ChatPPG.
Join free
Model

ActiNet (OxWearables activity-intensity classifier)

Parameters
10M
Dataset
CAPTURE-24
Year
2024
Form factor
Wrist
IMUOpen codeWeights: open
Cite
Figure or first page from the ActiNet (OxWearables activity-intensity classifier) source (opens the source in a new tab)
Acquah, Aidan et al., "ActiNet: An Open-Source Tool for Activity Intensity Classification of Wrist-Worn Accelerometry Using Self-Supervised Deep Learning", arXiv:2510.01712, CC BY 4.0
UnconfirmedCatalog paper date: 2024Added to atlas: Oct 17, 2026
CitationsView DOI record (opens in a new tab)GitHub stars16Repository-wide · checked Oct 19, 2026 (opens in a new tab)

Activity counts measure attention, not model quality. Missing counts are unknown. Publication status follows the linked source; a listed venue alone does not confirm peer review.

Released activity-intensity classifier from OxWearables: the self-supervised HARNet wrist-accelerometer encoder (harnet-ukb, base model) fine-tuned on CAPTURE-24, followed by an HMM and post-hoc sleep correction (per the ActiNet paper). Shipped as the actinet Python package; two classifiers are published (walmsley 4-class intensity, willetts).

Parameters
10M
Signals
IMU
Form factor
Wrist
License
Oxford Academic Use Licence (non-commercial)
Year
2024
Architecture

HARNet backbone (harnet30 from OxWearables/ssl-wearables): modified 18-layer 1D-conv ResNet-V2 feature extractor, 1,024-d output, input 900 x 3 (30 s at 30 Hz triaxial wrist ACC); all layers fine-tuned with a fully connected head on CAPTURE-24, then HMM smoothing and post-hoc sleep correction

Key results4 SOTA

All results

Results as reported by the authors; not independently reproduced. A SOTA star marks the best value among results reported under the same evaluation protocol; ties go to the earlier publication.

Embed this badge (4 badges for READMEs and project pages)
  • SOTA on ChatPPG: Activity recognition · CAPTURE-24
    Markdown
    [![SOTA on ChatPPG: Activity recognition · CAPTURE-24](https://chatppg.com/research/badges/actinet--har-capture-24-cohen-kappa.svg)](https://chatppg.com/research/models/actinet#key-results)
    HTML
    <a href="https://chatppg.com/research/models/actinet#key-results"><img src="https://chatppg.com/research/badges/actinet--har-capture-24-cohen-kappa.svg" alt="SOTA on ChatPPG: Activity recognition · CAPTURE-24" height="20"></a>
  • SOTA on ChatPPG: Activity recognition · WISDM
    Markdown
    [![SOTA on ChatPPG: Activity recognition · WISDM](https://chatppg.com/research/badges/actinet--har-wisdm-macro-f1-har.svg)](https://chatppg.com/research/models/actinet#key-results)
    HTML
    <a href="https://chatppg.com/research/models/actinet#key-results"><img src="https://chatppg.com/research/badges/actinet--har-wisdm-macro-f1-har.svg" alt="SOTA on ChatPPG: Activity recognition · WISDM" height="20"></a>
  • SOTA on ChatPPG: Activity recognition · RealWorld
    Markdown
    [![SOTA on ChatPPG: Activity recognition · RealWorld](https://chatppg.com/research/badges/actinet--har-realworld-har-macro-f1-har.svg)](https://chatppg.com/research/models/actinet#key-results)
    HTML
    <a href="https://chatppg.com/research/models/actinet#key-results"><img src="https://chatppg.com/research/badges/actinet--har-realworld-har-macro-f1-har.svg" alt="SOTA on ChatPPG: Activity recognition · RealWorld" height="20"></a>
  • SOTA on ChatPPG: Activity recognition · CAPTURE-24
    Markdown
    [![SOTA on ChatPPG: Activity recognition · CAPTURE-24](https://chatppg.com/research/badges/actinet--har-capture-24-macro-f1-har.svg)](https://chatppg.com/research/models/actinet#key-results)
    HTML
    <a href="https://chatppg.com/research/models/actinet#key-results"><img src="https://chatppg.com/research/badges/actinet--har-capture-24-macro-f1-har.svg" alt="SOTA on ChatPPG: Activity recognition · CAPTURE-24" height="20"></a>
At a glance
Type
FM adaptation
Modalities
IMU
Form factor
Wrist
Parameters
10M
Architecture
HARNet backbone (harnet30 from OxWearables/ssl-wearables): modified 18-layer 1D-conv ResNet-V2 feature extractor, 1,024-d output, input 900 x 3 (30 s at 30 Hz triaxial wrist ACC); all layers fine-tuned with a fully connected head on CAPTURE-24, then HMM smoothing and post-hoc sleep correction
Venue
Not reported
Year
2024
Authors
arXiv
Not reported
DOI
Not reported
Code
Open
Weights
Open
Licence
Oxford Academic Use Licence (non-commercial)
Caveats
Not reported
Datasets
Papers
Devices
Added
Oct 17, 2026
Updated
Oct 18, 2026

Key numbers

0.86
Cohen's κ ↑ · Human activity recognition (IMU)
CAPTURE-24
#1 / 1
0.818
Macro-F1 ↑ · Human activity recognition (IMU)
CAPTURE-24
#1 / 1
0.661
Macro-F1 ↑ · Human activity recognition (IMU)
RealWorld HAR
#1 / 1

How it works

How ActiNet (OxWearables activity-intensity classifier) works: Input signals → Model → Downstream tasksInput signals: IMU (Accelerometer, 3-axis). Dataset: CAPTURE-24. Model: ActiNet (OxWearables activity-intensity classifier), 10M parameters (Convolutional network). Downstream tasks: Activity recognition.Input signalsIMU: Accelerometer, 3-axisxyzIMUAccelerometer, 3-axisDataset: CAPTURE-24DatasetCAPTURE-24ModelActiNet (OxWearables…10M parametersConvolutional networkDownstream tasksActivity recognitionActivity recognition

Datasets linked in the hub; pretraining vs evaluation roles follow the paper.

Architecture
HARNet backbone (harnet30 from OxWearables/ssl-wearables): modified 18-layer 1D-conv ResNet-V2 feature extractor, 1,024-d output, input 900 x 3 (30 s at 30 Hz triaxial wrist ACC); all layers fine-tuned with a fully connected head on CAPTURE-24, then HMM smoothing and post-hoc sleep correction
Inputs
IMU
Approach
HAR · wrist accelerometer · HARNet · HMM · OxWearables · activity intensity

Devices· 2

Axivity AX3 photo
Axivity AX3
Axivity · Wrist
IMU
Photo: Open Movement, Newcastle University / project docs, CC BY 3.0
OMG Life Autographer photo
OMG Life Autographer
OMG Life · Other
camera
Photo: OMG Plc / Wikimedia Commons, CC BY 3.0

Full photo credits, sources and licences are on each device page.

Leaderboard rows· 4

All leaderboards
TaskDatasetMetricValueRank
Human activity recognition (IMU)CAPTURE-24Cohen's κ ↑0.86#1 / 1
Human activity recognition (IMU)CAPTURE-24Macro-F1 ↑0.818#1 / 1
Human activity recognition (IMU)RealWorld HARMacro-F1 ↑0.661#1 / 1
Human activity recognition (IMU)WISDM AR v1.1Macro-F1 ↑0.741#1 / 1

Datasets· 1

Papers· 1

Authors· 3

Aidan Acquah
Author
Aidan Acquah
Big Data Institute, University of Oxford, Oxford, UK; Nuffield Department of Population Health, University of Oxford, Oxford, UK
SC
Author
Shing Chan
Big Data Institute, University of Oxford, Oxford, UK; Nuffield Department of Population Health, University of Oxford, Oxford, UK
AD
Author
Aiden Doherty
Big Data Institute, University of Oxford, Oxford, UK; Nuffield Department of Population Health, University of Oxford, Oxford, UK

Cite

APA

ActiNet (OxWearables activity-intensity classifier). (2024).

BibTeX

@misc{actinet2024actinet,
  title         = {{ActiNet (OxWearables activity-intensity classifier)}},
  year          = {2024}
}

Built from the hub record. Check it against the publisher before submitting.

History

  1. Record updated

  2. Added to the ChatPPG research hub

Software: actinet is listed in Wearable Signal Tools, with its licence, commercial use and last commit.

Connections

Open in the graph explorer
Dataset: RealWorld HAR (uses dataset)RealWorld HARRealWorld…Dataset: CAPTURE-24 (uses dataset)CAPTURE-24CAPTURE-24Dataset: WISDM AR v1.1 (uses dataset)WISDM AR v1.1WISDM AR v…Signal: IMU (uses signal)IMUIMUPaper: ActiNet: an open-source tool for activity intensity classification of wrist-worn accelerometry using self-supervised deep learning (paper and model)ActiNet: an open-so…ActiNet: a…Author: Aidan Acquah (authored by)Aidan AcquahAidan Acqu…Author: Aiden Doherty (authored by)Aiden DohertyAuthor: Shing Chan (authored by)Shing ChanActiNet (OxWearables activity…ActiNet (OxWearab…
Signals 1
IMU

Same tasks, data and authors

Papers

Models

Datasets