ChatPPG.
Paper

TransfHAR: Self-Supervised Wrist Representations for On-Demand Activity Recognition

Aidan Bradshaw, Riku Arakawa, Xin Liu, Karan Ahuja
arXiv preprint2026IMU
arXiv Cite
PreprintFirst posted: Aug 16, 2026Added to atlas: Oct 11, 2026Source
Citations—Not available
GitHub stars—Not available
HF downloads—Not available

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.

TransfHAR pretrains on coarse wrist IMU activities like sitting and walking and transfers to fine-grained ones like snapping or stirring that were never in pretraining. It runs as a smartwatch app where users add their own activities from a few demonstrations. In a lab study with 10 participants it reached 86.7% balanced accuracy with five examples per activity.

  • Signals: Accelerometer
  • Result: Across three offline cross-dataset evaluations, TransfHAR matches or exceeds fully supervised baselines that use complete label sets with equal or additional sensor channels, by 6.2 balanced-accuracy points on average.
  • Result: In an in-lab study with 10 participants each performing seven novel wrist activities, TransfHAR reaches 86.7% balanced accuracy across participants with five examples per class and 90.4% when updated from a single one-minute recording per class.
At a glance
Venue
arXiv preprint
Published
2026
DOI
Not reported
arXiv
2608.15861 (opens in a new tab)
PMID
Not reported
Licence
Not reported
Signals
IMU
Form factor
Not reported
Models
Subjects
10 participants
Code
Not reported
Added
Oct 11, 2026

Abstract

Fine-grained wrist activity recognition can support applications such as procedural step guidance and context-aware assistance, yet acquiring labeled data for every new task, user, and activity granularity remains a bottleneck. We present TransfHAR, a self-supervised wrist IMU framework for on-demand, fine-grained activity recognition by learning transferable motion priors from global, unlabeled activities. We show that self-supervised pretraining on coarse wrist IMU activities (e.g., sitting, walking, exercise) learns motion structure rich enough to transfer to fine-grained manipulative, gestural, and procedural activities (e.g., snapping, stirring, waving) that are absent from pretraining. We implement TransfHAR as a real-time smartwatch application that lets users define and expand their own activity set for personalized recognition from only a few demonstrations. Across three offline cross-dataset evaluations, TransfHAR matches or exceeds fully supervised baselines that use complete label sets with equal or additional sensor channels, by 6.2 balanced-accuracy points on average. In an in-lab study with 10 participants each performing seven novel wrist activities, TransfHAR reaches 86.7% balanced accuracy across participants with five examples per class and 90.4% when updated from a single one-minute recording per class. These results indicate that broad self-supervised wrist pretraining provides an effective foundation for on-demand fine-grained activity recognition.

Verbatim fromarXiv· arXiv metadata (abstracts are released under CC0)

TransfHAR: Self-Supervised Wrist Representations for On-Demand Activity Recognition: ChatPPG research summary card

How it works

How InrtlWrist works: Setup → Method → ResultSetup: Cohort (10 participants); Signals (Accelerometer). Method: InrtlWrist. Result: Result (Across three offline cross-dataset evaluations, TransfHAR matches or exceeds fully supervised baselines that use complete label sets with equal or additional sensor channels, by 6.2 balanced-accuracy points on average.); Result (In an in-lab study with 10 participants each performing seven novel wrist activities, TransfHAR reaches 86.7% balanced accuracy across participants with five examples per class and 90.4% when updated from a single one-minute recording per class.).SetupCohort: 10 participantsCohort10 participantsSignals: AccelerometerxyzSignalsAccelerometerMethodInrtlWristResultResult: Across three offline cross-dataset evaluations, TransfHAR matches or exceeds fully supervised baselines that use complete label sets with equal or additional sensor channels, by 6.2 balanced-accuracy points on average.ResultAcross three offline cross-dataset…Result: In an in-lab study with 10 participants each performing seven novel wrist activities, TransfHAR reaches 86.7% balanced accuracy across participants with five examples per class and 90.4% when updated from a single one-minute recording per class.ResultIn an in-lab study with 10 participants…
Cohort
10 participants

Models introduced or evaluated· 1

Foundation model
InrtlWrist
masked-autoencoder wrist IMU encoder (IMU-Bench reference model)
IMU

Cite

@article{transfharselfsupervisedw2026,
  title = {{TransfHAR: Self-Supervised Wrist Representations for On-Demand Activity Recognition}},
  author = {Aidan Bradshaw and Riku Arakawa and Xin Liu and Karan Ahuja},
  year = {2026},
  journal = {arXiv preprint},
  eprint = {2608.15861}
}

History

  1. Added to the ChatPPG research hub

More on ChatPPG