Video-Based Person Re-identification with Improved Temporal Attention And Spatial Memory

Video-based Person Re-identification (reID) aims to retrieve and query videos of people with the same identity across multiple cameras. Temporal attention mechanism is widely used in video-based person re-identification, but it cant effectively use the effective features in poor-quality frames. In this paper, we propose a video-based Person Re-identification model with improved temporal attention and spatial memory(ITASM), which mainly includes encoder, spatial memory module, attention residual module, feature fusion optimization module and weight distribution module. In the feature fusion optimization module, we extract the effective features of each frame, and fuse them with the features output by the attention residual module, so that the final temporal sequence features will contain more effective features of poor-quality frames during the weighted summation. In this way, we strengthen the effective information in poor-quality frames, and solve the above problem to a certain extent. The validity of the model proposed in this paper has been verified on MARS and iLIDS-vid datasets.