05-30 (CVPR 2018) Diversity Regularized Spatiotemporal Attention for Video-based Person Re-identification