AI Paper Atlas

Research map / Adversarial robustness and security / Attacks on text and graphs

RecoverMark: Robust Watermarking for Localization and Recovery of Manipulated Faces

Haonan An, Xiaohui Ye, Guang Hua, Yihang Tao, Hangcheng Cao, Xiangyu Yu, Yuguang Fang

CVPR 2026 · Adversarial robustness and security › Attacks on text and graphs · approach: Neural network watermarking · 0 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
embed the face as a robust watermark in the background (RecoverMark) instead of fragile watermarksproactive manipulation localization · CelebA, Structpix2pix, regeneration attacklocalization F1: 0.105 (OmniGuard) -> 0.855Improved
RecoverMark vs OmniGuard under lattice attack (unseen)RecoverMark · CelebA, Structpix2pix, lattice attacklocalization F1: 0.113 -> 0.842Improved
RecoverMark vs EditGuard under lattice attackRecoverMark · CelebA, SD inpaint, lattice attacklocalization F1: 0.704 -> 0.756Improved
RecoverMark vs OmniGuard under patch removalRecoverMark · CelebA, SD inpaint, patch removallocalization F1: 0.911 -> 0.569Hurt
RecoverMark vs Imuge+ for face recovery under regenerationRecoverMark · CelebA, Structpix2pix, regeneration attackrecovery PSNR (dB): 7.252 -> 22.154Improved
RecoverMark vs Imuge+ for face recovery under patch removalRecoverMark · CelebA, Structpix2pix, patch removalrecovery PSNR and MS-SSIM: PSNR 14.445 -> 16.624; MS-SSIM 0.673 -> 0.467Mixed
train on CelebA, test on OOD FFHQRecoverMark · FFHQ (OOD), Instructpix2pix, regeneration attacklocalization F1: 0.855 -> 0.884No clear effect
add the distortion layer (stage 2 robustness training) (ablation)RecoverMark · CelebA, removal attacksface recovery (qualitative)Improved
introduce regeneration attack first in progressive training (ablation)RecoverMark · CelebA, regeneration attackrobustness (qualitative)Improved
increase face-to-image area ratio above 60% (ablation)RecoverMark · capacity evaluation, regeneration attackrecovery and background PSNRHurt

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