Smoothing Adversarial Domain Attack and P-Memory Reconsolidation for Cross-Domain Person Re-Identification

• Corresponding author bility in real-world scenarios. When applied to a new scenario, they suffer from serious performance degradation, e.g., from 92.0% to 47.5% on the Market-1501 dataset [52]. How to transfer the knowledge from a labeled source domain to an unlabeled target domain for person re-ID and how to exploit the specific knowledge from an unlabeled target domain to boost person re-ID methods become hot topics in the current person re-ID community. It is widely called cross-domain person re-ID.

Recently, lots of cross-domain person re-ID methods