Summary of Transferable Adversarial Face Attack with Text Controlled Attribute, by Wenyun Li et al.
Transferable Adversarial Face Attack with Text Controlled Attribute
by Wenyun Li, Zheng Zhang, Xiangyuan Lan, Dongmei Jiang
First submitted to arxiv on: 16 Dec 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary This paper proposes a novel Text Controlled Attribute Attack (TCA^2) to generate photorealistic adversarial impersonation faces guided by natural language. The method uses a category-level personal softmax vector to precisely guide impersonation attacks, and combines data and model augmentation strategies for transferable attacks on unknown target models. A generative model, Style-GAN, is used to synthesize impersonated faces with desired attributes. Experiments on two high-resolution face recognition datasets demonstrate the effectiveness of TCA^2, which can generate natural text-guided adversarial impersonation faces with high transferability. The method is also evaluated on real-world face recognition systems, such as Face++ and Aliyun. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper creates fake faces that look like someone else’s, but only if you tell it what kind of person the fake face should be (e.g., “a young woman”). This is useful for testing security cameras or other facial recognition systems. The new method works by using a special type of math to change how a computer sees a face. It can make the fake faces look really real, and it’s good at fooling most computers. The paper shows that this method can work on real-world cameras, which could be useful for testing or even causing mischief. |
Keywords
» Artificial intelligence » Face recognition » Gan » Generative model » Softmax » Transferability
