Saved in:
Bibliographic Details
Main Authors: Chen, Yan, Shang, Hanlin, Liu, Ce, Chen, Yuxuan, Li, Hui, Yuan, Weihao, Zhu, Hao, Dong, Zilong, Zhu, Siyu
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2506.13355
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911007415205888
author Chen, Yan
Shang, Hanlin
Liu, Ce
Chen, Yuxuan
Li, Hui
Yuan, Weihao
Zhu, Hao
Dong, Zilong
Zhu, Siyu
author_facet Chen, Yan
Shang, Hanlin
Liu, Ce
Chen, Yuxuan
Li, Hui
Yuan, Weihao
Zhu, Hao
Dong, Zilong
Zhu, Siyu
contents Video face restoration faces a critical challenge in maintaining temporal consistency while recovering fine facial details from degraded inputs. This paper presents a novel approach that extends Vector-Quantized Variational Autoencoders (VQ-VAEs), pretrained on static high-quality portraits, into a video restoration framework through variational latent space modeling. Our key innovation lies in reformulating discrete codebook representations as Dirichlet-distributed continuous variables, enabling probabilistic transitions between facial features across frames. A spatio-temporal Transformer architecture jointly models inter-frame dependencies and predicts latent distributions, while a Laplacian-constrained reconstruction loss combined with perceptual (LPIPS) regularization enhances both pixel accuracy and visual quality. Comprehensive evaluations on blind face restoration, video inpainting, and facial colorization tasks demonstrate state-of-the-art performance. This work establishes an effective paradigm for adapting intensive image priors, pretrained on high-quality images, to video restoration while addressing the critical challenge of flicker artifacts. The source code has been open-sourced and is available at https://github.com/fudan-generative-vision/DicFace.
format Preprint
id arxiv_https___arxiv_org_abs_2506_13355
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle DicFace: Dirichlet-Constrained Variational Codebook Learning for Temporally Coherent Video Face Restoration
Chen, Yan
Shang, Hanlin
Liu, Ce
Chen, Yuxuan
Li, Hui
Yuan, Weihao
Zhu, Hao
Dong, Zilong
Zhu, Siyu
Computer Vision and Pattern Recognition
Video face restoration faces a critical challenge in maintaining temporal consistency while recovering fine facial details from degraded inputs. This paper presents a novel approach that extends Vector-Quantized Variational Autoencoders (VQ-VAEs), pretrained on static high-quality portraits, into a video restoration framework through variational latent space modeling. Our key innovation lies in reformulating discrete codebook representations as Dirichlet-distributed continuous variables, enabling probabilistic transitions between facial features across frames. A spatio-temporal Transformer architecture jointly models inter-frame dependencies and predicts latent distributions, while a Laplacian-constrained reconstruction loss combined with perceptual (LPIPS) regularization enhances both pixel accuracy and visual quality. Comprehensive evaluations on blind face restoration, video inpainting, and facial colorization tasks demonstrate state-of-the-art performance. This work establishes an effective paradigm for adapting intensive image priors, pretrained on high-quality images, to video restoration while addressing the critical challenge of flicker artifacts. The source code has been open-sourced and is available at https://github.com/fudan-generative-vision/DicFace.
title DicFace: Dirichlet-Constrained Variational Codebook Learning for Temporally Coherent Video Face Restoration
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2506.13355