Supervised rPPG training requires high fidelity synchronous PPG waveform labels. However not alldatasets contain such high quality labels. In these cases we offer the option to train on synchronousPPG "pseudo" labels derived through a signal processing methodology as described by [41 ]. Theselabels are produced through POS-generated [ 31] PPG waveforms, which are then bandpass filteredaround the normal heart-rate frequencies (cut-off frequencies of 0.70 and 3.0 Hz), and finallyamplitude normalized using a Hilbert-signal envelope. The tight filtering and envelope normalizationresults in a strong periodic proxy signal, but at the cost of limited signal morphology.
Aplicar synthetic PPG signals in the videos.