Reproduce “Exploring Frequency Adversarial Attacks for Face Forgery Detection” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Shuai Jia, Chao Ma, Taiping Yao, Bangjie Yin, Shouhong Ding, Xiaokang Yang

A new approach for generating adversarial examples against face forgery detectors focuses on perturbing images in the frequency domain, leading to more visually imperceptible attacks. The proposed hybrid method achieved high success rates against diverse detectors, including state-of-the-art frequency-based models, and demonstrated improved transferability compared to existing spatial-domain attacks.

Key insights

  • Perturbing images in the frequency domain can generate more visually imperceptible adversarial examples while remaining highly effective against face forgery detectors.
  • A hybrid approach integrating both spatial and frequency domain perturbations enhances the generalization and transferability of adversarial attacks across different detector architectures.
  • Attacking all frequency bands generally yields the best performance, with low-frequency bands contributing significantly due to their content information.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.