Joint Learning of Single-Image and Cross-Image Representations for Person Re-identification | Synapse