Summary of Face Reconstruction Transfer Attack As Out-of-distribution Generalization, by Yoon Gyo Jung et al.
Face Reconstruction Transfer Attack as Out-of-Distribution Generalization
by Yoon Gyo Jung, Jaewoo Park, Xingbo Dong, Hojin Park, Andrew Beng Jin Teoh, Octavia Camps
First submitted to arxiv on: 2 Jul 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 A novel attack framework is proposed to reconstruct face images that can evade face recognition systems by transferring attacks on unseen encoders. The approach, termed Face Reconstruction Transfer Attack (FRTA), is formulated as an out-of-distribution (OOD) generalization problem and solved using Averaged Latent Search and Unsupervised Validation with pseudo target (ALSUV). ALSUV reconstructs faces by searching the latent space of StyleGAN2 through multiple latent optimization, latent optimization trajectory averaging, and unsupervised validation with a pseudo target. The effectiveness and generalizability of the method are demonstrated on widely used face datasets, accompanied by extensive ablation studies and qualitative, quantitative, and visual analyses. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper aims to create fake faces that can trick face recognition systems even if they’re not trained on those specific faces before. To do this, it uses a special kind of search called Averaged Latent Search and Unsupervised Validation with pseudo target (ALSUV) to find the right combination of features to make a fake face look like someone else. The method is tested on some big datasets and shown to be good at making fake faces that can fool these systems. |
Keywords
» Artificial intelligence » Face recognition » Generalization » Latent space » Optimization » Unsupervised
