Model

DailyActFormer-v2 (DAF; PULSE benchmark multimodal action-recognition baseline; SCUT)

Year
2026
Form factor
Body worn
Venue listedCatalog paper date: 2026Added to atlas: Oct 29, 2026

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.

DailyActFormer-v2 (DAF; PULSE benchmark multimodal action-recognition baseline; SCUT). NeurIPS 2026 Evaluations & Datasets Track (spotlight) 2026. Modalities: IMU, EMG, MoCap, eye-tracking, pressure.

Signals
IMU, EMG, eye-tracking, MoCap, pressure
Form factor
body-worn
Year
2026
Architecture

Per-modality stems → fusion + norm → positional embedding → Transformer encoder → pooling → multi-head classifier (verb / noun / composite / hand); optional S9 motion-primitive pre-training (F5) with whole-modality dropout, transferring everything except head.* (pulse/models/t2/seqpred.py, pulse/models/f5/pretrain.py)

Pretraining data

PULSE scenario S9 (pre-training volunteers disjoint from the T2 pool)

License

Code MIT; no weights released (cached encoders are built locally)

At a glance
Type
Task model
Form factor
Body worn
Parameters
Not reported
Architecture
Per-modality stems → fusion + norm → positional embedding → Transformer encoder → pooling → multi-head classifier (verb / noun / composite / hand); optional S9 motion-primitive pre-training (F5) with whole-modality dropout, transferring everything except head.* (pulse/models/t2/seqpred.py, pulse/models/f5/pretrain.py)
Venue
NeurIPS 2026 Evaluations & Datasets Track (spotlight)
Year
2026
arXiv
Not reported
DOI
Not reported
Code
Open
Weights
Not released
Licence
Code MIT; no weights released (cached encoders are built locally)
Caveats
Not reported
Datasets
Not reported
Added
Oct 29, 2026
Updated
Oct 29, 2026

Original description

Reference code of the paper "PULSE: A Synchronized Five-Modality Dataset for Sensorimotor Coordination in Long-Horizon Daily Activities" (NeurIPS 2026 Evaluations and Datasets Track, spotlight). It loads the public PULSE dataset, defines the volunteer-stratified 5-fold cross-validation splits, trains every baseline of the benchmark (T1-T7, F5), runs the F0 case study, and regenerates the benchmark tables, figures and numbers of the paper from the run results.

The benchmark data are the dataset's tag v1.1 (commit e27b1bb6bb35f714705ae8685507f71773c065eb, 3,001 files, 22.0 GB: sensor streams, annotations, metadata). All numbers in the paper were computed on that tag with the commands in section 3.

Verbatim fromHugging Face· Description from the Hugging Face card; licence as stated on the card

How it works

How DailyActFormer-v2 (DAF; PULSE benchmark multimodal action-recognition baseline; SCUT) works: Input signals → Model → Downstream tasksInput signals: IMU (Accelerometer, 3-axis); EMG (Electrophysiology); eye-tracking; MoCap. Dataset: PULSE. Model: DailyActFormer-v2 (DAF; PULSE benchmark multimodal action-recognition baseline; SCUT) (Transformer).Input signalsIMU: Accelerometer, 3-axisxyzIMUAccelerometer, 3-axisEMG: ElectrophysiologyEMGElectrophysiologyeye-trackingeye-trackingMoCapMoCapDataset: PULSEDatasetPULSEModel×NDailyActFormer-v2 (DAF; PULSE…Transformer

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

Pretraining data
PULSE scenario S9 (pre-training volunteers disjoint from the T2 pool)
Architecture
Per-modality stems → fusion + norm → positional embedding → Transformer encoder → pooling → multi-head classifier (verb / noun / composite / hand); optional S9 motion-primitive pre-training (F5) with whole-modality dropout, transferring everything except head.* (pulse/models/t2/seqpred.py, pulse/models/f5/pretrain.py)
Inputs
IMU, EMG, eye-tracking, MoCap, pressure

Datasets used

No public datasets linked yet.

Papers· 1

Cite this model

Cite the publication recorded for this model. Check it against the publisher before submitting.

DailyActFormer-v2 (DAF; PULSE benchmark multimodal action-recognition baseline; SCUT). (2026). NeurIPS 2026 Evaluations & Datasets Track.
@inproceedings{pulsedailyactfor2026dailyactformerv2,
  title         = {{DailyActFormer-v2 (DAF; PULSE benchmark multimodal action-recognition baseline; SCUT)}},
  year          = {2026},
  booktitle     = {NeurIPS 2026 Evaluations \& Datasets Track}
}

Source metadata: 2026

Record updated Oct 29, 2026

History

  1. Added to the ChatPPG research hub