Reproduce “SemBind: Binding Diffusion Watermarks to Semantics Against Black-Box Forgery 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
Xin Zhang, Zijin Yang, Kejiang Chen, Linfeng Ma, Weiming Zhang, Nenghai Yu
SemBind makes latent-based watermark verification depend on an image-derived semantic code, masking the watermark latent during generation and reversing the mask from the submitted image during verification. In a matched Stable Diffusion 2.1 COCO reprompting test, detection fell from 0.99 to 0.06 and bit accuracy from 0.9788 to 0.4894. This result is limited to the tested attack setting.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.