Table 2 Performance (MAE ± SE, in bpm) of methods across all datasets

From: The reliability of remote photoplethysmography under low illumination and elevated heart rates

Method

PURE

COHFACE

CHILL

DeepPhys7

3.1 ± 2.1

4.4 ± 1.3

9.1 ± 2.3

TS-CAN25

10.1 ± 5.5

4.1 ± 2.1

3.2 ± 0.2

PhysNet8

4.0 ± 1.9

1.6 ± 0.4

4.1 ± 1.3

rPPGNet9

8.8 ± 4.5

3.4 ± 1.8

2.0 ± 0.4

GREEN21

10.1 ± 2.9

7.1 ± 0.7

2.6 ± 0.6

CHROM6

8.9 ± 2.1

10.2 ± 0.6

1.8 ± 0.6

POS5

7.5 ± 2.0

11.8 ± 0.7

1.1 ± 0.1

ICA23

4.9 ± 2.0

7.4 ± 0.6

5.4 ± 0.9

Dummy

15.6 ± 2.2

9.7 ± 0.7

10.8 ± 0.7

  1. Deep learning-based methods (top) are evaluated using 10-fold cross-validation with participant-wise splits. Signal processing-based methods (bottom) are evaluated on the whole datasets. A dummy estimator (bottom) that always predicts the mean heart rate of the training set is included as a baseline. Bold values indicate the lowest MAE (best performance) for each dataset.
  2. SE standard error.
Back to article page