BadCM: Invisible Backdoor Attack Against Cross-Modal Learning.

Semantic-Aware Adversarial Training for Reliable Deep Hashing Retrieval.