face-rhythm documentation!¶
While many tools exist to transform behavioral videos into useful low-dimensional representations, very few both stay stable to small alterations in the recording conditions and require no pre-training of large models. Furthermore, few techniques capitalize on rich spectral data in creating factors. To tackle these existing restraints, we propose Face Rhythm, a novel behavioral quantification pipeline combining optic flow, spectral analysis, and tensor component analysis to robustly generate meaningful, low-dimensional factors capturing both spectral and spatial patterns in behavioral videos.