Reproduce “VideoShield: Regulating Diffusion-based Video Generation Models via Watermarking” 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
Runyi Hu, Jie Zhang, Yiming Li, Jiwei Li, Qing Guo, Han Qiu, Tianwei Zhang
VideoShield embeds watermark bits in diffusion noise during video generation, repeating them for robust recovery and using encrypted template bits for localization. DDIM inversion recovers bits; frame matching identifies reordered or unmatched frames, while spatial comparisons produce tamper masks. On ModelScope Crop&Drop, its five-metric average is 0.906 versus PSCC-Net’s 0.529, but falls to 0.761 on SVD and temporal results have no baseline.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.