Research map / Image generation / Editing and style transfer
ID-PreFeR: ID-Preserving Face Restoration with Mixed Data Quality
Chengxuan Zhu, Yuchen Hong, Qingnan Fan, Qi Zhang, Bingtao Fu, Jinxiu Liang, Jinwei Chen, Huaqi Zhang, Chao Xu, Boxin Shi
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 latent prior preservation loss (ablation) | ID-PreFeR · CelebRef | MANIQA: 0.624 -> 0.649 | Improved |
| add latent prior preservation loss (ablation) | ID-PreFeR · CelebRef | ID similarity: 0.746 -> 0.767 | Improved |
| add ID-quality disentanglement (ablation) | ID-PreFeR · CelebRef | MANIQA: 0.555 -> 0.649 | Improved |
| add ID-quality disentanglement (ablation) | ID-PreFeR · CelebRef | LPIPS: 0.279 -> 0.242 | Improved |
| add ID-quality disentanglement (ablation) | ID-PreFeR · CelebRef, #HQ=1 | ID similarity: 0.7632 -> 0.7613 | No clear effect |
| add ID-preserving sampling (ablation) | ID-PreFeR · CelebRef | ID similarity: 0.734 -> 0.767 | Improved |
| add ID-preserving sampling (ablation) | ID-PreFeR · CelebRef, #HQ=5 | ID similarity: 0.7338 -> 0.7671 | Improved |
| add ID-preserving sampling (ablation) | ID-PreFeR · Inference cost | inference time (s): 5.6 -> 29.0 | Hurt |
| add ID-preserving sampling (ablation) | ID-PreFeR · Inference cost | VRAM (GB): 24 -> 34.8 | Hurt |
| restrict LoRA to decoder cross-attention instead of all cross-attention (ablation) | ID-PreFeR · CelebRef | parameters (M): 11.82 -> 6.41 | Improved |
| restrict LoRA to decoder cross-attention instead of all CA and SA (ablation) | ID-PreFeR · CelebRef | ID similarity: 0.756 -> 0.767 | Improved |
| reduce number of HQ references from 5 to 1 (ablation) | ID-PreFeR · CelebRef | ID similarity: 0.7671 -> 0.7613 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- FaceMe — FaceMe: Robust Blind Face Restoration with Personal Identification (AAAI 2025) · better on 3 of 3 shared results
- ReF-LDM — ReF-LDM: A Latent Diffusion Model for Reference-based Face Image Restoration (NeurIPS 2024) · better on 3 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- PFStorer: Personalized Face Restoration and Super-Resolution (CVPR 2024 · similarity 0.91)
- FaceMe: Robust Blind Face Restoration with Personal Identification (AAAI 2025 · similarity 0.91)
- FaceA-Net: Facial Attribute-Driven ID Preserving Image Generation Network (AAAI 2025 · similarity 0.86)
- $\text{ID}^3$: Identity-Preserving-yet-Diversified Diffusion Models for Synthetic Face Recognition (NeurIPS 2024 · similarity 0.86)
- RealisID: Scale-Robust and Fine-Controllable Identity Customization via Local and Global Complementation (AAAI 2025 · similarity 0.86)
- ReF-LDM: A Latent Diffusion Model for Reference-based Face Image Restoration (NeurIPS 2024 · 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.
