Recent advancements in face restoration have led to high-quality outputs but often lack fidelity to the identity. This paper introduces PFStorer, a personalized face restoration approach using diffusion models. By personalizing a base restoration model with a few high-quality reference images, PFStorer achieves tailored restoration while retaining fine-grained details. The model balances between input image details and personalization, showcasing robust capabilities in real-world scenarios. A generative regularizer is employed to encourage the model to learn a robust neural representation of the identity. Training pipeline improvements enable super-resolution and alignment-free approaches.
To Another Language
from source content
arxiv.org
Key Insights Distilled From
by Tuomas Varan... at arxiv.org 03-14-2024
https://arxiv.org/pdf/2403.08436.pdfDeeper Inquiries