Reproduce “Semantic Image Attack for Visual Model Diagnosis” 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
Jinqi Luo, Zhaoning Wang, Chen Henry Wu, Dong Huang, Fernando De la Torre
SIA uses a differentiable face generator and target-model gradients to generate adversarial examples and diagnose vision-model vulnerabilities. It updates facial attributes and image pixels together, recording average attribute movement during attacks as sensitivity. In an eyeglasses-classifier ablation, joint updates reached 68.20% attack success, versus 32.67% for attribute-only updates. The evidence is limited to face-based white-box attacks on differentiable models.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.