Seq-Masks: Bridging the gap between appearance and gait modeling for video-based person re-identification.