[CVPR2026] V-Attack: Targeting Disentangled Value Features for Controllable Adversarial Attacks on LVLMs
By chatting or signing in you agree to the Terms and chat-message logging (revocable in History).