Skeleton-Graph: Long-Term 3D Motion Prediction From 2D Observations Using Deep Spatio-Temporal Graph CNNs

Several applications such as autonomous driving, augmented reality and\nvirtual reality require a precise prediction of the 3D human pose. Recently, a\nnew problem was introduced in the field to predict the 3D human poses from\nobserved 2D poses. We propose Skeleton-Graph, a deep spatio-temporal graph CNN\nmodel that predicts the future 3D skeleton poses in a single pass from the 2D\nones. Unlike prior works, Skeleton-Graph focuses on modeling the interaction\nbetween the skeleton joints by exploiting their spatial configuration. This is\nbeing achieved by formulating the problem as a graph structure while learning a\nsuitable graph adjacency kernel. By the design, Skeleton-Graph predicts the\nfuture 3D poses without divergence in the long-term, unlike prior works. We\nalso introduce a new metric that measures the divergence of predictions in the\nlong term. Our results show an FDE improvement of at least 27% and an ADE of 4%\non both the GTA-IM and PROX datasets respectively in comparison with prior\nworks. Also, we are 88% and 93% less divergence on the long-term motion\nprediction in comparison with prior works on both GTA-IM and PROX datasets.\nCode is available at https://github.com/abduallahmohamed/Skeleton-Graph.git\n