Research map / Video / Video generation
Dirichlet-Constrained Variational Codebook Learning for Temporally Coherent Video Face Restoration
Baoyou Chen, Ce Liu, Weihao Yuan, Zilong Dong, Siyu Zhu
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add Dirichlet probabilistic code aggregation instead of top-16 selection (ablation) | DicFace · VFHQ-Test BFR | PSNR / FVD: PSNR 28.505 -> 29.099; FVD 349.784 -> 336.015 | Improved |
| add Dirichlet probabilistic code aggregation instead of top-4 selection (ablation) | DicFace · VFHQ-Test BFR | FVD: 361.160 -> 336.015 | Improved |
| add Dirichlet prior with sparsity instead of unconstrained average weights (ablation) | DicFace · VFHQ-Test BFR | AKD: 2.128 -> 2.093 (5.7% reduction in AKD versus Average) | Improved |
| add Dirichlet prior with sparsity instead of unconstrained average weights (ablation) | DicFace · VFHQ-Test BFR | FVD: 342.141 -> 336.015 | Improved |
| add KL prior loss with alpha=1.0 (ablation) | DicFace · VFHQ-Test BFR | PSNR: 28.929 -> 29.099 | Improved |
| add KL prior loss with alpha=1.0 (ablation) | DicFace · VFHQ-Test BFR | FVD: 336.076 -> 336.015 | Improved |
| add KL prior loss with alpha=1.0 (ablation) | DicFace · VFHQ-Test BFR | TLME: 1.091 -> 1.107 | Hurt |
| increase Dirichlet prior alpha from 1.0 to 10.0 (ablation) | DicFace · VFHQ-Test BFR | FVD: 336.015 -> 344.922 | Hurt |
| decrease Dirichlet prior alpha from 1.0 to 0.0001 (ablation) | DicFace · VFHQ-Test BFR | FVD: 336.015 -> 337.365 | Hurt |
| train encoder and decoder instead of freezing them (ablation) | DicFace · VFHQ-Test BFR | image quality and temporal stability | Improved |
| DicFace versus best prior video face restoration method KEEP on TLME | DicFace · VFHQ-Test (blind video face restoration) | TLME: 1.156 -> 1.091 | Improved |
| DicFace versus PGTFormer on FVD | DicFace · VFHQ-Test (blind video face restoration) | FVD: 332.340 -> 336.015 | Hurt |
Closest papers by meaning
- Efficient Video Face Enhancement with Enhanced Spatial-Temporal Consistency (CVPR 2025 · similarity 0.87)
- Discrete Prior-Based Temporal-Coherent Content Prediction for Blind Face Video Restoration (AAAI 2025 · similarity 0.87)
- Latent Harmony: Synergistic Unified UHD Image Restoration via Latent Space Regularization and Controllable Refinement (NeurIPS 2025 · similarity 0.86)
- SeNM-VAE: Semi-Supervised Noise Modeling with Hierarchical Variational Autoencoder (CVPR 2024 · similarity 0.86)
- $InterLCM$: Low-Quality Images as Intermediate States of Latent Consistency Models for Effective Blind Face Restoration (ICLR 2025 · similarity 0.86)
- Dynamic Content Prediction with Motion-aware Priors for Blind Face Video Restoration (CVPR 2025 · similarity 0.85)
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.
