Motion-Excited Sampler: Video Adversarial Attack with Sparked Prior

TL;DR

This paper introduces a novel motion-aware adversarial attack method for video models, leveraging intrinsic movement patterns and regional motion to generate effective noise with fewer queries.

Contribution

It proposes a motion-excited sampler that utilizes relative motion as sparked prior to improve attack efficiency on video classification models.

Findings

Successfully attacks various video models with fewer queries.

Outperforms existing methods on four benchmark datasets.

Effectively exploits intrinsic movement patterns for adversarial attacks.

Abstract

Deep neural networks are known to be susceptible to adversarial noise, which are tiny and imperceptible perturbations. Most of previous work on adversarial attack mainly focus on image models, while the vulnerability of video models is less explored. In this paper, we aim to attack video models by utilizing intrinsic movement pattern and regional relative motion among video frames. We propose an effective motion-excited sampler to obtain motion-aware noise prior, which we term as sparked prior. Our sparked prior underlines frame correlations and utilizes video dynamics via relative motion. By using the sparked prior in gradient estimation, we can successfully attack a variety of video classification models with fewer number of queries. Extensive experimental results on four benchmark datasets validate the efficacy of our proposed method.