PPG-DaLiA — Reiss et al., Sensors 2019

Mirror of the official PPG-DaLiA dataset hosted on the UCI Machine Learning Repository (dataset 495).

UCI's own download endpoint occasionally truncates the response, so this mirror exists for robust re-download. The contents are byte-identical to the official UCI zip.

Citation

A. Reiss, I. Indlekofer, P. Schmidt, K. Van Laerhoven, "Deep PPG: Large-Scale Heart Rate Estimation with Convolutional Neural Networks", Sensors 19(14), 2019.

Always cite the original Reiss et al. paper and link the UCI dataset page.

Files

File	Size	Notes
ppg_dalia.zip	2.86 GB	Outer zip — contains data.zip (2.86 GB nested) + readme.pdf (818 KB)

Structure (inner data.zip)

PPG_FieldStudy/
  S1/  ...  S15/                  # 15 subjects
    S{n}.pkl                      # main pickle (~1.5 GB), see schema below
    S{n}_RespiBAN.h5              # chest signals at 700 Hz (HDF5)
    S{n}_E4.zip                   # wrist Empatica E4 raw
    S{n}_activity.csv             # activity transitions with start times
    S{n}_quest.csv                # questionnaire (age/gender/height/weight/skin/sport)
PPG_FieldStudy_readme.pdf

Pickle schema (S{n}.pkl)

{
  'rpeaks': (N_R,) int32,                     # ECG R-peak indices
  'signal': {
    'chest': {                                 # RespiBAN, fs=700 Hz
      'ACC':  (T, 3) float64,                  # 3-axis ACC
      'ECG':  (T, 1) float64,
      'EMG':  (T, 1) float64,
      'EDA':  (T, 1) float64,
      'Temp': (T, 1) float32,
      'Resp': (T, 1) float64,
    },
    'wrist': {                                 # Empatica E4
      'ACC':  (T_acc, 3) float64,              # fs=32 Hz
      'BVP':  (T_bvp, 1) float64,              # fs=64 Hz, PPG (BVP)
      'EDA':  (T_eda, 1) float64,              # fs=4 Hz
      'TEMP': (T_eda, 1) float64,              # fs=4 Hz
    }
  },
  'label':         (N_lbl,)        float64,    # HR labels (8 s window, 2 s slide)
  'activity':      (T_eda, 1)      float64,    # activity labels (fs=4 Hz aligned)
  'questionnaire': {WEIGHT, Gender, AGE, HEIGHT, SKIN, SPORT},
  'subject':       'S{n}',
}

For S1: chest T = 6,448,400 → session length = 9,212 s ≈ 153.5 min.

Activities

BASELINE / STAIRS / SOCCER / CYCLING / DRIVING / LUNCH / WALKING / WORKING / CLEAN_BASELINE + NO_ACTIVITY transitions, with start-time stamps in S{n}_activity.csv.