Reproduce “Diffusion-Based Image Editing for Breaking Robust Watermarks” 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
Yunyi Ni, Finn Carter, Ze Niu, Emily Davis, Bo Zhang
Diffusion regeneration attacks invisible image watermarks by adding noise and denoising watermarked images; decoder-guided sampling uses decoder gradients to steer denoising away from the embedded message. In tests, VINE exact-message recovery fell from 18% to 0% with guidance, while PSNR fell from 29.4 to 26.0 dB. This result assumes decoder access and applies to the test set, not all watermarks.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.