity recognition with adversarial network,” inICASSP 2021 - 2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2021, pp. 3330–3334

    work page 2021

  37. [37]

    DA-HAR: Dual adversarial network for environment-independent WiFi human activity recognition,

    L. Sheng, Y . Chen, S. Ning, S. Wang, B. Lian, and Z. Wei, “DA-HAR: Dual adversarial network for environment-independent WiFi human activity recognition,”Pervasive and Mobile Computing, vol. 96, p. 101850, 2023

    work page 2023

  38. [38]

    Privacy- preserving cross-environment human activity recognition,

    L. Zhang, W. Cui, B. Li, Z. Chen, M. Wu, and T. S. Gee, “Privacy- preserving cross-environment human activity recognition,”IEEE Trans- actions on Cybernetics, vol. 53, no. 3, pp. 1765–1775, 2023

    work page 2023

  39. [39]

    A ConvNet for the 2020s,

    Z. Liu, H. Mao, C.-Y . Wu, C. Feichtenhofer, T. Darrell, and S. Xie, “A ConvNet for the 2020s,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), June 2022, pp. 11 976–11 986

    work page 2022

  40. [40]

    Deep residual learning for image recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), June 2016

    work page 2016

  41. [41]

    Searching for MobileNetV3,

    A. Howard, M. Sandler, G. Chu, L.-C. Chen, B. Chen, M. Tan, W. Wang, Y . Zhu, R. Pang, V . Vasudevan, Q. V . Le, and H. Adam, “Searching for MobileNetV3,” inProceedings of the IEEE/CVF International Confer- ence on Computer Vision (ICCV), October 2019

    work page 2019

  42. [42]

    Focal loss for dense object detection,

    T.-Y . Lin, P. Goyal, R. Girshick, K. He, and P. Dollar, “Focal loss for dense object detection,” inProceedings of the IEEE International Conference on Computer Vision (ICCV), Oct 2017

    work page 2017

  43. [43]

    A deep learning-based human identification system with Wi-Fi CSI data augmentation,

    H. Mo and S. Kim, “A deep learning-based human identification system with Wi-Fi CSI data augmentation,”IEEE Access, vol. 9, pp. 91 913– 91 920, 2021

    work page 2021