
A novel attack framework, LogoStyleFool, is proposed to fool video recognition systems by superimposing a stylized logo on the input video, achieving superior attack performance and semantic preservation compared to existing patch-based attacks.


coremsg

attacking-video-recognition-systems-via-adversarial-logo-style-transfer


Attacking Video Recognition Systems via Adversarial Logo Style Transfer


title_rewrite


StyleFool, a black-box video adversarial attack framework, leverages style transfer to efficiently fool video classification systems with unrestricted perturbations.


efficient-black-box-video-adversarial-attacks-via-style-transfer


Efficient Black-box Video Adversarial Attacks via Style Transfer
