Research map / Image generation / Super-resolution
Principal Uncertainty Quantification With Spatial Correlation for Image Restoration Problems
Omer Belhasin, Yaniv Romano, Daniel Freedman, Ehud Rivlin, Michael Elad
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 |
|---|---|---|---|
| use principal-component basis instead of pixelwise basis (E-PUQ vs im2im-uq) | interval-based uncertainty region · Colorization local 8x8 | uncertainty volume: 1.6e-1 -> 2.3e-3 | Improved |
| use principal-component basis instead of pixelwise basis (E-PUQ vs Conffusion) | interval-based uncertainty region · Super-resolution local 8x8 | uncertainty volume: 8.8e-2 -> 1.3e-2 | Improved |
| use principal-component basis instead of pixelwise basis (E-PUQ vs Conffusion) | interval-based uncertainty region · Inpainting local 8x8 | uncertainty volume: 2.7e-1 -> 1.9e-2 | Improved |
| use dimension-adaptive PCs (DA-PUQ) instead of full E-PUQ | PUQ · Colorization local 8x8 | uncertainty volume: 2.3e-3 -> 2.4e-11 | Improved |
| use dimension-adaptive PCs (DA-PUQ) instead of full E-PUQ | PUQ · Inpainting local 8x8 | uncertainty volume: 1.9e-2 -> 3.6e-8 | Improved |
| apply DA-PUQ globally on whole image instead of locally on 8x8 patches | DA-PUQ · Colorization | uncertainty volume: 2.4e-11 -> 1.2e-13 | Improved |
| apply DA-PUQ globally instead of im2im-uq pixelwise | DA-PUQ · Colorization global 128x128 | uncertainty volume: 1.4e-1 -> 1.2e-13 | Improved |
| RDA-PUQ instead of DA-PUQ (fewer PCs to construct) | PUQ · Colorization local 8x8 | number of PCs constructed K: 100 -> 11.8 | Improved |
| RDA-PUQ instead of DA-PUQ | PUQ · Colorization global | number of PCs constructed K: 100 -> 22.3 | Improved |
| RDA-PUQ instead of DA-PUQ | PUQ · Colorization global | uncertainty volume: 1.2e-13 -> 3.1e-13 | Hurt |
| increase patch size from 1x1 to 8x8 in E-PUQ (ablation) | E-PUQ · CelebA-HQ colorization, super-resolution, inpainting | uncertainty volume | Improved |
| allow reconstruction error (DA-PUQ, beta 0.05) compared to exact reconstruction | PUQ · Local 8x8 all tasks | reconstruction risk: 0 -> 2.5e-2 | Hurt |
Closest papers by meaning
- BUFF: Bayesian Uncertainty Guided Diffusion Probabilistic Model for Single Image Super-Resolution (AAAI 2025 · similarity 0.86)
- Uncertainty Visualization via Low-Dimensional Posterior Projections (CVPR 2024 · similarity 0.86)
- On the Posterior Distribution in Denoising: Application to Uncertainty Quantification (ICLR 2024 · similarity 0.85)
- The Perception-Robustness Tradeoff in Deterministic Image Restoration (ICML 2024 · similarity 0.84)
- FiRe: Fixed-points of Restoration Priors for Solving Inverse Problems (CVPR 2025 · similarity 0.84)
- QUTCC: Quantile Uncertainty Training and Conformal Calibration for Imaging Inverse Problems (NeurIPS 2026 · similarity 0.84)
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.
