Spatiotemporal Attention on Sliced Parts for Video-based Person Re-identification

Video-based person re-identification has been the focus of computer vision technology owing to its great demands in reality. The key to this work is to mine useful information of each video sequence at spatial and temporal level. In this paper, we propose a new solution for video-based person re-identification with spatial and temporal attention on sliced parts of images. The spatial attention tackles the problem of region-level noise such as occlusions, background changes, misalignments and the temporal attention helps discover complementary information at sequencelevel. By combining both attention mechanisms, our solution surpasses the state-of-the-art algorithms on three popular video-based person re-identification datasets by a large margin.