Reproduce “Perceptually Constrained Adversarial Attacks” 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
Muhammad Zaid Hameed, Andras Gyorgy
The authors constrain structural similarity index (SSIM) in image-classification attacks to retain image structure while changing a classifier’s prediction. They compare SSIM and SSIM_E with ENet on adversarially trained defenses. On CIFAR-10, both SSIM attacks reach 100% success without an SSIM threshold; against Feature defense, SSIM exceeds ENet by 20 points at SSIM ≥ 0.9, but does not guarantee imperceptibility.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.