Reproduce “A study of the effect of JPG compression on adversarial images” 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

Gintare Karolina Dziugaite, Zoubin Ghahramani, Daniel M. Roy

The authors test whether JPG recompression can partly reverse adversarial image perturbations. Using FGSM attacks on pretrained OverFeat with ImageNet validation images, they compare accuracy before attack, after attack, and after recompression at JPG quality 75. For ε=1/255, accuracy rises from 0.23 to 0.48, below the clean-image 0.58; recovery shrinks for larger tested perturbations.

Summary by alphaXiv.

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