Reproduce “Reject and Report: Forgery-Evident Undetectable Watermarks for Generative Image Models” 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

Yihan Wang, Haochen Sun, Xi He, Yaoliang Yu, N. Asokan

Tripwire-WM adds a semantic commitment to diffusion-watermark magnitudes, allowing verification to distinguish content-changing forgeries that retain a watermark. It hashes a draft image’s semantics, embeds the commitment in magnitudes, and checks it after inversion. On MS-COCO with GS++, it alerted on 93.9% of reprompting and 100.0% of imprinting attacks; results are setting-specific and depend on the secret hash and sign-layer security.

Summary by alphaXiv.

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