Fetching the paper…

AnyAttack: Towards Large-scale Self-supervised Adversarial Attacks on Vision-language Models · Around