REGINA - Reasoning Graph Convolutional Networks in Human Action\n Recognition

It is known that the kinematics of the human body skeleton reveals valuable\ninformation in action recognition. Recently, modeling skeletons as\nspatio-temporal graphs with Graph Convolutional Networks (GCNs) has been\nreported to solidly advance the state-of-the-art performance. However,\nGCN-based approaches exclusively learn from raw skeleton data, and are expected\nto extract the inherent structural information on their own. This paper\ndescribes REGINA, introducing a novel way to REasoning Graph convolutional\nnetworks IN Human Action recognition. The rationale is to provide to the GCNs\nadditional knowledge about the skeleton data, obtained by handcrafted features,\nin order to facilitate the learning process, while guaranteeing that it remains\nfully trainable in an end-to-end manner. The challenge is to capture\ncomplementary information over the dynamics between consecutive frames, which\nis the key information extracted by state-of-the-art GCN techniques. Moreover,\nthe proposed strategy can be easily integrated in the existing GCN-based\nmethods, which we also regard positively. Our experiments were carried out in\nwell known action recognition datasets and enabled to conclude that REGINA\ncontributes for solid improvements in performance when incorporated to other\nGCN-based approaches, without any other adjustment regarding the original\nmethod. For reproducibility, the REGINA code and all the experiments carried\nout will be publicly available at https://github.com/DegardinBruno.\n