Deep Cross-Modality Alignment for Multi-Shot Person Re-IDentification