So I'm working on this rPPG features for PAD. Right now, I implemented three different algorithms to retrieve the pulse signal, and they are all based on what I previously did in bob.rppg.base (and hence, bob.pad.face should then depend on it eventually).
At the moment, all pulse extraction algorithm are implemented as bob.pad.face.Extractor, but the more I think about it, I think that they should be implemented as bob.pad.face.Preprocessor for mainly two reasons:
Solves the problem with rotated videos directly (instead of having an "empty" preprocessor copying data)
Gives the ability to then extract features from the pulse (i.e. frequency spectrum, for instance)
Guillaume HEUSCH authored