Universal 3-Dimensional Perturbations for Black-Box Attacks on Video Recognition Systems

Universal 3-Dimensional Perturbations for Black-Box Attacks on Video Recognition Systems

Explore a 16-minute IEEE conference talk on developing universal 3-dimensional perturbations for black-box attacks on video recognition systems. Delve into adversarial tags, previous video attacks, and the proposed threat model. Learn about the innovative ProcedureNoise framework, techniques for improving processor abilities, and methods to address boundary effects. Examine expandable results, family models, and adverse examples. Gain insights into real-time performance and attacks against main differences in video recognition systems.