A Graph Attention Spatio-temporal Convolutional Network for 3D Human Pose Estimation in Video