Pith. sign in

REVIEW 4 major objections 4 minor 33 references

A Multi-Modal IoT Node for Energy-Efficient Environmental Monitoring with Edge AI Processing

T0 review · 4 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read Small sensor node runs 143 hours on one 600 mAh battery

desk verdict A solid, citable hardware-integration paper whose abstract muddles two different operating points: the 42% energy saving comes from a low-accuracy 192×192 config, while the 143h/15.5mW headline uses the 512×512 model that actually consumes more energy than the streaming baseline. read the letter →

arxiv 2507.14165 v2 pith:VY7YFT5N submitted 2025-07-08 eess.SP cs.SYeess.IVeess.SY

Philip Wiese , Victor Kartsch , Marco Guermandi , Luca Benini This is my paper · ORCID
classification eess.SPcs.SYeess.IVeess.SY
keywords environmentalmonitoringedgeAITinyMLoccupancydetectionmulti-modalsensingultra-low-powerSoCGAP9indoorairquality
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 claims that a single compact IoT board, 17 by 38 millimeters, can carry eleven environmental sensors plus an RGB camera and still run for days on one small battery while doing AI on the device. The authors report an average system power of 15.5 mW in a representative deployment, which corresponds to more than 143 hours of operation on a 600 mAh, 3.7 V battery. They show that a quantized YOLOv5-based head detector running on the GAP9 chip uses 42 percent less energy than streaming raw camera frames over Bluetooth Low Energy. They also combine occupancy detection with adaptive sampling of the environmental sensors to build a full-stack indoor air quality monitor. If these numbers hold, battery-powered multi-modal sensing with local inference becomes practical without cloud processing.

What carries the argument

The load-bearing mechanism is a dual-SoC base board: GAP9, a nine-core RISC-V ultra-low-power system-on-chip with an NE16 hardware accelerator for quantized neural networks, paired with an nRF5340 wireless SoC and a power-management IC. A modular shield adds the eleven sensors, including CO2, VOC, barometric pressure, ambient light, UV, temperature, humidity, IMU, GNSS, microphone, and camera. The argument is carried by an energy accounting chain: 2.63 mJ per 192-by-192 inference, 490 mJ for a complete sensor readout, a 1 mW assumed sleep power, and a 600 mAh battery, which together produce the claimed lifetime.

What would settle it

Perform a full discharge test with a fresh 600 mAh, 3.7 V battery, running one 512x512 inference every 2 s and one complete sensor readout every 60 s, and measure hours until shutdown; a result well under 143 h would show that sleep power, PMIC efficiency, or battery capacity differs from the assumptions.

Watch

Extended reading notes

Core claim

The central claim is that edge AI and multi-modal sensing can coexist on an MCU-class node without sacrificing battery life. In the paper's implementation, the HM0360 camera feeds GAP9, which performs debayering, white balancing, downscaling, quantized YOLOv5p inference, and non-maximum suppression, then forwards only occupancy counts to the nRF5340 for wireless reporting. The 192-by-192 input variant is the most efficient, giving 13.6 percentage points of mAP@50 per millijoule, and the full cycle of one image inference plus one readout of all sensors is estimated at 929 mJ. That yields an average power of 15.5 mW and a runtime over 143 hours on a 600 mAh cell, and the paper contrasts this with Raspberry Pi-based people counting at roughly 380 times higher power.

Load-bearing premise

The lifetime claim rests on unmeasured assumptions: a constant 1 mW system sleep power, a 600 mAh battery delivering its full rated energy without conversion or aging losses, and measured per-cycle energies staying valid over the whole discharge.

Editorial extensions

If this is right

  • Occupancy monitoring no longer requires sending images off the node, so a deployment on a small rechargeable battery can run for nearly six days.
  • Adding more sensing modalities does not force higher wireless power draw, because on-device inference filters the high-bandwidth camera data before transmission.
  • During periods of low occupancy the sampling rate can be reduced, so real deployments may last longer than the conservative always-occupied estimate.
  • The same platform has enough on-chip compute to host additional workloads, making on-device forecasting of indoor air quality a plausible next step.

Reading between the lines

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

  • The 143-hour figure is an arithmetic projection based on assumed sleep power and battery efficiency; the paper reports no end-to-end discharge test, so the true lifetime depends on those assumptions holding.
  • A head-to-head comparison against a low-power camera node with separate sensors would show how much of the advantage comes from the GAP9's accelerator versus from system-level integration and adaptive sampling.
  • The 42 percent energy saving is measured against raw-frame BLE streaming; comparing against JPEG-compressed transmission or motion-triggered capture would give a smaller advantage, since those alternatives also spend on-device energy.
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 presents SENSEI, a compact 17x38 mm2 multi-modal environmental IoT node built around the GAP9 SoC and an nRF5340 radio, integrating 11 sensors plus an RGB camera. The authors train and quantize a YOLOv5-based occupancy detector (YOLOv5p), measure per-frame power/energy across input resolutions from 64x64 to 512x512, report a 42% energy saving versus raw BLE image streaming at the 192x192 operating point, and combine occupancy detection with periodic environmental sensor readout in a full-stack scenario. They estimate an average system power of 15.5 mW and a runtime of about 143 h on a 600 mAh, 3.7 V battery. The paper also characterizes the environmental sensors indoors and outdoors and compares the platform with Raspberry Pi-based systems.

Significance. If the reported claims hold, the platform is a useful contribution: it packs an unusually wide sensor set with a programmable ultra-low-power AI accelerator in a small form factor, and the measured power traces in Table I and Figure 3 are internally consistent and reproducible from the stated numbers. The yawning gap in system power versus Raspberry Pi-class baselines (Table II) is meaningful, and the open-hardware release of the base board and shield is a concrete strength. However, the headline claims are currently presented more strongly than the evidence supports: the 143 h lifetime is an extrapolation from assumed sleep power and a single sensor-readout energy measurement, and the 42% energy saving is tied to a low-accuracy operating point that is not the one used in the full-system demonstration. With measured lifetime data and a more careful framing of the energy-saving claim, the work would be a solid systems contribution.

major comments (4)
  1. [V.B and Conclusion] The 143 h runtime in the abstract and conclusion is an arithmetic extrapolation, not a measured quantity. In Section V.B the system-level average of 15.5 mW is obtained from (30 x 12.99 mJ + 490 mJ + 60 mJ)/60 s under the assumption of a 1 mW sleep power and a fixed 2 s/60 s sampling schedule, and no end-to-end discharge test is reported. The 600 mAh battery is assumed to deliver its full rated energy with no PMIC conversion, BLE transmission, or aging losses. The manuscript should either report a measured lifetime or qualify the abstract and conclusion as estimates based on these assumptions.
  2. [IV.B and Table I] The 42% energy saving is specific to the 192x192 input (Table I, mAP50=35.1%), where 2.63 mJ/frame plus 2 mJ sleep gives 4.63 mJ versus the 7.86 mJ BLE-streaming baseline. The full-system demonstration in Section V.B uses the 512x512 network (mAP50=84.7%), whose 12.99 mJ/frame alone exceeds the 7.86 mJ baseline; at that operating point, on-device inference is not energy-saving relative to the paper's own streaming baseline. Moreover, the 7.86 mJ baseline transmits 320x240 grayscale images, not the 512x512 color inputs used in the full system, so the comparison conflates resolution and processing. The abstract and contribution list should attach the 42% figure to its operating point, and the full-system comparison should use a resolution- and content-matched streaming baseline.
  3. [V.B and Abstract] Adaptive sample rates are announced in the abstract and Section I but are not implemented or measured anywhere. Section V.B conservatively fixes the image interval at 2 s and the environmental readout at 60 s and assumes the room remains occupied; the only discussion of contextual sampling is a single sentence in Section V.A. Either implement the adaptive policy and report its measured power and lifetime, or remove the claim from the abstract and contribution list.
  4. [V.A and V.B] The full-system average power is dominated by the 490 mJ 'total energy required to read out all sensors' from Section V.A, yet the manuscript gives no breakdown of how this value was obtained, over what duration, or whether it includes heater warm-up transients for the SCD41, BME680, and SGP41. Because this single value contributes about 8.2 mW of the 15.5 mW average, the measurement methodology and the dependence of this energy on the sampling interval need to be documented.
minor comments (4)
  1. [IV.B] The phrase 'excluding the power datacenter power' is unclear; it should read 'excluding the power consumed by the data center' or 'excluding off-device processing power.'
  2. [IV.A] The manuscript does not report the number of training/validation images or the test split used to compute the mAP values in Table I; adding this information would improve reproducibility.
  3. [IV.B, V.B, Conclusion] The lifetime figures are variously stated as 'over 40 days,' 'almost 6 days,' and 'over 143 h'; please keep the qualification 'estimated' consistent across these statements so readers do not confuse extrapolation with measurement.
  4. [Table II] Table II compares platforms on total power only; adding a column for measured energy per sensing or inference cycle would make the comparison more direct and fairer given the different sampling rates.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the headline figures are arithmetic consequences of measured power traces, and self-citations are acknowledged hardware reuse.

full rationale

The paper makes no theoretical derivation; its headline numbers are arithmetic consequences of independently measured quantities. The 42% energy saving is computed as (7.86 − 4.63)/7.86 mJ from a measured raw-streaming baseline and measured per-frame GAP9/camera/memory energy, with a stated 1 mW sleep assumption (Section IV.B). The 143 h runtime follows from measured 929 mJ per 2 s/60 s cycle and the rated 600 mAh battery energy (Section V.B). No parameter is fitted to reproduce a desired result, and no 'prediction' is defined in terms of the quantity it claims to predict. Self-citations to the SENSEI base board [12] and prior PULP/BioGAP work [15] are hardware-reuse acknowledgments, explicitly footnoted as 'not developed as part of this work,' and they do not carry the argument. The apparent tension between the 42% saving at 192×192 input and the 512×512 full-system demo is a representativeness/consistency question, not a circularity: the energy figures at both operating points are measured inputs, and the paper does not use the 42% claim as an input to the 143 h estimate. No step of the derivation chain reduces to its own input, so no circularity is present.

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

The central energy and lifetime claims rest on a handful of hand-chosen assumptions: a 1 mW sleep power, 10 mW BLE transmission power, maximum BLE throughput, fixed sampling intervals, and full battery energy conversion. The mAP results are treated as representative of occupancy accuracy without a described test split. No free parameters are fitted to make a theory match data, and no theoretical entities are introduced.

free parameters (4)
  • System sleep power = 1 mW
    Hand-assumed in Section V.B for all idle time; the 143 h runtime scales inversely with it.
  • BLE transmission power = 10 mW
    Assumed BLE transmission power in Section IV.B for the raw-streaming baseline and long-term estimates; higher power would change the 42% figure.
  • BLE throughput = maximum BLE throughput, not quantified
    Raw image transmission time is computed at maximum BLE throughput; real throughput lowers baseline energy and changes the claimed saving.
  • Sampling schedule = occupancy 2 s, sensors 60 s
    Fixed in Section V.B for the lifetime estimate; the adaptive sampling advertised in the abstract is not evaluated.
assumptions (4)
  • domain assumption Battery energy equals 600 mAh times 3.7 V with no conversion loss
    Used in Sections IV.B and V.B to convert battery capacity into runtime; ignores PMIC efficiency, discharge curve, and aging.
  • domain assumption Component power measurements are additive and representative
    Section IV.A measures GAP9, memory, and camera in isolation; Section V.B adds the 490 mJ sensor readout measured from the nRF to form the 929 mJ per-cycle estimate.
  • domain assumption mAP on the Roboflow human-head dataset transfers to real occupancy counting
    Section IV.A trains on [33]; no test-set split or deployment-time counting accuracy is reported.
  • ad hoc to paper Raw uncompressed BLE image streaming is an appropriate energy baseline
    Section IV.B defines this baseline and excludes off-device processing; it is the paper's own choice and drives the 42% savings headline.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Multi-Modal IoT Node for Energy-Efficient Environmental Monitoring with Edge AI Processing." pith.science (2026). https://pith.science/paper/VY7YFT5N

@misc{pith2026250714165,
  author       = {Pith},
  title        = {Pith review of: A Multi-Modal IoT Node for Energy-Efficient Environmental Monitoring with Edge AI Processing},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VY7YFT5N}},
  note         = {Machine review of arXiv:2507.14165}
}
read the original abstract

The widespread adoption of Internet of Things (IoT) technologies has significantly advanced environmental monitoring (EM) by enabling cost-effective and scalable sensing solutions. Concurrently, machine learning (ML) and artificial intelligence (AI) are introducing powerful tools for the efficient and accurate analysis of complex environmental data. However, current IoT platforms for environmental sensing are typically limited to a narrow set of sensors, preventing a comprehensive assessment of environmental conditions and lacking sufficient computational capabilities to support the deployment of advanced ML and AI algorithms on the edge. To overcome these limitations, we introduce a compact (17x38 mm2), multi-modal, MCU-based environmental IoT node integrating 11 sensors, including CO2 concentration, volatile organic compounds (VOCs), light intensity, UV radiation, pressure, temperature, humidity, visual sensing via an RGB camera, and precise geolocation through a GNSS module. It features GAP9, a parallel ultra-low-power system-on-chip, enabling real-time, energy-efficient edge processing of advanced ML models directly on-device. We implemented a YOLOv5-based occupancy detection pipeline (0.3 M parameters, 42 MOP per inference), demonstrating 42% energy savings over raw data streaming. Additionally, we present a smart indoor air quality (IAQ) monitoring setup that combines occupancy detection with adaptive sample rates, achieving operational times of up to 143 h on a single compact 600 mAh, 3.7 V battery. Our platform lays the groundwork for innovative applications such as predictive indoor IAQ, enabling efficient AI-driven on-edge forecasting for energy-efficient and autonomous, proactive pollution-mitigation control strategies

Figures

Figures reproduced from arXiv: 2507.14165 by the authors.

Figure 1
Figure 1. Our platform combines a compact dual-SoC base board (a) with a modular sensing shield (b) to enable energy-efficient, near-sensor computation for [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. System setup for on-device occupancy detection. The camera is [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Evaluation of runtime performance and energy efficiency across [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Environmental sensor readings in two test scenarios: (a) indoor office [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

33 extracted references · 32 canonical work pages

  1. [1]

    Degrees matter: Understanding the science of climate change,

    United Nations, “Degrees matter: Understanding the science of climate change,” n.d. [Online]. Available: https://www.un.org/en/climatechange/ science/climate-issues/degrees-matter TABLE II COMPARISON OF SOA ENVIRONMENTAL IOT MONITORING SYSTEMS WITH EMBEDDED PROCESSING CAPABILITIES Author Sensors Embedded Processor Application Power Moursi et al. [4] 3 – 5...

    work page

  2. [2]

    Urbantwin: Creating a digital twin for sustainable and resilient swiss cities,

    UrbanTwin, “Urbantwin: Creating a digital twin for sustainable and resilient swiss cities,” 2023. [Online]. Available: https://urbantwin.ch/

    work page 2023

  3. [3]

    A comparative review on multi-modal sensors fusion based on deep learning,

    Q. Tang, J. Liang, and F. Zhu, “A comparative review on multi-modal sensors fusion based on deep learning,” Signal Processing, vol. 213, p. 109165, 2023

    work page 2023

  4. [4]

    An iot enabled system for enhanced air quality monitoring and prediction on the edge,

    A. S. Moursi et al. , “An iot enabled system for enhanced air quality monitoring and prediction on the edge,” Complex & Intelligent Systems, vol. 7, p. 2923–2947, 2021

    work page 2021

  5. [5]

    Machine learning at the network edge: A survey,

    M. S. Murshed et al., “Machine learning at the network edge: A survey,” ACM Computing Surveys (CSUR) , vol. 54, no. 8, pp. 1–37, 2021

    work page 2021

  6. [6]

    On-device domain learning for keyword spotting on low-power extreme edge embedded systems,

    C. Cioflan et al. , “On-device domain learning for keyword spotting on low-power extreme edge embedded systems,” in 2024 IEEE 6th International Conference on AI Circuits and Systems (AICAS) , 2024, pp. 6–10

    work page 2024

  7. [7]

    Endoscopy image classification for wireless capsules with cnns on microcontroller-based platforms,

    P. Busia, A. Pinna, and P. Meloni, “Endoscopy image classification for wireless capsules with cnns on microcontroller-based platforms,” in International Workshop on Design and Architectures for Signal and Image Processing. Springer, 2025, pp. 57–68

    work page 2025

  8. [8]

    Embedded machine learning using microcontrollers in wearable and ambulatory systems for health and care applications: A review,

    M. S. Diab and E. Rodriguez-Villegas, “Embedded machine learning using microcontrollers in wearable and ambulatory systems for health and care applications: A review,” IEEE Access , vol. 10, pp. 98 450– 98 474, 2022

    work page 2022

Show all 33 references
  1. [9]

    Towards energy-aware tinyml on battery-less iot devices,

    A. Sabovic et al. , “Towards energy-aware tinyml on battery-less iot devices,” Internet of Things , vol. 22, p. 100736, 2023

    2023

  2. [10]

    A comprehensive survey on wireless sensor node hardware platforms,

    F. Karray et al. , “A comprehensive survey on wireless sensor node hardware platforms,” Computer Networks, vol. 144, pp. 89–110, 2018

    2018

  3. [11]

    Microcontroller unit-based wireless sensor network nodes: A review,

    M. Wu, Y . Zhang, and L. T. Yang, “Microcontroller unit-based wireless sensor network nodes: A review,” Sensors, vol. 22, no. 22, p. 8937, 2022

    2022

  4. [12]

    Biogap-ultra: A modular edge-ai platform for wear- able multimodal biosignal acquisition and processing,

    S. Frey et al. , “Biogap-ultra: A modular edge-ai platform for wear- able multimodal biosignal acquisition and processing,” arXiv preprint arXiv:2508.13728, 2025

    2025

  5. [13]

    Iot-enabled smart agriculture: architecture, applica- tions, and challenges,

    V . K. Quy et al. , “Iot-enabled smart agriculture: architecture, applica- tions, and challenges,” Applied Sciences, vol. 12, no. 7, p. 3396, 2022

    2022

  6. [14]

    Ag-iot for crop and environment monitoring: Past, present, and future,

    N. Chamara et al., “Ag-iot for crop and environment monitoring: Past, present, and future,” Agricultural Systems, vol. 203, p. 103497, 2022

    2022

  7. [15]

    Biogap: A 10-core fp-capable ultra-low power iot processor, with medical-grade afe and ble connectivity for wearable biosignal processing,

    S. Frey et al. , “Biogap: A 10-core fp-capable ultra-low power iot processor, with medical-grade afe and ble connectivity for wearable biosignal processing,” in 2023 IEEE International Conference on Omni- layer Intelligent Systems (COINS) . IEEE, 2023, pp. 1–7

    2023

  8. [16]

    Structural health monitoring system with narrow- band iot and mems sensors,

    F. Di Nuzzo et al. , “Structural health monitoring system with narrow- band iot and mems sensors,” IEEE Sensors Journal, vol. 21, no. 14, pp. 16 371–16 380, 2021

    2021

  9. [17]

    Wireless sensor networks and machine learning meet climate change prediction,

    T. Anh Khoa et al. , “Wireless sensor networks and machine learning meet climate change prediction,” International Journal of Communica- tion Systems, vol. 34, no. 3, p. e4687, 2021

    2021

  10. [18]

    A survey on matching theory for distributed computation offloading in iot-fog-cloud systems: Perspectives and open issues,

    H. Tran-Dang and D.-S. Kim, “A survey on matching theory for distributed computation offloading in iot-fog-cloud systems: Perspectives and open issues,” IEEE Access, vol. 10, pp. 118 353–118 369, 2022

    2022

  11. [19]

    A low-cost platform for environmental smart farming monitoring system based on iot and uavs,

    F. A. Almalki et al. , “A low-cost platform for environmental smart farming monitoring system based on iot and uavs,” Sustainability, vol. 13, no. 11, p. 5908, 2021

    2021

  12. [20]

    Monitoring water quality metrics of ponds with iot sensors and machine learning to predict fish species survival,

    M. M. Islam et al. , “Monitoring water quality metrics of ponds with iot sensors and machine learning to predict fish species survival,” Microprocessors and microsystems, vol. 102, p. 104930, 2023

    2023

  13. [21]

    Thingspeak-based environmental monitoring sys- tem using iot,

    N. Sindhwani et al., “Thingspeak-based environmental monitoring sys- tem using iot,” in Proceedings of the 2022 Seventh International Con- ference on Parallel, Distributed and Grid Computing (PDGC) . IEEE, 2022, pp. 675–680

    2022

  14. [22]

    Microcontroller unit-based wireless sensor network nodes: A review,

    A. Khalifeh et al., “Microcontroller unit-based wireless sensor network nodes: A review,” Sensors, vol. 22, no. 22, 2022

    2022

  15. [23]

    Visual sensors hardware platforms: A review,

    D. G. Costa, “Visual sensors hardware platforms: A review,” IEEE Sensors Journal, vol. 20, no. 8, pp. 4025–4035, 2020

    2020

  16. [24]

    Tinyml is coming: Edge ai trends, architectures, and challenges,

    L. Song et al. , “Tinyml is coming: Edge ai trends, architectures, and challenges,” ACM Computing Surveys (CSUR) , 2024

    2024

  17. [25]

    Smart air quality monitoring iot-based infrastructure for industrial environments,

    L. Garc ´ıa et al. , “Smart air quality monitoring iot-based infrastructure for industrial environments,” Sensors, vol. 22, no. 23, p. 9221, 2022

    2022

  18. [26]

    Deep learning prediction of thunderstorm severity using remote sensing weather data,

    Y . Essa et al., “Deep learning prediction of thunderstorm severity using remote sensing weather data,” IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing , vol. 15, pp. 4004– 4013, 2022

    2022

  19. [27]

    Predicting algorithm of thunderstorm days in the northern region of chile using convolution neural network,

    J. Monta ˜na et al. , “Predicting algorithm of thunderstorm days in the northern region of chile using convolution neural network,”IEEE Access, vol. 12, pp. 116 017–116 025, 2024

    2024

  20. [28]

    Assessment of indoor air quality in academic buildings using iot and deep learning,

    M. Al-Khafajiy et al. , “Assessment of indoor air quality in academic buildings using iot and deep learning,” Sustainability, vol. 14, no. 12, p. 7015, 2022

    2022

  21. [29]

    Enabling smart agriculture by implementing artificial intelligence and embedded sensing,

    A. Sharma et al., “Enabling smart agriculture by implementing artificial intelligence and embedded sensing,” Computers & Industrial Engineer- ing, vol. 165, p. 107936, 2022

    2022

  22. [30]

    Enabling precision agriculture through embedded sensing with artificial intelligence,

    D. Shadrin et al. , “Enabling precision agriculture through embedded sensing with artificial intelligence,” IEEE Transactions on Instrumenta- tion and Measurement , vol. 69, no. 7, pp. 4103–4113, 2020

    2020

  23. [31]

    Automated pest detection with dnn on the edge for precision agriculture,

    A. Albanese, M. Nardello, and D. Brunelli, “Automated pest detection with dnn on the edge for precision agriculture,” IEEE Journal on Emerging and Selected Topics in Circuits and Systems , vol. 11, no. 3, pp. 458–467, 2021

    2021

  24. [32]

    Adaptive indoor people-counting system based on edge ai computing,

    M.-H. Yen et al. , “Adaptive indoor people-counting system based on edge ai computing,” IEEE Transactions on Emerging Topics in Computational Intelligence, vol. 8, no. 1, pp. 255–263, 2023

    2023

  25. [33]

    dataset human head dataset,

    C. A. C. Zuleta, “dataset human head dataset,” apr 2023. [Online]. Avail- able: https://universe.roboflow.com/carlos-alberto-castro-zuleta-cnopi/ dataset-human-head

    2023

Pith tools

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