Rank-GCN for Robust Action Recognition

We present a robust skeleton-based action recognition method with graph convolutional network (GCN) that uses the new adjacency matrix, called Rank-GCN. In Rank-GCN, biggest change from previous approaches is how matrix generated to accumulate features neighboring nodes by re-defining “adjacency.” The which we call rank ranking all according metrics including Euclidean distance of interest, whereas GCNs methods used only 1-hop construct adjacency. By adopting find not performance improvements but also robustness against swapping, location shifting and dropping certain nodes. fact human-made wins deep-learning-based implies there are still some parts need touch humans. expect our Rank-GCN can make especially when predicted human joints less accurate unstable.