AI Paper Atlas

Research map / Image generation / Autoregressive and latent generation

Attention to Neural Plagiarism: Diffusion Models Can Plagiarize Your Copyrighted Images!

Zihang Zou, Boqing Gong, Liqiang Wang

ICCV 2025 · Image generation › Autoregressive and latent generation · approach: Robust diffusion watermarking · 1 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add attention-perturbation shims instead of plain regeneration (Regen) with same noisy startNeural Plagiarism attack · MS-COCO DwtDctSvd, late startbit accuracy: 0.64 -> 0.52Improved
add attention-perturbation shims instead of RegenNeural Plagiarism attack · MS-COCO RivaGAN, late startbit accuracy: 0.60 -> 0.56No clear effect
add shims with early start from inverse latent vs RegenNeural Plagiarism attack · MS-COCO DwtDctSvd, early startFID: 21.36 -> 71.14Improved
add shims with early start from inverse latent vs RegenNeural Plagiarism attack · MS-COCO RivaGAN, early startbit accuracy: 0.59 -> 0.52Improved
start from inverse latent instead of noisy latent in late-start setting (ablation)Neural Plagiarism attack · invisible watermark removalsemantic changeHurt
attack Stable Signature watermark with latent shimsNeural Plagiarism attack · MS-COCO Stable SignatureT@1%F: 1.00 -> 0.00Improved
attack Tree-Ring watermark with latent shimsNeural Plagiarism attack · MS-COCO Tree-RingT@1%F: 1.00 -> 0.98No clear effect
replace RivaGAN watermark with new watermark after attackNeural Plagiarism attack · RivaGAN ambiguity attackbit accuracy: BA(w*) 0.75 -> 1.00; BA(w) 1.00 -> 0.75Improved
embed alternative watermark channel after attackNeural Plagiarism attack · Tree-Ring ambiguity attackT@1%F: T@1%F(w*) 0.00 -> 1.00; T@1%F(w) 1.00 -> 0.99Improved
vary shim timestep (t=100 to t=700) on real copyrighted images (ablation)Neural Plagiarism attack · real-world copyrighted imagessemantic modificationImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy