Protective perturbations defend personalized diffusion models by inducing CLIP embedding misalignment that triggers shortcut learning of noise patterns; a red-teaming method using image restoration and contrastive decoupling learning aims to break this protection.
As evident from the figure, our proposed decoupled learning (CDL) approach signifi- cantly enhances the quality compared to the case with per- turbations
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2024 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Rethinking and Red-Teaming Protective Perturbation in Personalized Diffusion Models
Protective perturbations defend personalized diffusion models by inducing CLIP embedding misalignment that triggers shortcut learning of noise patterns; a red-teaming method using image restoration and contrastive decoupling learning aims to break this protection.