Research map / Image generation / Editing and style transfer
Learning Dual Memory Dictionaries for Blind Face Restoration
Xiaoming Li, Shiguang Zhang, Shangchen Zhou, Lei Zhang, Wangmeng Zuo
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 generic memory dictionary (128 items) instead of plain CNN (ablation) | DMDNet · CelebRef-HQ x4 | PSNR: 26.31 -> 28.97 | Improved |
| add generic memory dictionary (128 items) instead of plain CNN (ablation) | DMDNet · CelebRef-HQ x8 | PSNR: 24.69 -> 27.37 | Improved |
| increase generic dictionary size from 32 to 128 items (ablation) | DMDNet · CelebRef-HQ x8 | PSNR: 26.65 -> 27.37 | Improved |
| increase generic dictionary size from 128 to 256 items (ablation) | DMDNet · CelebRef-HQ x4 | PSNR: 28.97 -> 28.98 | No clear effect |
| add forward update to the generic dictionary (ablation) | DMDNet · CelebRef-HQ x4 | PSNR: 28.60 -> 28.97 | Improved |
| add backward update to the generic dictionary (ablation) | DMDNet · CelebRef-HQ x4 | PSNR: 28.36 -> 28.97 | Improved |
| add backward update to the generic dictionary (ablation) | DMDNet · CelebRef-HQ x8 | PSNR: 26.63 -> 27.37 | Improved |
| initialize dictionary from features then forward update, instead of random noise with backward update only (ablation) | DMDNet · CelebRef-HQ x4 | PSNR: 27.67 -> 28.97 | Improved |
| use attention-based dictionary transform module instead of best-match value (ablation) | DMDNet · CelebRef-HQ x4 | PSNR: 28.24 -> 28.97 | Improved |
| use multi-scale transform modules (2T instead of 1T) (ablation) | DMDNet · CelebRef-HQ x4 | FID: 12.11 -> 10.77 | Improved |
| add generic dictionary to specific-only model (ablation) | DMDNet · CelebRef-HQ x4 | PSNR: 28.43 -> 28.97 | Improved |
| add specific dictionary to generic-only model (ablation) | DMDNet · CelebRef-HQ x8 | PSNR: 27.03 -> 27.37 | Improved |
Closest papers by meaning
- Blind Face Restoration via Deep Multi-scale Component Dictionaries (ECCV 2020 · similarity 0.93)
- RestoreFormer++: Towards Real-World Blind Face Restoration From Undegraded Key-Value Pairs (TPAMI 2023 · similarity 0.90)
- FaceMe: Robust Blind Face Restoration with Personal Identification (AAAI 2025 · similarity 0.88)
- Blind Face Restoration under Extreme Conditions: Leveraging 3D-2D Prior Fusion for Superior Structural and Texture Recovery (AAAI 2024 · similarity 0.87)
- Dual Associated Encoder for Face Restoration (ICLR 2024 · similarity 0.86)
- ReF-LDM: A Latent Diffusion Model for Reference-based Face Image Restoration (NeurIPS 2024 · similarity 0.86)
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.
