Research map / Image generation / Super-resolution
Extreme Face Super-Resolution through Identity Fitting and Decoupling
Jiarui Yang, Hang Guo, Wen Huang, Shu-Tao Xia, Tao Dai
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 personalized ID-embedding fine-tuning to the pretrained IDFSR | IDFSR (PT) · CelebRef-HQ 8x | PSNR: 24.29 -> 28.85 | Improved |
| add personalized ID-embedding fine-tuning to the pretrained IDFSR | IDFSR (PT) · CelebRef-HQ 8x | IDS: 0.3173 -> 0.2242 | Improved |
| add personalized ID-embedding fine-tuning to the pretrained IDFSR | IDFSR (PT) · CelebRef-HQ 16x | PSNR: 23.32 -> 24.09 | Improved |
| add personalized ID-embedding fine-tuning to the pretrained IDFSR | IDFSR (PT) · CelebRef-HQ 16x | IDS: 0.5227 -> 0.3625 | Improved |
| add ID embedding to the diffusion SR model (case4 removes it) (ablation) | IDFSR full model · component ablation (Fig. 9) | ID consistency | Improved |
| add style embedding to the diffusion SR model (case5 removes it) (ablation) | IDFSR full model · component ablation (Fig. 9) | facial attribute realism | Improved |
| mask the LR image before concatenation (case2 vs case1 plain concatenation) (ablation) | vanilla diffusion SR · component ablation (Fig. 9) | perceptual quality | Improved |
| increase the number of reference images for ID fine-tuning from 2 to 10 (ablation) | IDFSR (FT) · CelebHQ-Ref test set | pixel, ID and attribute consistency | Improved |
| degrade the reference image progressively until reference-free (ablation) | IDFSR (FT) · 16x progressive reference degradation | ID similarity: 0.3426 -> 0.5272 | Hurt |
| use personalized ID embedding versus DiffFace for emotion consistency | IDFSR (FT) · CelebRef-HQ 16x | emotion accuracy: 67.4% -> 66.2% | 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.
- DifFace — DifFace: Blind Face Restoration With Diffused Error Contraction (TPAMI 2024) · better on 1 of 1 shared result
- PGDiff — PGDiff: Guiding Diffusion Models for Versatile Face Restoration via Partial Guidance (NeurIPS 2023) · better on 1 of 1 shared result
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
- Super-Resolving Very Low-Resolution Face Images With Supplementary Attributes (CVPR 2018 · similarity 0.89)
- Low-Light Face Super-resolution via Illumination, Structure, and Texture Associated Representation (AAAI 2024 · similarity 0.87)
- FaceMe: Robust Blind Face Restoration with Personal Identification (AAAI 2025 · similarity 0.86)
- FaithfulFaces: Pose-Faithful Facial Identity Preservation for Text-to-Video Generation (NeurIPS 2026 · similarity 0.85)
- When StyleGAN Meets Stable Diffusion: a W+ Adapter for Personalized Image Generation (CVPR 2024 · similarity 0.85)
- FaceA-Net: Facial Attribute-Driven ID Preserving Image Generation Network (AAAI 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.
