m monad-knowledge FIIT STU · open research vault
dataset

ut-har

Also known as: UT-HAR, UT HAR, UTHAR, UT-HAR dataset

Description

UT-HAR (Yousefi et al., University of Toronto) is one of the earliest publicly released WiFi CSI activity-recognition datasets and remains the de facto baseline in CSI HAR papers. It contains continuous CSI traces of seven daily activities (e.g. lie down, fall, walk, run, sit down, stand up, pick up) collected on commodity Intel 5300 NICs, sliced into per-activity segments.

Modality / size

  • Modality: WiFi CSI from Intel 5300 NIC, 30 subcarriers.
  • Subjects / scenarios: small set of subjects in a single indoor environment.
  • Labels: 7 activity classes.

Used by (papers)

  • Quasi-universal CSI HAR baseline, cited next to Widar, NTU-Fi, SignFi in benchmark tables.
  • Used in the SenseFi benchmark suite.

Notes

  • Public dataset; available from the original Yousefi et al. release.
  • Distinct from the CSI-HAR-Dataset by Moshiri et al. (see csi-har-dataset); some papers conflate them so check the citation when in doubt.
Where it appears

4 vault papers evaluate on this dataset

Titles and DOIs only — no abstracts, no analyses.

  • SenseFi: A library and benchmark on deep-learning-empowered WiFi human sensing 2023 DOI ↗
  • Deep Learning-Enhanced Human Sensing with Channel State Information: A Survey 2026 DOI ↗
  • A survey on CSI-based Wi-Fi sensing datasets and models with a focus on reproducibility 2026 DOI ↗
  • Context-Aware Predictive Coding: A Representation Learning Framework for WiFi Sensing 2024 DOI ↗
Co-mentioned with
Convolutional Neural Network×4 Widar×4 Deep Learning×4 Activity Recognition×3 GRU×3 Long Short-Term Memory×3 SignFi×3 Intel 5300×3 Fall Detection×3 Gesture Recognition×3